Rapid Modelling for Increasing Competitiveness: Tools and Mindset

Rapid Modelling for Increasing Competitiveness Gerald Reiner Editor Rapid Modelling for Increasing Competitiveness T...

Author: Gerald Reiner

362 downloads 1573 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Rapid Modelling for Increasing Competitiveness

Gerald Reiner Editor

Rapid Modelling for Increasing Competitiveness Tools and Mindset

123

Editor Prof. Dr. Gerald Reiner Institut de l’Entreprise (IENE) Faculté des Sciences Économiques Université de Neuchâtel Rue A.-L. Breguet 1 2000 Neuchâtel Switzerland [email protected]

ISBN 978-1-84882-747-9 e-ISBN 978-1-84882-748-6 DOI 10.1007/978-1-84882-748-6 Springer Dordrecht Heidelberg London New York British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Control Number: 2009929380 c Springer-Verlag London Limited 2009 Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms of licenses issued by the Copyright Licensing Agency. Enquiries concerning reproduction outside those terms should be sent to the publishers. The use of registered names, trademarks, etc., in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore free for general use. The publisher makes no representation, express or implied, with regard to the accuracy of the information contained in this book and cannot accept any legal responsibility or liability for any errors or omissions that may be made. Cover design: eStudioCalamar, Figueres/Berlin Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

Foreword

A Perspective on Two Decades of Rapid Modeling It is an honor for me to be asked to write a foreword to the Proceedings of the 1st Rapid Modeling Conference. In 1987, when I coined the term “Rapid Modeling” to denote queuing modeling of manufacturing systems, I never imagined that two decades later there would be an international conference devoted to this topic! I am delighted to see that there will be around 40 presentations at the conference by leading researchers from around the world, and about half of these presentations are represented by written papers published in this book. I congratulate the conference organizers and program committee on the success of their efforts to hold the first ever conference on Rapid Modeling. Attendees at this conference might find it interesting to learn about the history of the term Rapid Modeling in the context it is used here. During the fall of 1986 I was invited to a meeting at the Headquarters of the Society of Manufacturing Engineers (SME) in Dearborn, Michigan. By that time I had successfully demonstrated several industry applications of queuing network models at leading manufacturers in the USA. Although in principle the use of queuing networks to model manufacturing systems was well known in the OR/MS community and many papers had been published, the actual use of such models by manufacturing professionals was almost nonexistent. Instead, discrete-event simulation had been popularized through aggressive marketing by software companies, and if manufacturing managers wanted an analysis of their systems it was usually done by simulation. A few researchers, including myself, were trying to change this situation and include queuing models in the suite of tools used by manufacturing analysts. In the 1970s Professor Jim Solberg of Purdue had demonstrated that a Flexible Manufacturing System (FMS) could be modeled as a closed queuing network; he implemented a simple calculation on a hand calculator which he took with him as he visited manufacturers and impressed them with his quick predictions of throughput and bottlenecks. He called his model CAN-Q, for computer-aided network of queues. Motivated by his success, but finding his single-class model to be too simplistic, Dr. Rick Hildebrant (of the Draper Lab at MIT) and I developed an extension v

vi

Foreword

of multiclass mean-value analysis (MVA) and implemented it in a software package called MVAQ. We demonstrated the practical value of queuing models by using MVAQ to quickly analyze and improve the throughput of an FMS at Hughes Aircraft Company during a production crisis (Suri and Hildebrant, 1984) - yes, the very same company formed by Howard Hughes that was seen in the recent movie called “The Aviator”! We then proceeded to use MVAQ at several FMS installations around the USA. However, I was not satisfied with the closed network model because FMSs were still rare and most manufacturing facilities operated more like open networks. At the same time Jackson networks were too limiting in their assumptions. So I homedin on the node decomposition approach developed and refined by Buzacott, Shanthikumar, Whitt and a few others, and in 1985 I wrote a software package called ManuPlan (in Fortran!). By the time of the 1986 meeting at SME, I had already demonstrated applications of ManuPlan not just by me, but by manufacturing analysts working at IBM, Alcoa, Digital Equipment Corporation, Pratt & Whitney Aircraft, and other companies - in other words, industry users were working with queuing models! Following this success, the software package was made more professional by a software company, Network Dynamics, Inc. (NDI). Eventually Dr. Gregory Diehl of NDI implemented it on a PC (the package was called ManuPlan II) and then greatly improved its interface to the latest version, MPX. A detailed perspective on these and many other developments in the area of queuing models of manufacturing systems can be found in the article, “From CAN-Q to MPX” (Suri et al, 1995). Anyway, let’s get back to the fall of 1986. The successes at IBM and other companies had been published in several articles and had caught the eye of the continuing education staff at SME. They wanted to know if I could teach a class to manufacturing professionals that would show them how to use queuing models to analyze manufacturing systems. The key question they had was, what would be the benefits of using this approach compared to simulation, and how could we convince people in industry to attend this class? (They didn’t think the term “queuing models” would do much to attract industry people!) At the SME meeting I explained that simulation models took weeks or months to build, debug and verify - in those days there was no interactive model-building software, and simulation models were built by writing detailed programming code. Further, even after the model was ready, it often took several hours for a single run. In other words, evaluation of a single set of decisions could take hours, and if you wanted to evaluate a number of alternatives it could take days or even weeks. I explained to the SME personnel that queuing models required only a few inputs, were easy to build, and just needed a few seconds to evaluate each scenario. As I went through these explanations, it suddenly dawned on me: what queuing models offered was a technique for rapid modeling of manufacturing systems - and the term Rapid Modeling was born! The SME team was convinced, and as a result I taught the first-ever continuing education class (known as a “Master Class” in Europe) on Rapid Modeling at SME Headquarters during April 28-30, 1987 (see Suri, 1987), also see Fig.1). Soon after that, I began using the acronym RMT for Rapid Modeling Technique and documented its advantages in

Foreword

vii

a 1988 article in Manufacturing Engineering (Suri, 1988). I also continued to teach Rapid Modeling classes at SME and at various manufacturing companies. In spite of these efforts, however, Rapid Modeling continued to play only a minor role in the modeling of manufacturing systems. This changed with a major advance in business strategy - and more significantly for this conference, when one of the members of this conference’s program committee invited me to collaborate with her on some projects. The year was 1988 and George Stalk had just published his article on Time-Based Competition (TBC). Professor Suzanne de Treville (then at the Helsinki University of Technology) invited me to Finland to work with Finnish companies on helping them reduce lead times and implement TBC, using the MPX software as an analysis tool. During these assignments we arrived at a critical insight: Rapid Modeling was the ideal tool to help companies reorganize their operations to reduce lead time. Static capacity models, or decision tools such as linear programming, did not show the tradeoffs between utilization and lead time; on the other hand simulation models were too complex and took too long to build, and managers could not wait that long to make time-sensitive strategic decisions. Rapid Modeling tools clearly showed senior managers the key tradeoffs involved and helped them to quickly justify and implement decisions to reduce their lead times (see De Treville, 1992 for examples of how Rapid Modeling benefited the projects in Finland).

Fig. 1 The first public use of the term Rapid Modeling at a continuing education class taught by the author at the Society of Manufacturing Engineers in 1987

From this point on, in our publications and classes we focused on the advantages of Rapid Modeling for lead time reduction (for example, see Suri, 1989). These ad-

viii

Foreword

vantages were further emphasized with the development of Quick Response Manufacturing (QRM) - a refinement of TBC with a specific focus on manufacturing enterprises (Suri, 1998). For instance, at the Center for Quick Response Manufacturing, during our work with around 200 manufacturing companies during the past 15 years (see www.qrmcenter.org) we have found that Rapid Modeling is an invaluable tool to help companies reduce lead time and implement QRM. But enough about the past - let us look to the future. It is very encouraging to see an entire conference organized around the theme of Rapid Modeling, and to see that researchers from around the world will be presenting papers at this conference. Further, it is even more encouraging to see Rapid Modeling being extended beyond manufacturing systems - for example, to supply chain modeling, to container terminals and logistics management, to service processes, and even to venture capital firms and courts of law. All these events speak well for the future of Rapid Modeling. Finally, as one who promoted the Rapid Modeling concept as a tool to help manufacturing companies become more competitive, it is truly heartening to see that leading researchers in Europe have decided to use Rapid Modeling as a core concept in their EU project on “Keeping Jobs in Europe”(see Project Keeping Jobs In Europe, 2009). Once again, I congratulate the conference organizers and the program committee on the rich set of papers that have been put together here. I wish all the participants a fruitful conference, and I would also like to wish all these researchers success in the application of their Rapid Modeling concepts to many different fields. Neuchˆatel, May 2009

Rajan Suri

References De Treville S (1992) Time is money. OR/MS Today 19(5):30–4 Project Keeping Jobs In Europe (2009) Keeping jobs in eu: Rapid modeling for the manufacturing and service industry. URL http://www2.unine.ch/iene-kje Suri R (1987) Rapid modeling techniques: Evaluating manufacturing system decisions. In: A Hands-on Course, SME World Headquarters, Dearborn, MI Suri R (1988) RMT puts manufacturing at the helm. Manufacturing Engineering February:41–44 Suri R (1989) Lead time reduction through rapid modeling. Manufacturing Systems 7:66–68 Suri R (1998) Quick response manufacturing: a companywide approach to reducing lead times. Productivity Press Suri R, Hildebrant R (1984) Modeling flexible manufacturing systems using meanvalue analysis. Journal of Manufacturing Systems Vol. 3(1):27–38 Suri R, Diehl G, de Treville S, Tomsicek M (1995) From CAN-Q to MPX: Evolution of queuing software for manufacturing. Interfaces 25(25):128–150

Preface

Rapid Modelling - Increasing Competitiveness - Tools and Mindset Despite the developments in the field of lead time reduction over the past 25 years, long lead times continue to have a negative impact on companies’ business results, i.e., customer dissatisfaction, loss of market shares, and missed opportunities to match supply and demand. Increased global competition requires companies to seek out new ways of responding to volatile demand and increased customer requirements for customization with continuously shorter lead times. Manufacturing companies, as well as service firms, in the developed economies are in the doldrums because low responsiveness make them vulnerable to low-cost competitors. Companies that are equipped for speed, with innovative processes, will outperform their slower competitors in many industries but the knowledge concerning lead time reduction, which has been developed globally, has yet to be combined into a unified theory. The purpose of this proceedings volume of selected papers presented at the 1st rapid modelling conference “Increasing Competitiveness - Tools and Mindset” is to give a state of the art overview about actual works in the field of rapid modelling in combination with lead time reduction. Furthermore, new developments will be discussed. In general, Rapid Modelling is based on queuing theory but other mathematical modelling techniques as well as simulation models to facilitate the transfer of knowledge from theory to application are of interest in this context as well. The interested reader, e.g., • researchers in the fields of – – – – –

operations management production management supply chain management operations research or industrial engineering as well as

ix

x

Preface

• practitioners with any connection to either – manufacturing or – service operations should have a good overview about what is going on in this field. The objective of this conference is to provide an international, multidisciplinary platform for researchers and practitioners to create and exchange knowledge on increasing competitiveness through speed. Lead time reduction (through techniques ranging from quick response manufacturing to lean production) is achieved through a mutually reinforcing combination of changed mindset and analytical tools. We accepted papers that contribute to these themes in the form of: • • • •

Theory Pieces and Reviews Modelling and Simulation Case Study and Action Research Survey and Longitudinal Research

Based on these research methods, the proceedings volume has been divided into four chapters and brings together papers which present different research methodologies. These papers are allocated based on their primary research methodology. All papers passed through a double-blind referee process to ensure their quality. Therefore, this book should serve as a valid source for research activities in this field. While the RMC09 (1st rapid modelling conference “Increasing Competitiveness - Tools and Mindset”) takes place at the University of Neuchˆatel, located in the heart of the city of Neuchˆatel, Switzerland, it is based on a collaboration with the project partners within our IAPP Project (No. 217891), see also http://www.unine.ch/ienekje. We are happy to have brought together authors from Algeria, Australia, Austria, Bahrain, Belgium, England, Finland, France, Germany, Hungary, Italy, Sweden, Switzerland and the United States of America.

Acknowledgement We would like to thank all those who contributed to the conference and this proceedings volume. First, we wish to thank all authors and presenters for their contribution. Furthermore, we appreciate the valuable help from the members of the international scientific board, the referees and our sponsors (see the Appendix for the appropriate lists). In particular, our gratitude goes to our support team at Enterprise Institute at the University of Neuchˆatel, Gina Fiore Walder who organized all the major and minor aspects of this conference project. Ulf Richter, who handled the promotion process as well as the scientific referee process. Gina Fiore Walder, Yvan Nieto, Gil Gomes dos Santos supported by Arda Alp and Boualem Rabta handled the majority of the text reviews as well as the formating work with LaTex. Ronald Kurz created the logo

Preface

xi

of our conference and he took over the development of the conference homepage http://www.unine.ch/rmc09. Furthermore, we would like to give special thanks to Professor Rajan Suri, the founding director of the Center for Quick Response Manufacturing, University of Wisconsin-Madison, USA, who supported the development of our conference with valuable ideas, suggestions and hints. Furthermore, he authored the forward of this book based on his leading expertise in the field of Rapid Modelling as well as Quick Response Manufacturing. Finally, it has to be mentioned that the conference as well as the book are supported by the EU SEVENTH FRAMEWORK PROGRAMME - THE PEOPLE PROGRAMME - Industry-Academia Partnerships and Pathways Project (No. 217891) “How revolutionary queuing based modelling software helps keeping jobs in Europe. The creation of a lead time reduction software that increases industry competitiveness and supports academic research.” Neuchˆatel, May 2009

Gerald Reiner

Contents

Part I Theory Pieces and Review 1

Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

3

2

Queueing Networks Modeling Software for Manufacturing . . . . . . . . 15 Boualem Rabta, Arda Alp and Gerald Reiner

3

A Review of Decomposition Methods for Open Queueing Networks . 25 Boualem Rabta

Part II Modelling and Simulation 4

Parsimonious Modeling and Forecasting of Time Series drifted by Autoregressive Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Akram M. Chaudhry

5

Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal (Bejaia’s Harbor) . . . . . . . . . . . . . . . . . . . . . . . . . . 53 D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

6

A Dynamic Forecasting and Inventory Management Evaluation Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Johannes Fichtinger, Yvan Nieto and Gerald Reiner

7

Performance Evaluation of Process Strategies Focussing on Lead Time Reduction Illustrated with an Existing Polymer Supply Chain Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

8

79

A Framework for Economic and Environmental Sustainability and Resilience of Supply Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

xiii

xiv

9

Contents

An Integrative Approach To Inventory Control . . . . . . . . . . . . . . . . . . 105 Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 No´emi Kall´o and Tam´as Koltai 11 Integrating Kanban Control with Advance Demand Information: Insights from an Analytical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Ananth Krishnamurthy and Deng Ge 12 Rapid Modelling in Manufacturing System Design Using Domain Specific Simulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Doug Love and Peter Ball 13 The Best of Both Worlds - Integrated Application of Analytic Methods and Simulation in Supply Chain Management . . . . . . . . . . . 155 Reinhold Schodl 14 Rapid Modeling In A Lean Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Nico J. Vandaele and Inneke Van Nieuwenhuyse Part III Case Study and Action Research 15 The Impact of Lean Management on Business Level Performance and Competitiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei 16 Reducing Service Process Lead-Time Through Inter-Organisational Process Coordination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 Henri Karppinen and Janne Huiskonen 17 Is There a Relationship Between VC Firm Business Process Flow Management and Investment Decisions? . . . . . . . . . . . . . . . . . . . . . . . . 209 Jeffrey S. Petty and Gerald Reiner 18 What Causes Prolonged Lead-Times in Courts of Law? . . . . . . . . . . . 221 Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a 19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Ralf Elbert and Robert Sch¨onberger Part IV Survey and Longitudinal Research 20 Measuring the Effects of Improvements in Operations Management . 249 Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

Contents

xv

21 Managing Demand Through the Enablers of Flexibility: The Impact of Forecasting and Process Flow Management . . . . . . . . . . . . . 265 Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner 22 Threats of Sourcing Locally Without a Strategic Approach: Impacts on Lead Time Performances . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Ruggero Golini and Matteo Kalchschmidt 23 Improving Lead Times Through Collaboration With Supply Chain Partners: Evidence From Australian Manufacturing Firms . . . . . . . . 293 Prakash J. Singh A

International Scientific Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

B

Sponsors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309

List of Contributors

Smail Adjabi Laboratory LAMOS, University of B´ejaia, Targa Ouzemour, 6000 B´ejaia, Algeria e-mail: [email protected] Djamil A¨ıssani Laboratory LAMOS, University of B´ejaia, Targa Ouzemour, 6000 B´ejaia, Algeria e-mail: lamos bejaia.hotmail.com Arda ALP Enterprise Institute, University of Neuchˆatel, Rue A.L. Breguet 1, CH-2000 Neuchˆatel, Switzerland e-mail: [email protected] Peter Ball Department of Manufacturing, Cranfield University, Cranfield, Bedford, MK43 0AL, U.K. e-mail: [email protected] Gerhard Bauer Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria e-mail: [email protected] T. Benkhellat Laboratory LAMOS, University of B´ejaia, Targa Ouzemour, 6000 B´ejaia, Algeria Vedran Capkun HEC School of Management, 1, rue de la Liberation, 78351 Jouy-en-Josas cedex, France e-mail: [email protected] Akram M. Chaudhry College of Business Administration, University of Bahrain, P.O.Box #32038, Sakhir, Kingdom of Bahrain, Middle East xvii

xviii

List of Contributors

e-mail: [email protected], [email protected] M. Cherfaoui Laboratory LAMOS, University of B´ejaia, Targa Ouzemour, 6000 B´ejaia, Algeria Suzanne de Treville University of Lausanne, Faculty of Business and Economics, Internef 315, CH-1015 Lausanne, Switzerland e-mail: [email protected] Krisztina Demeter Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary e-mail: [email protected] Ralf Elbert University of Technology Berlin, Chair of Logistics Services and Transportation, sec. H. 94 main building, Room H 9181, Straße des 17. Juni 135, 10623 Berlin, Germany e-mail: [email protected] Johannes Fichtinger Institute for Production Management, WU Vienna, Nordbergstraße 15, A-1090 Wien, Austria e-mail: [email protected] Deng Ge University of Wisconsin-Madison, Department of Industrial and Systems Engineering, 1513 University Avenue, Madison, WI 53706, USA e-mail: [email protected] Dominik Gl¨aßer Institut de l’entreprise, Universit´e de Neuchˆatel, Rue A.-L. Breguet 1, CH-2000 Neuchˆatel, Switzerland e-mail: [email protected] Ruggero Golini Department of Economics and Technology Management, Universit`a degli Studi di Bergamo, Viale Marconi 5, 24044 Dalmine (BG), Italy e-mail: [email protected] Ari-Pekka Hameri Ecole des HEC, University of Lausanne, Internef, Lausanne 1015, Switzerland e-mail: [email protected] Philip Hedenstierna Logistics Research Group, University of Sk¨ovde, 541 28 Sk¨ovde, Sweden Per Hilletofth Logistic Research Group, University of Sk¨ovde, 541 28 Sk¨ovde, Sweden e-mail: [email protected]

List of Contributors

xix

Olli-Pekka Hilmola Lappeenranta Univ. of Tech., Kouvola Unit, Prikaatintie 9, 45100 Kouvola, Finland Ulrich Hoffrage University of Lausanne, Faculty of Business and Economics, Internef 614, CH-1015 Lausanne, Switzerland e-mail: [email protected] Janne Huiskonen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN-53851 Lappeenranta, Finland Werner Jammernegg Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria e-mail: [email protected] Istv´an Jenei Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary e-mail: [email protected] Matteo Kalchschmidt Department of Economics and Technology Management, Universit`a di Bergamo, Viale Marconi 5, 24044 Dalmine, Italy e-mail: [email protected] No´emi Kall´o Department of Management and Corporate Economics, Budapest University of Technology and Economics, M¨uegyetem rkp. 9. T. e´ p. IV. em., 1111 Budapest, Hungary e-mail: [email protected] Henri Karppinen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN-53851 Lappeenranta, Finland e-mail: [email protected] Tam´as Koltai Department of Management and Corporate Economics, Budapest University of Technology and Economics, M¨uegyetem rkp. 9. T. e´ p. IV. em., 1111 Budapest, Hungary e-mail: [email protected] Ananth Krishnamurthy University of Wisconsin-Madison, Department of Industrial and Systems Engineering, 1513 University Avenue, Madison, WI 53706, USA e-mail: [email protected]

xx

List of Contributors

D´avid Losonci Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary e-mail: ([email protected] Doug Love Aston Business School, Aston University, Birmingham, B4 7ET, U.K. e-mail: [email protected] Zsolt Matyusz Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary e-mail: [email protected] N. Medjkoune Laboratory LAMOS, University of B´ejaia, Targa Ouzemour, 6000 B´ejaia, Algeria Yvan Nieto Institut de l’entreprise, Universit´e de Neuchˆatel, Rue A.-L. Breguet 1, CH-2000 Neuchˆatel, Switzerland e-mail: [email protected] Petra Pekkanen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN-53851 Lappeenranta, Finland e-mail: [email protected] Jeffrey S. Petty Lancer Callon Ltd., Suite 298, 56 Gloucester Road, UK-SW7 4UB London, United Kingdom e-mail: [email protected] Timo Pirttil¨a Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN-53851 Lappeenranta, Finland Boualem Rabta Enterprise Institute, University of Neuchatel, Rue A.-L. Breguet 1, CH-2000 Neuchatel, Switzerland e-mail: [email protected] Gerald Reiner Institut de l’entreprise, Universit´e de Neuchˆatel, Rue A.-L. Breguet 1, CH-2000 Neuchˆatel, Switzerland e-mail: [email protected] Heidrun Rosiˇc Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria e-mail: [email protected]

List of Contributors

xxi

Reinhold Schodl Capgemini Consulting, Lassallestr. 9b, 1020 Wien, Austria e-mail: : [email protected] Robert Sch¨onberger University of Technology Darmstadt, Chair of Clusters & Value Chain,Darmstadt University of Technology, Hochschulstrasse 1, 64289 Darmstadt, Germany e-mail: [email protected] Prakash J. Singh Department of Management & Marketing, University of Melbourne-Parkville, 3010, Australia e-mail: [email protected] Nico J. Vandaele Research Center for Operations Management, Department of Decision Sciences and Information Management, K.U. 3000 Leuven, Belgium e-mail: [email protected] Inneke Van Nieuwenhuyse Research Center for Operations Management, Department of Decision Sciences and Information Management, K.U. 3000 Leuven, Belgium e-mail: [email protected] Lawrence A. Weiss McDonough School of Business, Georgetown University, Old North G01A, Washington, DC 20057-1147, USA e-mail: [email protected]

Part I

Theory Pieces and Review

Chapter 1

Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

Abstract In this paper, we consider the impact of cognitive illusions on decision making in the operations management field, in areas ranging from product and process development to project management. Psychologists have studied the effects of overconfidence, the planning fallacy, illusions of control, anchoring, confirmation bias, hindsight bias, and associative memory illusions on individual judgment, thinking, and memory in many experiments, but little research has focused on operations management implications of these biases and illusions. Drawing on these psychological findings we discuss several of these cognitive illusions and their impact on operations managers, plant workers, technicians and engineers alike in a variety of operational settings. As in other contexts, these cognitive illusions are quite robust in operations management, but fortunately the impact of selected illusions can be substantially reduced through debiasing techniques. The examples discussed in this paper highlight the need for more operations-management-based research on the impact of cognitive illusions on decision making.

Suzanne de Treville University of Lausanne, Faculty of Business and Economics, Internef 315, CH-1015 Lausanne T´el´ephone : 021 692 33 41, e-mail: [email protected] Ulrich Hoffrage University of Lausanne, Faculty of Business and Economics, Internef 614, CH-1015 Lausanne e-mail: [email protected] Jeffrey S. Petty Lancer Callon Ltd., Suite 298, 56 Gloucester Road, UK-SW7 4UB London, e-mail: [email protected]

3

4

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

1.1 Introduction People play an integral role in any operation, from process development to execution, assessment, and improvement. Because people are involved, most decisions exhibit some bias, as individuals use heuristics to simplify the decision-making process. Although such biases are not usually considered in managing and evaluating operations, they have a major impact on the decisions that are made, as well as how learning from decisions occurs. Cognitive illusions “lead to a perception, judgment, or memory that reliably deviates from reality” (Pohl 2004: 2). This deviation is referred to as a cognitive bias. Such illusions or biases happen randomly, tend to be robust and hard to avoid, and are difficult – sometimes impossible – to eliminate. There have been occasional references to the scarcity of literature on cognitive illusions or biases in the OM field (e.g., Mantel et al, in press; Schweitzer and Cachon, 2000). These papers refer to one or two cognitive biases, but do not present a large sample of biases that have been studied in the cognitive psychology literature.

1.2 Cognitive Illusions We begin with a bias that is fundamental to operations management: the planning fallacy (an example of overconfidence). We then consider the illusion of control, anchoring and adjustment, hindsight bias, confirmation bias, and associative memory illusion. We close with a brief discussion of debiasing techniques.

1.2.1 Overconfidence: The Planning Fallacy Overconfidence occurs when “our confidence in our judgments, inferences, or predictions is too high when compared to the corresponding accuracy (Hoffrage 2004: 235). One specific type of overconfidence is the planning fallacy, according to which “predicted completion times for specific future tasks tend to be more optimistic than can be justified by the actual completion times or by the predictors’ general beliefs about the amount of time such tasks usually take“ (Buehler et al 1994, 2002: 250). The planning fallacy results in substantial underestimation of timing or cost for a given project or task due to cognitive malfunctions (Buehler et al, 2002; Kahneman and Tversky, 1979; Lovallo and Kahneman, 2003). It is particularly likely for projects that are perceived as a linear set of well-understood tasks, which, however, is often not the case, as we describe shortly. The planning fallacy also has a temporal aspect: The more time we have to plan, the more overconfident we are and the more likely we are to underestimate the demands of the project (Sanna and Schwarz, 2004). Furthermore, the phenomenon increases with incentives (Buehler et al, 2002). This cognitive illusion is particu-

1 Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions

5

larly robust, occurring even when the decision-maker is completely aware of the phenomenon (Buehler et al, 2002). This phenomenon plays a fundamental role in operations management, in areas ranging from delivery to product and process development, to project management. Things always take longer than anyone expected, there is always a scramble to get things pulled together right before the final deadline, and no amount of planning or organization seems to eliminate this bias. Can insights from cognitive psychology inform operations management theory concerning how to improve lead time performance? Or, could operations management theory bring new insights to theory concerning the planning fallacy? Breaking projects into small pieces has been observed to keep projects more on schedule through creating the tension required to keep people focused on due dates (van Oorschot et al, 2002). While this might be feasible with the new product development projects studied by these authors, it would not work for repetitive operations (manufacturing or service). Furthermore, van Oorschot et al. noted that estimates for smaller project packages are more accurate, but the overall project time remains excessively long. Responding to lead times that are longer than expected by increasing our estimation of lead times leads to the “planning loop” (Suri, 1998); Longer estimates reduce the quality of forecasts, increasing mismatches between production and demand, placing additional demands on the system to respond to actual customer needs, resulting in higher utilization and longer lead times. This is consistent with the psychological realization mentioned early that the more time available, the worse the overconfidence. Historically, lead time estimation has been treated as a rational, linear computation. Suri (e.g., 1998) and Hopp and Spearman (1996) used queuing theory to illustrate the complexity of process dynamics, explaining part of the divergence between the expected simplicity and actual complexity of calculating lead times. These complex system dynamics may amplify the cognitive bias implied by the planning fallacy, thus partially explaining why in operations management we so consistently fail to get our lead times right. Furthermore, exploration of the interaction between the cognitive and computational aspects of lead time estimation may lead to new insights concerning this cognitive illusion. Most managers do not understand the impact of bottleneck utilization, lot size, layout, and system variability on lead time (Suri, 1994). As lead times not only increase but explode with utilization, it is not surprising that lead times exceed expectation in the majority of operations, especially given the common emphasis on maintaining high utilization. Therefore, an understanding of the mathematical principles that drive lead times might serve as a model for the cognitive processes involved in the planning fallacy.

6

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

1.2.2 Illusion of Control Illusion of control occurs when an individual overestimates his or her personal influence on an outcome (Thompson, 2004). The illusion increases in magnitude with perceived skill, emphasis on success or failure, and need for a given outcome, as well as in contexts where skill coexists with luck, as people use random positive outcomes to increase their skill attributions (Langer and Roth, 1975). Consider an experienced worker who is choosing whether to follow process documents in completing a task. The illusion of control implies that the worker may believe that process outcomes will be better if he or she draws from experience and intuition rather than relying on the standard operating procedures (SOPs). Interestingly enough, this worker may well believe that other workers should follow the SOPs (for a discussion of worker insistence that co-workers follow SOPs, see Barker, 1993; Graham, 1995). Times when the worker has carried out a process change that has coincided with an improved yield (success, whether or not due to that change vs. normal process variability) will tend to increase this illusion of control. Polaroid’s efforts to introduce Statistical Process Control were hindered by workers’ illusions of control. Workers believed that process outcomes would be better if they were allowed to choose their own machine settings and make adjustments as they deemed necessary, rather than shutting down the machine and waiting for maintenance if process data indicated that the machine was going out of statistical control. This was in spite of data demonstrating substantially increased yields when machines were maintained to maximize consistency (Wheelwright et al, 1992). More generally, workers prey to the illusion of control are unlikely to embrace process standardization and documentation. Entrepreneurs may well demonstrate an illusion of control when it comes to developing the operations for their new venture. E Ink, for example, was a new venture originating from the MIT Media Lab that had invented an electronic ink, opening the door to “radio paper” and an eventual electronic newspaper that would have the look and feel of a traditional newspaper, but that would be updateable from newspaper headquarters. The attitude of the founders was that developing the new product was difficult, but that operations would be relatively easy-a classic illusion of control. Basic operations management problems (such as yield-to-yield variation) kept the company in survival mode for the better part of a decade (Yoffie and Mack, 2005). Had the founders made operations a priority from the start, they may well have been profitable many years earlier. The good news is that the illusion of control can be virtually eliminated by the intrusion of reality, which creates circumstances requiring individuals to systematically estimate the actual control that they have in a process (Thompson, 2004). In other words, before standardizing and documenting processes, or before designing new processes, it is worth carrying out an assessment exercise so that people have a clear understanding of their true abilities and control level.

1 Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions

7

1.2.3 Anchoring and Hindsight Bias Anchoring and adjustment, that is, predicting or estimating relative to some anchor (Mussweiler et al, 2004), is a heuristic that is often evoked to explain cognitive biases that can be observed in various aspects of operations management. Anchors may be used individually when making decisions, or collectively across an organization as a benchmark for success or failure, often without regard for their relevance or impact on a given situation. In the operations management field we often use anchoring to make an operations strategy more powerful; Consider “Zero Defects” or “Zero Inventories,” “Justin-Time” or “lean production,” and “Six Sigma,” all of which have in common use of a keyword anchor that powerfully sets direction. The positive aspect of these anchors is that a direction or standard for the company has been established, it may, however, be set with such force that later efforts to moderate are unfruitful. Hopp and Spearman (1996), for example, described the confusion that resulted from use of the Zero Defects or Zero Inventories slogans, as companies responded by excessively slashing inventories or setting completely unrealistic quality objectives (as has also occurred as companies that should be striving for percent or parts-perthousand defects vainly strive for the ppm or even parts per billion implied by being six standard deviations from the process mean). The term lean is so powerful that companies may become overenthusiastic about removing slack resources (required for creativity, flexibility, or improvement efforts, e.g., Lawson, 2001) from processes (De Treville and Antonakis, 2006; Rinehart et al, 1997). Anchoring has been observed to move companies away from rational inventory policies (Schweitzer and Cachon, 2000), and to shift companies from constantly striving for improvement to just working toward meeting a set standard Tenbrunsel et al (e.g., that individuals stop seeking to save the environment and simply work to meet environmental standards, 2000). An interesting example of anchoring in the field of operations management comes from the shift in attitude toward the level of defects in a process over the past couple of decades. Twenty years ago, a classroom discussion of defect levels might include student claims that if the optimum is 10% defects and we are aiming for 2%, we are going to make less money than we should.“ Fast forward to today’s classroom, where a similar comment might beif the optimum is 300 parts per million (ppm) defects and we are aiming for 50 ppm” In other words, referring to percent vs. ppm anchors decision-makers as they set process improvement goals. Anchoring influences process experimentation. Consider a conveyor belt that carries product through a furnace that is limiting the overall capacity of the process. In thinking through how to increase throughput for the furnace operation, process development engineers may limit their experiments if they anchor their analysis to the existing process, rather than taking a new look at how the process is run. In the case of the conveyor, for example, it might be possible to almost double the output by stacking pieces on the belt, which would require both slowing the belt and increasing the temperature.

8

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

On a larger scale, Campbell Soup anchored their process development activities to canned soup production in designing a line for microwaveable soups. This anchoring was not even questioned until a new project manager was brought in who had frozen food rather than canned soup experience, and was therefore free to break free from the anchor (unfortunately too late to save the project, Wheelwright and Gill, 1990).

1.2.4 Confirmation Bias “Whenever people search for, interpret, or remember information in such a way that the corroboration of a hypothesis becomes likely, independent of its truth, they show a confirmation bias.” (Oswald and Grosjean 2004: 93). Confirmation bias represents a “type of cognitive bias toward confirmation of the hypothesis under study. To compensate for this observed human tendency, the scientific method is constructed so that we must try to disprove our hypotheses” (Wikipedia, 2006). This type of bias becomes more likely when the “hypotheses tested are already established or are motivationally supported” (Oswald and Grosjean 2004: 93). Watkins and Bazerman (2003) described several disasters that would have been easily preventable had individuals not fallen prey to confirmation and related biases. “As managers estimate the likelihood of an event’s occurrence, they may overestimate the representativeness of a piece of information and draw inaccurate conclusions” (Bazerman, 2005). This also implies that information that is easily available may well have a greater impact on the decision made than it should: Whether decision-makers notice or ignore a piece of information often depends on how that information is presented (Mantel et al, in press). One of the early demonstrations of the confirmation bias came from an experiment in which subjects were shown the sequence 2, 4, 6, and asked to find the rule that generated the sequence. Subjects were to propose their own triples, learn from the experimenter whether the sequence conformed to the rule, and specified the rule as soon as they thought they had discovered it. The actual rule was “any ascending sequence.” Many subjects, however, assumed a rule of the form of n+2, and generated sequences of this form to confirm their guess. Such subjects were quite surprised to learn that their specified rule was wrong, in spite of the fact that they had only received positive and confirming feedback. Arriving at the correct rule required that subjects select examples that would disconfirm their beliefs, but this did not come naturally (Wason, 1960). Compare this phenomenon to an employee who has an idea about how to improve a process. As demonstrated by Wason, such an employee is more likely to suggest experiments to demonstrate that the idea works than to seek problems that may arise. Furthermore, implementation of insufficiently tested ideas is a primary source of production variability (Edelson and Bennett, 1998). The choice by Campbell Soup engineers to create a microwaveable soup process that resembled a canned soup line (argued in the preceding section to demonstrate

1 Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions

9

anchoring) also provides an example of confirmation bias. Just as respondents in Wason (1960) experiment did not seek examples that would disconfirm the intuitive rule suggested by the initial sequence, so the Campbell Soup team made no apparent efforts to test their assumption that all soup lines should look like canned soup lines until the new project manager brought them a new model of reality (Wheelwright and Gill, 1990). There are many examples of confirmation biases in the operations management field. We mentioned previously how illusion of control led to unmanageable new processes in the Campbell Soup case (Wheelwright and Gill, 1990). The ability of these managers to dispel these illusions of control by an injection of reality was hindered by subsequent confirmation bias: Although it was clear that nothing was working as it should, management could not read the signals, nor did they perceive the costs to be excessive, and continued to invest in develop of the new processes. This example illustrates how cognitive illusions can coexist and reinforce each other. The confirmation bias emerges in decisions about outsourcing (for a discussion of confirmation-related biases in outsourcing, see Mantel et al, in press). Managers considering outsourcing have often already made up their minds about whether a supplier is capable of meeting their needs, so that they do not really consider the possibility that the supplier might fail. Confirmation bias can hinder communication and theory development, as opposing camps only consider information that supports their pet hypothesis, as can be seen in the case of lean production. A given factory is likely to have lean proponents and opponents, both of whom can produce substantial quantities of data supporting their viewpoint. The ability to really make use of conflicting data to falsify the opposing theory, however, appears to be the scarce supply (e.g., De Treville and Antonakis, 2006).

1.2.5 Associative Memory Illusion “In recollecting some target event from the past, people will often confuse events that happened before or after the target even with the event itself,” with some illusions involving remembrance of events that never actually occurred (Bartlett 1932/95; Roediger III and Gallo 2004: 309). In managing operations, memory plays an important role. When was the last time we did a furnace profile or maintained that machine? How has that supplier been performing over the past year? Does it seem like the process is stable? What has been going on with operators and repetitive strain injuries? The list goes on and on. The constant updating of memories plays an important role in adaptive learning, and is almost impossible to prevent or control (Roediger III and Gallo, 2004). That memory is constantly reconstructed based on our theories, beliefs, and subsequent experiences demonstrates the importance of excellent record-keeping and patience with those who remember differently. Associative memory illusions are related to illusions of change or stability (Wilson and Ross, 2004), referring to in-

10

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

accurate comparisons of past and present states. Individuals, for example, often erroneously believe that improvement has occurred simply because of involvement in an improvement activity. Consider MacDuffie’s description of Ford’s improvement activities: “[Reporting forms] appear to be used more to report on the activity level of the subsystem group, to show that the required processes are being fulfilled, rather than to diagnose, systematically, the “root cause” and possible solutions to a problem. When a problem recurs, seldom is it reanalyzed, and rarely are earlier actions reassessed. With past activities already documented and reported, the key is to generate new documentation, to provide proof of continued activity. Thus, “continuous improvement” becomes less a process of incremental problem resolution than a process of energetic implementation of intuitively selected solutions” (MacDuffie, 1997, 185). In other words, the assumption of these managers is that activity = improvement, and this is seldom tested.

1.3 Debiasing Techniques Once cognitive biases have been identified, what debiasing techniques exist to reduce their impact? In this section we briefly examine some tools that may contribute to debiasing.

1.3.1 Inside or Outside Views In considering ways to reduce the planning fallacy, it is useful to differentiate between an inside (focusing on singular information) and an outside view (focusing on distributional information). One reduces the dominance of the singular (i.e., subjective probabilities for single cases) over the distributional approach (i.e., estimation of frequency distribution parameters) by eliciting predictions for aggregate sets of events rather than single cases. Asking for aggregate frequencies rather than singlecase estimates has proven to reduce or eliminate a number of cognitive illusions (Gigerenzer et al, in press). Unfortunately however, Buehler et al. (2002: 269) reported that using this method to debias the planning fallacy (asking questions in the form “In how many cases like this will you be able to keep the deadline?” rather than “What is your subjective probability of being able to keep the deadline?”) was not successful, speculating that it takes a sophisticated view to conceive of a given task as a sample of a more general reference class. It seems that participants adapt an inside view to make estimates about a given project, only then inferring a frequency response from this individual case. Beginning with an outside view to arrive at an inside view appears to be unnatural for average individuals.

1 Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions

11

1.3.2 Consideration of Alternative Scenarios Evidence is inconclusive concerning the impact of asking individuals to consider alternative scenarios. On one hand, encouraging people to consider other, more pessimistic scenarios generally reduces overconfidence, both in knowledge and in prediction (Koriat et al, 1980). Not surprisingly, scenario planning has become popular as a forecasting tool in many business and organizational contexts (for a review, see Schoemaker, 1993). On the other hand, Buehler et al (2002) reported that in their studies the planning fallacy has resisted this technique, obviously because individuals’ best-guess predictions were overly influenced by the most optimistic scenarios, thereby downplaying the more pessimistic (and, unfortunately, often more realistic) scenarios.

1.3.3 Premortem Exercise We suggest that a technique called the Premortem exercise (Klein 2003: 98-101) may be more successful in overcoming or reducing the planning fallacy. This method starts with the assumption that a project or plan has failed. Not just a bit, but in a big way: It has turned out to be a catastrophe or disaster. Participants in the exercise take this failure as a given and provide reasons why it happened. This procedure relieves the participants from the (usually self-imposed) constraint that they must not say anything unpleasant, depressing, or potentially hurtful to their colleagues. The aim is to compile a long list of hidden assumptions that turned out to be wrong, or of weaknesses and key vulnerabilities in a plan. Once this list has been established, managers are enabled to take such “unforeseeable” events into account when planning, incorporating buffers and contingencies. Although in our experience the premortem technique has been quite successful in debiasing the planning fallacy, we are not aware of studies that have systematically explored its use.

1.3.4 Recall-Relevance Manipulation Getting participants to use their past experience to calibrate their time judgments has proven to be successful in empirical verification. Buehler et al (1994) required participants to first indicate the date and time they would finish a computer assignment if they finished it as far before its deadline as they typically completed assignments. In a second step, participants were asked to recall a plausible scenario from their past experience that would result in their completing the computer assignment at the typical time. Based on these estimations, they were to make predictions about completion times. This “recall-relevance” manipulation successfully reduced the optimistic bias constituting the planning fallacy.

12

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

1.3.5 Incorporation of Lead Time Reduction Theory Combining an understanding of the mathematical relationships that drive lead times may be helpful in reducing the planning fallacy. Recall that the planning fallacy is particularly likely in situations where individuals perceive a project as a set of linear, straightforward tasks. Although operations often appear to be just such a set of linear and well-understood tasks, the underlying system dynamics are neither linear nor straightforward. We propose that a good understanding of the relationship between utilization, lot size, variability, and layout to lead time might encourage incorporation of a buffer at bottleneck operations, more attention to lot size and variability reduction, and implementation of layouts that support flow. Perhaps these operations management concepts will eventually inform the literature on the planning fallacy.

1.4 Conclusions This paper considered many of the biases and cognitive illusions which are relevant to the field of operations management and will continue to have an effect on operations as long as people are involved in the decision-making process. Cognitive psychologists have developed theories and conducted empirical research that can serve as a theoretical foundation for operations-based research. As demonstrated in the case examples cited, these biases occur in both start-up and established ventures, and across all levels of the companies. The planning fallacy and anchoring effect appear to dominate operations-related activities, but developing an understanding of each of the cognitive illusions presented in this paper in the operations management context may improve the quality of decisions in our field, as well as facilitate learning.

References Barker J (1993) Tightening the iron cage: Concertive control in self-managing teams. Administrative Science Quarterly 38:408–437 Bartlett F (1932/95) Remembering: A study in experimental and social psychology. Cambridge University Press Bazerman M (2005) Judgment in managerial decision making, 6th edn. John Wiley and Sons, New York Buehler R, Griffin D, Ross M (1994) Exploring the Planning Fallacy”: Why people underestimate their task completion times. Journal of Personality and Social Psychology 67:366–366 Buehler R, Griffin D, Ross M (2002) Inside the planning fallacy: The causes and consequences of optimistic time predictions. In: Gilovich DGT, Kahneman D

1 Managerial Decision Making and Lead Times: The Impact of Cognitive Illusions

13

(eds) Heuristics and Biases: The Psychology of Intuitive Judgement, Cambridge University Press, New York, pp 250–270 De Treville S, Antonakis J (2006) Could lean production job design be intrinsically motivating? Contextual, configurational, and levels-of-analysis issues. Journal of Operations Management 24(2):99–123 Edelson N, Bennett C (1998) Process discipline: How to maximize profitability and quality through manufacturing consistency. Quality Resources, New York Gigerenzer G, Hertwig R, Hoffrage U, Sedlmeier P (in press) Cognitive illusions. In: Smith P (ed) Handbook of experimental economics results, North Holland/Elsevier Press Graham L (1995) On the line at Subaru-Isuzu: The Japanese model and the American worker. ILR Press, Ithaca, NY Hoffrage U (2004) Overconfidence. In: Pohl R (ed) Cognitive Illusions, Psychology Press, Hove, East Sussex, pp 235–254 Hopp WJ, Spearman ML (1996) Factory Physics: Foundations of Manufacturing Management. Irvin Inc., Chicago Kahneman D, Tversky A (1979) Intuitive Prediction: Biases and Corrective Procedures. TIMS Studies in Management Science 12:313–327 Klein G (2003) The Power of Intuition. Currency/Doubleday Koriat A, Liechtenstein S, Fischhoff B (1980) Reasons for confidence. Journal of Experimental Psychology: Human Learning and Memory 6(2):107–118 Langer E, Roth J (1975) Heads I win, tails its chance: The illusion of control as a function of the sequence of outcomes in a purely chance task. Journal of Personality and Social Psychology 32(6):951–955 Lawson M (2001) In praise of slack: Time is of the essence. Academy of Management Executive 15(3):125–135 Lovallo D, Kahneman D (2003) Delusions of success. How optimism undermines executives’ decisions. Harvard Business Review 81(7):56 MacDuffie J (1997) The road to” Root Cause”: Shop-floor problem-solving at three auto assembly plants. Management Science 43(4):479–502 Mantel S, Tatikonda M, Liao Y (in press) A behavioral study of supply manager decision-making: Factors influencing make versus buy evaluation. Journal of Operations Management 24(6):822–838 Mussweiler T, Englich B, Strack F (2004) Anchoring Effect. In: Pohl R (ed) Cognitive illusions, Psychology Press (UK), Hove, East Sussex, pp 183–200 van Oorschot K, Bertrand J, Rutte C (2002) A longitudinal empirical study of the causes of lateness of new product development projects. URL http://www2.ipe.liu.se/rwg/igls/igls2002/Paper111.pdf Oswald M, Grosjean S (2004) Confirmation bias. In: Pohl R (ed) Cognitive Illusions, Psychology Press (UK), Hove, East Sussex, pp 79–96 Pohl R (2004) Introduction: Cognitive illusions. In: Pohl R (ed) Cognitive Illusions, Psychology Press (UK), Hove, East Sussex, pp 1–20 Rinehart J, Huxley C, Robertson D (1997) Just another car factory?: Lean production and its discontents. Cornell University Press, Ithaca, NY

14

Suzanne de Treville, Ulrich Hoffrage and Jeffrey S. Petty

Roediger III H, Gallo D (2004) Associative memory illusions. In: Pohl R (ed) Cognitive Illusions, Psychology Press, East Sussex, pp 309–326 Sanna L, Schwarz N (2004) Integrating Temporal Biases. Psychological Science 15(7):474–481 Schoemaker P (1993) Multiple scenario development: Its conceptual and behavioral foundation. Strategic Management Journal 14(3):193–213 Schweitzer M, Cachon G (2000) Decision bias in the newsvendor problem with a known demand distribution: Experimental evidence. Management Science 46(3):404–420 Suri R (1994) Common misconceptions and blunders in implementing quick response manufacturing. Proceedings of the SME AUTOFACT ’94 Conference, Detroit, Michigan, November Suri R (1998) Quick response manufacturing: A companywide approach to reducing lead times. Productivity Press Tenbrunsel A, Wade-Benzoni K, Messick D, Bazerman M (2000) Understanding the influence of environmental standards on judgments and choices. Academy of Management Journal 43(5):854–866 Thompson S (2004) Illusions of control. In: Pohl R (ed) Cognitive Illusions, Psychology Press, Hove, East Sussex, pp 113–126 Wason P (1960) On the failure to eliminate hypotheses in a conceptual task. The Quarterly Journal of Experimental Psychology 12(3):129–140 Watkins M, Bazerman M (2003) Predictable surprises: The disasters you should have seen coming. Harvard Business Review 81(3):72–85 Wheelwright SC, Gill G (1990) Campbell Soup Company. In: Harvard Business School case 9-690-051, Cambridge, MA, p 23 Wheelwright SC, Bowen HK, Elliott B (1992) Process control at Polaroid. In: Harvard Business School case 9-693-047, Cambridge, MA, p 17 Wikipedia (2006) Confirmation bias. URL http://en.wikipedia.org /wiki/Confirmation bias. Wilson A, Ross M (2004) 21 Illusions of change or stability. In: Pohl R (ed) Cognitive Illustions, Psychology Press (UK), Hove, East Sussex, pp 379–396 Yoffie DB, Mack BJ (2005) E Ink in 2005. In: Harvard Business School case 9-705506, Cambridge, MA, p 24

Chapter 2

Queueing Networks Modeling Software for Manufacturing Boualem Rabta, Arda Alp and Gerald Reiner

Abstract This paper reviews the evolution of queueing networks software and its use in manufacturing. In particular, we will discuss two different groups of software tools. First, there are queueing networks software packages which require a good level of familiarity with the theory. In the other hand, there are some packages designed for manufacturing where the model development process is automated. Issues related to practical considerations will be adressed and recommendations will be given.

2.1 Introduction In a period of continuous change in global business environment, organizations, large and small, are finding it increasingly difficult to deal with, and adjust to the demands for such changes (Bosilj-Vuksic et al, 2007). In order to improve performance of a complex manufacturing system, the dynamic dependencies need to be understood well (e.g., utilization, variability, lead time, throughput, WIP, operating expenses, quality, etc). In this manner rapid modeling techniques like queueing theory, can be applied to improve such an understanding. For instance queueing Boualem Rabta Entreprise Institute, University of Neuchˆatel, Rue A.L. Breguet 1, CH-2000 Neuchˆatel, Switzerland. e-mail: [email protected] Arda Alp Entreprise Institute, University of Neuchˆatel, Rue A.L. Breguet 1, CH-2000 Neuchˆatel, Switzerland. e-mail: [email protected] Gerald Reiner Entreprise Institute, University of Neuchˆatel, Rue A.L. Breguet 1, CH-2000 Neuchˆatel, Switzerland. e-mail: [email protected]

15

16

Boualem Rabta, Arda Alp and Gerald Reiner

networks are useful to model and measure the performance of manufacturing systems and also complex service processes. Apparently, queuing-theory-based software packages for manufacturing processes (e.g., MPX) automate model development process and help users (e.g. managers, academics) identify relatively easy analytical insights (Vokurka et al, 1996). Queuing software can be used by industrial analysts, managers, and educators. It is also a good tool to help students understand factory physics along with modeling and analysis techniques (see, e.g., de Treville and Van Ackere, 2006). Despite certain challenges over of queuing-theory-based modeling (e.g. need strong mathematical background, hard to maintain certain level of understanding on theories), training in queuing-theory-based modeling is likely to yield better competitiveness in lead time reduction (de Treville and Van Ackere, 2006). Business executives do not always make the best possible decisions. That is, managers can fail to understand the implications of mathematical laws and take actions that increase lead times (see, de Treville and Van Ackere, 2006; Suri, 1998). Complex real life service and manufacturing systems have a number of specific features as compared to ’simplistic cases’, posing important methodological challenges. Basic queuing theory provides key insights to practitioners but not complete and depth understanding of the system. Also the complexity of queuing theory based methods has caused companies to use other tools (e.g. simulation) rather than queuing theory. Finally, queuing theory becomes that much popular in academic and research areas, especially for operations modeling, because complexity and size of the real life problems can be reduced to relatively simple yet complex enough models. Compared to a similar simulation model, those will be in less detail, lacking transition behavior of the system but on the other hand simple and sufficient enough to make a decision (de Treville and Van Ackere, 2006). Basically, relatively simple and quick solutions are much more preferred as an initial system analysis or for quick decisions. The rest of this paper is organized as follows: In Section 22.2, we give a brief review of the evolution of queuing network theory, focusing on decomposition methods. In Section 22.3, we list selected queueing software packages. All of them are freely available for download on the Internet. Some manufacturing software packages based on queueing theory are presented in Section 2.4. Finally, we provide a conclusion and give recommendations in Section 2.5.

2.2 Queueing Networks Theory Queueing networks have been extensively studied in literature since Jackson’s seminal paper (Jackson, 1957). The first significant results were those of Jackson (Jackson, 1957, 1963) who showed that under special assumptions (exponnential interarrival and service times, markovian routing, first-come-first-served descipline,...) a queueing network may be analyzed by considering its stations each in isola-

2 Queueing Networks Modeling Software for Manufacturing

17

tion (product form). Gordon and Newell showed that the product form solution also holds for closed queuing networks (i.e., networks where the number of jobs is fixed) with exponential interarrival and service durations (Gordon and Newell, 1967). Those results have been extended in (Baskett et al, 1975) and (Kelly, 1975) to other special cases (open, closed and mixed networks of queues with multiple job classes and different service disciplines). Since this kind of results was possible only under restrictive assumptions, other researchers tried to extend product form solutions to more general networks (decomposition methods). Several authors ( Kuehn (1979),Whitt (1983), Pujolle and Wu (1986), Gelenbe and Pujolle (1987) and Chylla (1986) among others) proposed decomposition procedures for open G/G/1 (G/G/m) queueing networks. Closed networks of queues have also been analyzed by decomposition (see, e.g., Marie, 1979). This approach has been modified in different ways since though (e.g., multiple job classes (Bitran and Tirupati, 1988; Whitt, 1994). In (Kim, 2004) and (Kim et al, 2005) it is shown that the classical Whitt’s decomposition method performs poorly in some situations (high variability and heavy traffic) and the innovations method is proposed as improvement, by replacing relations among squared coefficients of variability with approximate regression relationships among in the underlying point processes. This relationships allow to add information about correlations. It seems that the application of this method gives satisfactory results in various cases. However, there are still some situations where the existing tools fail. Other approaches which have been proposed include diffusion approximations (Reiser and Kobayashi, 1974) and Brownian approximations (Dai and Harrison, 1993; Harrison and Nguyen, 1990; Dai, 2002). Queueing theory is a well-known method for evaluating the performance of manufacturing systems under the influence of randomness (see, e.g., Buzacott and Shanthikumar, 1993; Suri, 1998). The randomness mainly comes from natural variability of interarrival times and service durations. Queueing networks modeling has its origins in manufacturing applications: Jackson’s papers (Jackson, 1957, 1963) targeted the analysis of job shops; a class of discrete manufacturing systems. Suri et al (1993) gave a detailed survey of analytical models for manufacturing including queueing network models. Govil and Fu (1999) presented a survey on the use of queueing theory in manufacturing. Shanthikumar et al (2007) surveyed applications of queuing networks theory for semiconductor manufacturing systems and discussed open problems.

2.3 Queueing Networks Software The developed theory motivated the development of many software packages for the analysis of queueing networks. These packages suppose a good level of familiarity with queueing theory. There are some early packages that were based on original algorithms. The Queueing Network Analyzer (QNA) has been proposed by Whitt as implementation of his two-node decomposition method (Whitt, 1983). QNET

18

Boualem Rabta, Arda Alp and Gerald Reiner

is another software package for performance analysis of queueing networks. It is the implementation of the analysis algorithm based on Brownian approximation of queueing networks (Dai and Harrison, 1993; Harrison and Nguyen, 1990) (motivated by heavy traffic theory). This package is written in text mode and its source code is available for free download. However, it seems that since mid 90s this software has not been rewritten and it is easy to guess that its use has remained very limited. See also, Govil and Fu (1999) for description of other queueing network packages. PEPSY-QNS /WinPEPSY: It has been developed at the University of ErlangenNurnberg in early 90s. It has a comfortable and easy to use graphical environment. This package includes more than 50 different analysis algorithms. The Windows version (WinPEPSY) has particular features : a user friendly graphical interface, a graphical network editor, charts for results... QNAT : The Queueing Network Analysis Tool (QNAT) is a Windows graphical package for analysing a wide variety of queueing networks. QNAT uses Mathematica as its computing platform and can handle general configurations of open and closed networks of both finite and infinite capacity queues. Incorporation of forkjoin nodes, multiclass customers, mixed customer classes and blocking mechanisms of different types are some of the other features available in this software tool. RAQS : Rapid Analysis of Queueing Systems (RAQS) is a Windows graphical queueing software (Kamath et al, 1995) based on the Whitt’s QNA method and its version in Segal and Whitt. (1989). It also implements decomposition algorithms for closed queuing networks and for tandem finite buffer queueing networks. It’s freely available for download. RAQS’ user interface provides less explanation for inexperienced users. Most probably, input and output interfaces are more suitable for experienced users who owns considerable amount of knowledge on basics of queueing theory. QTS : Queueing Theory Software, is written as Excel spreadsheet for solving a wide range of queueing models and other probability models (Markov chains, birth and death processes,...). The software is based on the textbook of Gross et al (2008). One advantage of this software is that the user has all-in-one model and several performance indicators (e.g., server utilization, mean number of jobs in the system and in the queue, mean waiting time in the system and in the queue...) in a simple sheet. JMT : The Java Modeling Tools is a free open source suite implementing several algorithms for the exact, asymptotic and simulative analysis of queueing network models. Models can be described either through wizard dialogs or with a graphical interface. The workload analysis tool is based on clustering techniques. The JMT tool is user-friendly including a visual design tool. Also, visual sliding buttons for simulation parameters (e.g. avg. arrival rate, avg. service time, buffer size and simulation time) makes what-if analysis easy for the user. Notice that those packages implement known (published) algorithms and are all freely available for download (some are open source, e.g., QTS, JMT). The difference is then in the number of implemented algorithms (the number of network types which can be analyzed), the user interface and the presentation of the results.

2 Queueing Networks Modeling Software for Manufacturing

19

Table 2.1 Download links for some free QN software WinPEPSY RAQS QST JMT

http://www7.informatik.uni-erlangen.de/ prbazan/pepsy/download.shtml http://www.okstate.edu/cocim/raqs/raqs.htm http://www.geocities.com/qtsplus/ (Also: http://qtsplus4calc.sourceforge.net/) http://jmt.sourceforge.net/

The important question is whether these software tools are practical and capable enough to satisfy the complex industry needs. Moreover, among the majority of functionalities that they offer, which one is suitable under which circumstances? When performing in a practical context the user of this kind of software is assumed to have an acceptable level of knowledge in queueing theory. The modeling has to be done separately and the results are generally given in a brute form. It is obvious that those drawbacks do not permit a wide use in a company given that managers are in general not queueing specialists.

2.4 Queueing Networks Software for Manufacturing Additionally the previous software tools, more specific software packages were designed for manufacturing based on queueing networks theory. Such modeling aid is automatic and embedded in the software and provides the user a unique ability to model the manufacturing system without worrying about the theoretical side. They are particularly suitable for use by industrials with little or no queueing knowledge. Snowdon and Ammons (1988) survey eight queueing network packages existing at that time. Some of the queueing network software packages are public domain while others are commercially sold by a software vendor. CAN-Q is a recursive algorithm for solving a product-form stochastic model of production systems (Co and Wysk, 1986) based on the results of Jackson and Gordon and Newell. A version of QNA supporting some features of manufacturing systems has also been proposed Segal and Whitt. (1989) but there are no indices that this package has been sold as commercial product or distributed for large use. Other early packages include Q-LOTS (Karmarkar et al, 1985), MANUPLAN (Suri et al, 1986) and Operations Planner (Jackman and Johnson, 1993). MANUPLAN includes an embedded dynamic model that is based on queueing network theory and provides common performance results such as WIP, tool utilization, production rate. The tool also provides trade-off analysis among inventory levels, flow times, reliability of the tools, etc. (Suri et al, 1986). MPX is perhaps the most popular software package in its category. It is the successor of MANUPLAN. Users greatly appreciate the speed of calculations and the ease of modeling despite of several missing improvements possibilities in its behavior and interface. The exact MPX’s algorithm is not published. Apparently, it uses the classical decomposition algorithm (Whitt, 1983) coupled to the operator/workstation algorithm (Suri et al, 1993) with some changes to support additional

20

Boualem Rabta, Arda Alp and Gerald Reiner

features. It also provides a procedure to compute optimal lot sizes and transfer batch sizes. Still that the existing software model is quite generic and does not integrate high level of complexity. For instance, MPX does not provide support for some manufacturing features like finite capacity of buffers, service disciplines other than firstcome-first-served and dynamic lot sizing nor for some popular production systems (e.g., Kanban). On the other hand several industries prefer to use systems design software such as SAP-APO, IBM’s A-Team, etc., (Pinedo, 2002) and those generate their solution based on heuristics, relaxations or approximations different than queueing software solutions. However, usually those approaches have limitations. Their performance change based on certain settings and in general, user needs to complete several experiments to determine the most suitable algorithm. Additionally computation speed becomes one of the most important practical considerations. Instead of those all-inone, multi functional software designs, queueing software can provide quick and easy solutions while covering dynamics and related effects but not higher levels of system details (Suri et al, 1995).

2.5 Further Remarks When using queueing networks software in a practical setting, the resulting models are less accurate and detailed than simulation and give no insights into transition behavior, but they often suffice as decision support tools and can yield results that are useful in real-world applications (de Treville and Van Ackere, 2006). They provide a rapid and easy way to understand systems’ dynamics and predict their performance, in the opposition of complex simulation models which necessitate vast amount of modeling, advanced knowledge and computer time. It is important in today’s world to be able to rapidly evaluate different alternatives as manufacturing systems are in continuous change. This software packages are also an important tool for training and teaching the impact of some decisions on lead time and cost reduction. Queueing networks software is still has limited usage in practical complex manufacturing applications. It is not mature for practitioners how a queueing software can cover complex industry related constraints among with several tradeoffs regarding to several performance objectives. Other issues like data requirements may also be the cause. Software that passes the test of accuracy and detail can fail miserably in the field because it requires data beyond what are easily available (Suri et al, 1995). Those are basically limitations related to practical implications. Close contact between researchers and industrial users has been critical to the growth in use of the software. Emphasis on such contact, along with better linkages to operational systems, will ensure continued growth of manufacturing applications of queuing software (Suri et al, 1995). The use of the software in education may also help to enlarge its use in companies. When students realize the usefulness of

2 Queueing Networks Modeling Software for Manufacturing

21

this tool, it becomes natural that they will use it after they join work in the industry or they become managers. When recognizing the importance of those tools and the opportunities they offer, the existing software packages are still limited in their modeling capabilities. It is important for software creators for enlarging the usability of their packages to offer support of different real manufacturing systems. While handing problems of modeling, a specified software design should be based on realistic assumptions (i.e. buffers capacity, priority rules, integration of forecasting and inventory policies). The combination of queueing networks analysis with statistical and optimization tools... can provide better solutions and attract more practical applications. The presentation of the computations’ output is also an important factor. Customizable reports and graphical charts help to better understand the results. It should be also possible for the software to provide some insights in the interpretation of the results and to warn the user about the limits of its performance (for example, MPX shows a warning when the utilization is very high saying that the results may not be accurate). Performance measures given by queueing packages are based on only steady-state value measurements given as the average values of such measures WIP, flow time. However, it can be desired to have variance (or variability) information about the output performance measures. Also, the provided average values are just approximate and it may be useful to provide trustable bounds for them. The success of a software package depends on many factors other than the accuracy of its computational method. Users look for a powerful tool with evidence of efficiency but also a user-friendly, easy-to-learn and well supported product (documentation and tutorial, demo version, consultancy/training course). The integration with other packages like spreadsheet packages, statistical packages, DBMS, legacy applications, ERP... is also a highly desired feature. Finally, the ability of the software to import/export data from/to other packages allows the users to gain in time and effort. Acknowledgements This work is supported by the SEVENTH FRAMEWORK PROGRAMME - THE PEOPLE PROGRAMME - Marie Curie Industry-Academia Partnerships and Pathways Project (No. 217891) ”Keeping jobs in Europe”

References Baskett F, Chandy K, Muntz R, Palacios F (1975) Open, closed and mixed networks of queues with different classes of customers. Journal of the ACM 22(2):248–260 Bitran G, Tirupati D (1988) Multiproduct queueing networks with deterministic rout-ing: Decomposition approach and the notion of interference. Management Science 34(1):75–100 Bosilj-Vuksic V, Ceric V, Hlupic V (2007) Criteria for the evaluation of business process simulation tools. Interdisciplinary Journal of Information, Knowledge and Management 2:73–88

22

Boualem Rabta, Arda Alp and Gerald Reiner

Buzacott JA, Shanthikumar JG (1993) Stochastic Models of Manufacturing Systems. Prentice-Hall, Englewood Cliffs, NJ Chylla P (1986) Zur Modellierung und approximativen Leistungsanalyse von Vielteilnehmer-Rechensystemen. Dessertation, Faculty for Mathematics and Computer Science, Technical university Munich Co HC, Wysk RA (1986) The robustness of can-q in modelling automated manufacturing systems. International journal of production research 24(6):1485–1503 Dai J, Harrison J (1993) The qnet method for two-moment analysis of closed manufacturing systems. Annals of Applied Probability 3(4):968–1012 Dai W (2002) A brownian model for multiclass queueing networks with finite buffers. Journal of Computational and Applied Mathematics 144(1–2):145–160 Gelenbe E, Pujolle G (1987) Introduction to queueing networks. John Wiley, Chichester Gordon W, Newell G (1967) Closed queueing systems with exponential servers. Operations Research 15(2):254–65 Govil M, Fu M (1999) Queueing theory in manufacturing : A survey. Journal of Manufacturing Systems 18(3):214–240 Gross D, Shortle JF, Thompson JM, Harris CM (2008) Fundamentals of Queueing Theory, 4th edn. John Wiley & Sons, Inc. Harrison JM, Nguyen V (1990) The qnet method for two-moment analysis of open queueing networks. Queueing Sytems 6(1):1–32 Jackman J, Johnson E (1993) The role of queueing network models in performance evaluation of manufacturing systems. Journal of the Operational Research Society 44(8):797–807 Jackson J (1963) Jobshop-like queueing systems. Management Science 10(1):131– 142 Jackson JR (1957) Networks of waiting lines. Operations Research 5(4):518–521 Kamath M, Sivaramakrishnan S, Shirhatti G (1995) Raqs: A software package to support instruction and research in queueing systems. Proceedings of 4th Industrial Engineering Research Conference, IIE, Norcross, GA pp 944–953 Karmarkar US, Kekre L, Freeman S (1985) Lotsizing and leadtime performance in a manufacturing cell. Interfaces 15(2):l–9 Kelly FP (1975) Networks of queues with customers of different types. Journal of Applied Probability 12(3):542–554 Kim S (2004) The heavy-traffic bottleneck phenomenon under splitting and superposition. European Journal of Operational Research 157(3):736–745 Kim S, Muralidharan R, O’Cinneide C (2005) Taking account of correlation between streams in queueing network approximations. Queueing Systems 49(3– 4):261–281 Kuehn PJ (1979) Approximate analysis of general networks by decomposition. IEEE Transactions on Communications 27(1):113–126 Marie R (1979) An approximate analytic method for general queueing networks. IEEE Transactions on Software Engineer 5(5):530–538 Pinedo M (2002) Scheduling: Theory, Algorithms, and Systems, 2nd edn. PrenticeHall Inc.

2 Queueing Networks Modeling Software for Manufacturing

23

Pujolle G, Wu A (1986) A solution for multiserver and multiclass open queueing networks. Information Systems and Operations Research 24(3):221–230 Reiser M, Kobayashi H (1974) Accuracy of diffusion approximations for some queueing networks. IBM Journal of Research and Development 18(2):110–124 Segal M, Whitt W (1989) A queueing network analyzer for manufacturing. Proceedings of the 12th International Teletraffic Congress, Torino, Italy, June 1988 pp 1146–1152 Shanthikumar J, Ding S, Zang M (2007) Queueing theory for semiconductor manufacturing systems : A survey and open problems. IEEE Transactions on Automation Science and Engiheering 4(4):513–522 Snowdon JL, Ammons JC (1988) A survey of queueing network packages for the analysis of manufacturing systems. Manufacturing Review 1(1):14–25 Suri R (1998) Quick Response Manufacturing. Productivity Press, Portland, OR Suri R, Diehl GW, Dean R (1986) Quick and easy manufacturing systems analysis using manuplan. Proceedings of the Spring HE Conference, Dallas, TX pp 195– 205 Suri R, Sanders J, Kamath M (1993) Performance Evaluation of Production Networks, vol 4: Logistics of Production and Inventory, Elsevier, pp 199–286 Suri R, Diehl GWW, de Treville S, Tomsicek MJ (1995) From can-q to mpx: Evolution of queuing software for manufacturing. Interfaces 25(5):128–150 de Treville S, Van Ackere A (2006) Equipping students to reduce lead times: The role of queuing-theory-based modeling. Interfaces 36(2):165–173 Vokurka RJ, Choobineh R, Vadi L (1996) A prototype expert system for the evaluation and selection of potential suppliers. International Journal of Operations & Production Management 16(12):106–127 Whitt W (1983) The queuing network analyzer. Bell Systems Technical Journal 62(9):2779–2815 Whitt W (1994) Towards better multi-class parametric-decomposition approximations for open queueing networks. Annals of Operations Research 48(3):221–248

Chapter 3

A Review of Decomposition Methods for Open Queueing Networks Boualem Rabta

Abstract Open queueing networks are useful for modeling and performance evaluation of complex systems such as computer systems, communication networks, production lines and manufacturing systems. Exact analytical results are available only in few situations with restricted assumptions. In the general case, feasible solutions can be obtained only through approximations. This paper reviews performance evaluation methods for open queueing systems with focus on decomposition methods.

3.1 Introduction Open queueing networks (OQN) are useful for modeling and performance evaluation of complex systems such as computer systems, communication networks, production lines and manufacturing systems. A Queueing network consists of several connected service stations. It is called open if customers can enter from outside and also r the system. A single station (or a node) queueing system consists of a queueing buffer of finite or infinite size and one or more identical servers. We will focus on unrestricted networks where each station has an infinite waiting capacity. Customers arrive from an external source to any station and wait for an available server. After being served, they move to the next station or leave the system. Performance evaluation of open queueing networks has been addressed through : • Exact methods : analytical results are available only in few situations with simple assumptions and particular topologies (Jackson networks). Many classes of networks have no known closed-form solutions. Boualem Rabta Entreprise Institute, University of Neuchatel, Rue A.-L. Breguet 1, CH-2000 Neuchatel (Switzerland) e-mail: [email protected]

25

26

Boualem Rabta

Fig. 3.1 An example of open queueing network (a) and a single station (b)

• Approximation methods including : diffusion approximations, mean value analysis, operational analysis, exponentialization approximations and decomposition methods. • Simulation and related techniques : This is perhaps, the most popular approach to evaluate the performance of queueing networks. Although more realistic and detailed, it can be cumbersome to optimize, and its accuracy is strongly dependent on the quality of the calibration data. First relevant analytical results for OQN were presented by Jackson (1957) who considered a special category (called thereafter Jackson networks) and showed that the joint distribution of the number of customers in the network is the product of the marginal distributions of the number of customers in each station (i.e., a product form solution). This kind of results allows us to analyze the network by considering each station individually. Product form results have been extended to a few situations (e.g. Kelly, 1975) but for general networks, product form solutions are not possible. Therefore, approximations are the only feasible solution. On the other hand, some networks have state spaces that are so large that certain analysis techniques, while theoretically possible, are impractical (Baldwin et al, 2003). The most frequently used approximation methods to analyze open queueing networks have been decomposition methods. According to this approach, the dimension of the network is reduced by breaking it down to sub-networks and analyzing each sub-network in isolation. The decomposition approach assumes that the sub-networks can be treated as being stochastically independent and that the input to each sub-network is a renewal process. Then, the analysis involves three basic steps : 1. decomposition of the network into sub-networks (in most cases, individual stations), 2. analysis of each sub-network and the interaction between the sub-networks, 3. recomposition of the results to compute the network performance.

3 A Review of Decomposition Methods for Open Queueing Networks

27

The parameters of each subnetwork depend on the state of other subnetworks and thus acknowledge the correlation with other subnetworks. The main difficulty lies in obtaining good approximations for these parameters. While the theory of single-station queues finds its origins in Erlang’s work on telecommunications at the beginning of the 20th century, the analysis of networks of queues began in the 1950s. Initial results appeared in Jackson (1954) who considered a system of two stations in tandem. Jackson (1957, 1963) analyzed a class of open queueing networks with Poisson external arrivals, exponential service times and Markovian routing of customers, and showed that the equilibrium probability distribution of customers could be obtained through node-by-node decomposition. Kelly (1975, 1976) extended Jackson’s work by including customers of several classes and different service disciplines. Similar results were presented by Barbour (1976). Baskett et al (1975) presented the most comprehensive results at the time for the classical models. First surveys of queueing network theory include Lemoine (1977) and Koenigsberg (1982). Lemoine discussed an overview of equilibrium results of general Jackson networks and the methodology which has been employed to obtain those results. Disney and Konig (1985) presented an extensive survey covering the seminal works of Jackson and the extensions of Kelly, including a bibliography of more than 300 references. Suri et al (1993) examined performance evaluation models for different manufacturing systems including production lines (tandem queues), assembly lines (arborescent queues), job-shops (OQN),... Buzacott and Shanthikumar (1992, 1993), Bitran and Dasu (1992) and Bitran and Morabito (1996) analyzed both performance evaluation models and optimization models for queueing networks. Bitran and Dasu (1992) discussed strategic, tactical and operational problems of manufacturing systems based on the OQN methodology, with a special attention to design and planning models for job-shops. Govil and Fu (1999) presented a survey on the use of queueing theory in manufacturing. Shanthikumar et al (2007) surveyed applications of queuing networks theory for semiconductor manufacturing systems and discussed open problems. Also, some software packages for the analysis of manufacturing systems are based on queueing networks theory. For instance, Manuplan and MPX (Suri et al, 1995) implement decomposition methods.

3.2 Jackson Networks 3.2.1 Single Class Jackson Networks When interarrival and service times are exponential, we refer to the network as a Jackson network. Here, the network is composed of several interconnected M/M/m stations with first–come–first–served (FCFS) discipline of service and an infinite queue capacity (n + 1 the number of stations in the system where station 0 rep-

28

Boualem Rabta

resents the external world to the network). Then, each station j is described by 3 parameters : The number of servers in the station, m j . The external arrival rate of customers to station j, λ0 j . The expected service rate, μ j . A customer who finishes the service at station i, moves to station j with probability ri j where, 0 ≤ ri j ≤ 1, ∀i, j = 0, .., n and ∑nj=0 ri j = 1, ∀i = 0, .., n. Thus, r0 j is the probability that a customers enters directly from outside to station j and r j0 is the probability that a customer leaves the network after just completing service at station j. Denote by λ j the overall arrival rate to station j and by λ the overall arrival rate to the whole network. By a result of Burke (1956) and Reich (1957) we know that the output of an M/M/m queue in equilibrium is Poisson with the same rate as the input process. Thus, n

λ j = λ0 j + ∑ ri j λi , ∀ j = 1..n,

(3.1)

i=1

is a system of linear equations known as traffic rate equations. The state of the system is defined as a vector x = (x1 , x2 , .., xn ) where x j is the number of customers in station j (customers in queue and in service). Under the assumption that the system reaches stationary regime, denote by π j (x j ) the probability of station j being in state x j and by π (x1 , x2 , .., xn ) the probability of the system being in state x = (x1 , x2 , .., xn ). Jackson (1957) showed that : n

π (x1 , x2 , .., xn ) = ∏ π j (x j ), j=1

with π j is the steady state distribution of the classical M/M/m j queueing system : ⎧ xj ⎪ ⎨ π j (0)m j ρ j if x j ≤ m j , xj! π j (x j ) = mj xj ⎪ ⎩ π j (0)m j ρ j if x j > m j , mj! where, ρ j is the expected utilization of station j, given as :

ρj =

∞

λj

∑ π j (k) = 1 − π j(0) = μ j m j , 0 ≤ ρ j < 1.

k=1

This result says that the network acts as if each station could be viewed as an independent M/M/m queue. In fact, it can be shown (Disney, 1981) that, in general, the actual internal flow in these kinds of networks is not Poisson (as long as there is any kind of feedback). Nevertheless, the previous relation still holds (see, Gross and Harris, 1998). The expected waiting time in queue at station j is the given by :

3 A Review of Decomposition Methods for Open Queueing Networks

E(W j ) =

29

mj

ρ j (m j ρ j ) π j (0). λ j (1 − ρ j )2 m j !

The expected number of visits to station j : E(V j ) =

λj . λ0

(3.2)

where λ0 = ∑ni=1 λ0i . Finally, the expected lead time E(T ) (or cycle time) for an arbitrary customer, that is, the total time spent by a customer in the network from its arrival moment to its final departure, is given by : n 1 . E(T ) = ∑ E(V j ) E(W j ) + μj j=1 Note that the model in Jackson (1963) allows for arrival and service rates to depend on the state of the system. Whitt (1999) proposed a time-dependent and state-dependent generalization of a Jackson queueing network to model a telephone call center. For each station j, external arrivals λ j (t, x), service rates μ j (t, x) and routing probabilities r ji (t, x), i = 1, .., n depend upon the time t and the state x = (x1 , x2 , .., xn ) of the system. The Markovian structure makes it possible to obtain a time-dependent description of performance as the solution of a system of ordinary differential equations, but the network structure induces a very large number of equations, tending to make the analysis intractable. The author presented a framework for decomposition approximations by assuming the transition intensities of the underlying Markov chain to be of a product form.

3.2.2 Multiple Class Jackson Networks Baskett et al (1975) treated multiclass Jackson networks and obtained product form solutions for three service disciplines : processor sharing, ample service and last– come–first–served with preemptive resume servicing. Customers are allowed to switch classes after completing service at a station. The external input may be state dependent and service distributions can be of the phase type. They also considered multiple server first–come–first–served stations where customers of different classes have the same rate of exponentially distributed service times. See the discussion in Kleinrock (1976, Sec. 4.12). Reiser and Kobayashi (1974) generalized the result of Baskett et al. by assuming that customer routing transitions are characterized by a Markov chain decomposable into multiple subchains. Kelly (1975, 1976, 1979) also extended Jackson’s results to multiple class queueing networks. The type of a customer is allowed to influence his choice of path through the network and, under certain conditions, his service time distribution at each queue. Kelly’s model allows for different service disciplines. Even though, the equilibrium probability has a product form Disney and Konig (see also 1985).

30

Boualem Rabta

Let I be the number of customer classes. Customers of type i arrive to the network as a Poisson process with rate λ (i) and follow the route (i)

(i)

(i)

r1 , r2 , ..., r fi (i)

(i)

where r j is the j-th station visited by this type and r fi is the last station visited before leaving the system. At station j, customers have an exponentially distributed service requirement where requirements at stations visited by a customer of a particular class, are independent and those at all stations for all customers are mutually independent and independent of the arrival processes. If queue j contains k j customers then the expected service requirement for the (l) customer in position l is 1/μ j . Also, x j = (v jl , s jl ) (l = 1, ..., k j ) indicates that the l-th customer in the queue is of type v jl and is has reached the stage s jl along its route. X j = (x j1 , x j2 , ..., x jk j ) denotes the state of station j. The state of the network is represented by X = (X1 , X2 , ..., Xn ). It is then proved (Kelly, 1975; Disney and Konig, 1985) that the equilibrium distribution is given by : n

π (X) = ∏ π j (X j ) j=1

where

kj

π j (X j ) = B j ∏

α (v jl , s jl ) (l)

μj

l=1

Bj =

baj

∞

∑

(l) a a=0 ∏l=1 μ j I

,

,

fi

b j = ∑ ∑ α j (i, s), i=1 s=1

α j (i, s) =

(i)

λ (i) if rs = j, . 0 otherwise.

Let N j , ( j = 1, .., n) be the stationary queue lengths in equilibrium. their stationary probabilities are : k B jb j j

P Nj = k j = k . (l) j ∏l=1 μ j The equilibrium departure process of class i is a Poisson process with rate λ (i) and the departure processes of the different classes are mutually independent (Kelly, 1976). Although these results are interesting, practical implementations are difficult due to the size of the state space (Bitran and Morabito, 1996).

3 A Review of Decomposition Methods for Open Queueing Networks

31

The previous model (Kelly, 1976) supposes deterministic routing. The general routing is considered in Kelly (1975). Based on the fact that nonnegative probability distributions can be well approximated by finite mixtures of gamma distributions, he further conjectured that many of his results can be extended to include general service time distributions. This conjuncture is proved by Barbour (1976). Gross and Harris (1998, Sec. 4.2.1) exposed a multiclass network where customers are served by m j exponential servers at station j, with the same service rate for all classes and first–come–first–served discipline. In this case, waiting time is the same for all customer classes. It is suggested to first solve the traffic equations separately for each customer class and then add the resulting arrival rates. Denote by (l) λ0 j the external arrival rate of customers of class l from outside to station j and let (l)

ri, j be the probability for a customer of class l to move to station j after completing (l)

the service at station i. Solving the traffic rate equation (3.1) yields λ j , j = 1, .., n for each class l; i.e., the overall arrival rate of customers of class l to station j. We (l) then obtain, λ j = ∑Il=1 λ j . Using M/M/m j results, we obtain the average number L j of customers in station j (the average waiting time can be obtained by Little’s formula). The average number of customers of class l in station j is then given by : (l)

Lj =

(l)

λj

(i)

∑Ii=1 λ j

L j.

3.3 Generalised Jackson Networks 3.3.1 Single Class Generalized Jackson Networks When the interarrival or service times (or both) are not exponential, we talk about a generalized Jackson network. Decomposition methods try to extend the independence between stations and Jackson’s product form solution to general open networks. The individual stations are analyzed as independent GI/G/m queues after approximating arrival processes by renewal processes. This approach involves : • Combination of the input of each station : arrivals from outside and from other stations are merged to produce an arrival flow to the station. • Analysis of each station as independent GI/G/m : compute performance measures and departures. • Splitting up departures from each station : decomposition of the overall departure flow into departure flows to other station and to outside. In general, distributions are specified by two first moments (the mean and the squared coefficient of variation). This approach was first proposed by Reiser and Kobayashi (1974) and improved by Sevcik et al (1977), Kuehn (1979), Shanthikumar and Buzacott (1981), Albin (1982), Whitt (1983a) among others.

32

Boualem Rabta

Fig. 3.2 Basic steps of decomposition

3.3.1.1 GI/G/1 Open Queueing Network Suppose we have n internal stations in the network with one server at each station. For a station j, external interarrival times a0 j and services times s j are independent and identically distributed (i.i.d) with general distributions. Define the following notations :

λ0 j expected external arrival rate. ca0 j scv (squared coefficient of variation) or variability of external interarrival time (ca0 j =

V(a0 j ) ). E(a0 j )2

μ j expected service rate (μ j = 1/E(s j )). cs j scv or variability of service time (cs j =

V(s j ) ). E(s j )2

After completing service at station i, a customer moves to station j with probability ri j or leaves the network with probability ri0 . Suppose that there is no immediate feedback (rii = 0, i = 1, .., n). Similarly to Jackson networks, we obtain exact (combined) expected arrival rates from the traffic rate equations (3.1). Necessary and sufficient conditions for stability of this network are well known. They are that at each station the total arrival rate must be less than the service rate. See Borovkov (1987) or, for a modern treatment and further references Dai (1995). Merging arrivals : The asymptotic method (Sevcik et al, 1977) and the stationary-interval method (Kuehn, 1979) may be used to determine ca j , i.e., the merged interarrival time variV(a )

ability (ca j = E(a j)2 , λ j = E(a1 j ) ). Moreover, the asymptotic method is asymptotij cally correct as ρ j → 1 (heavy traffic intensity) and the stationary-interval method is asymptotically correct when the arrival process tends to a Poisson process (Bitran and Morabito, 1996). Let cai j be the interarrival time variability at station j from station i. Based on the asymptotic method, ca j is a convex combination of cai j given by :

3 A Review of Decomposition Methods for Open Queueing Networks

ca j =

n λ0 j λi j ca0 j + ∑ cai j . λj i=1 λ j

33

(3.3)

Albin (1982, 1984) suggested an approximation to ca j based on a convex combination between the previous value and the one obtained by the stationary interval method. Whitt (1983b) substituted the stationary interval method by a Poisson process and obtained : n λi j cai j + 1 − w j (3.4) ca j = w j ∑ i=0 λ j where wj =

1 1 + 4(1 − ρ j)2 (v j − 1) vj =

1 λi j ∑ni=0 ( λ j )2

Computing departures : The squared coefficient of variation cd j of the inter-departure stream from station j is computed by Marshall’s formula: cd j = ca j + 2ρ 2j cs j − 2ρ j (1 − ρ j )

E(W j ) . E(S j )

Using the Kraemer and Langenbach-Belz (1976) approximation for the expected waiting time E(W j ) at G/G/1 nodes, cd j = ρ 2j cs j + (1 − ρ 2j )ca j . Splitting departures : Under the assumption of Markovian routing, the departure stream from station j is split. The squared coefficient of variation cd ji of the departure stream from station j to station i is given by cd ji = r ji cd j + 1 − r ji. Analysis of single nodes : The expected waiting time E(W j ) in station j may be estimated by the KLB formula (Kraemer and Langenbach-Belz, 1976) : E(W j ) =

ρ j (ca j + cs j )g(ρ j , ca j , cs j ) , 2μ j (1 − ρ j )

34

Boualem Rabta

where,

g(ρ j , ca j , cs j ) =

exp 1

−2(1−ρ j )(1−ca j )2 3ρ j (ca j +cs j )

if ca j < 1, if ca j ≥ 1.

For other approximations of E(W j ) see, e.g., Shanthikumar and Buzacott (1981) and Buzacott and Shanthikumar (1993). The expected lead time E(T ) for a customer (including waiting times and service times) is given by : n 1 E(T ) = ∑ E(V j )(E(W j ) + ), μ j j=1 where E(V j ) is the expected number of visits to station j given by (3.2). 3.3.1.2 GI/G/m Open Queueing Network Suppose that in station j, there are m j (m j ≥ 1) identical servers. The following system of equations (Whitt, 1983a) allows us to determine ca j for each station : n

ca j = α j + ∑ cai βi j , for j = 1, .., n i=1

where

αj = 1 + wj

n

p0 j ca0 j − 1 + ∑

i=1

pi j (1 − ri j + ri j ρi2 yi )

βi j = w j pi j ri j (1 − ρi2) with w j is defined by (3.4) and pi j = yi = 1 +

λi j λi = ri j λj λj

max{csi , 0.2} − 1 . √ mi

The expressions for α j and βi j follow from considerations of the merging and splitting of customers streams and the impact of service time variability on the squared coefficient of traffic streams departing from a station, as opposed to that of incoming stream. The expected waiting time at station j is given by : E(W j ) =

ca j + cs j Wj 2

3 A Review of Decomposition Methods for Open Queueing Networks

35

where W j is the expected waiting time for a M/M/m j queue. Many other approximations formulas for the mean waiting time in GI/G/m system are given in Bolch et al (2006, Sec.6.3.6). Creating and Combining Customers : The method described in this section allows for customers creation and combination by using a multiplication factor γ j at each station j (Whitt, 1983a). Eliminating immediate feedback : For those stations where r j j > 0 it is advantageous to consider the successive visits of a customer as one longer visit, that is, a customer gets its total service time continuously. The stations’ parameters are changed as follows (Kuehn, 1979) :

μ ∗j = μ j (1 − r j j ) cs∗2 j = r j j (1 − r j j )cs j ri∗j = ri j /(1 − r j j ), i = j. A proof of en exact analogy between stations with and without feedback with respect to the distribution of queue lengths and mean sojourn times was given by Takacs (1963) in the case of G/M/1 stations. The extension to general arrival processes is an approximation. It has been shown by simulation that this reconfiguration step of the network yields good accuracy, whereas the analysis without this step results in considerable inaccuracies (Kuehn, 1979). Further details may be found in Whitt (1983a,b) and Suri et al (1993). Manufacturing systems: To meet needs in the manufacturing environment, this method has been modified to represent machine breakdowns, batch service, changing lot sizes and product testing with associated repair and partial yields (Segal and Whitt, 1989). 3.3.1.3 The Bottleneck Phenomenon Suresh and Whitt (1990) showed that for tandem queues, for example, the original Whitt’s procedure performs well for all except the last station, which is a bottleneck. That is, the expected waiting time at the bottleneck station is underapproximated. The heavy-traffic bottleneck phenomenon can be described as a relatively large number in queue, observed when external arrivals are highly variable and a bot-

36

Boualem Rabta

tleneck station is visited after jobs go through stations with moderate traffic (Kim, 2005). Whitt (1995) suggested an enhancement to the parametric-decomposition method for generalized Jackson networks. Instead of using a variability parameter for each arrival process, he proposed the use of a variability function for each arrival process; i.e., the variability parameter should be regarded as a function of the traffic intensity of a queue to which the arrival process might go. Dai et al (1994) proposed a hybrid method for analyzing generalized Jackson networks that employs both decomposition approximation and heavy traffic theory; the sequential bottleneck method, in which an open queueing network is decomposed in a set of groups of queues, i.e., not necessarily individual queues. 3.3.1.4 PH/PH/1(/K) Open Queueing Network Haverkort (1995, 1998) modified Whitt’s network by using PH/PH/1(/K) queues instead of GI/G/1 queues, so that the individual queues can be solved exactly using matrix-geometric techniques. In another step, he also allowed for the inclusion of finite capacity queues. Sadre et al (1999) extended this work by removing a few approximate steps in the decomposition procedure. In particular, they used exact results for the departure process of PH/PH/1/K queues, as first developed by Bocharov and Naumov (1986). 3.3.1.5 Open Queueing Network with Correlated Input As mentioned before, in most existing decomposition algorithms for open networks, the output of a queue is usually approximated as a renewal process, which becomes the arrival process to the next queue. Since the correlations of network traffic may have a considerable impact on performance measures, they must be captured to some extent by the employed traffic descriptors. Heindl (2001) considered a general tandem network where the internal traffic processes are described as semi-Markov processes (SMPs) and Markov modulated Poisson processes (MMPPs). Heindl and Telek (2002) presented a decomposition methodology based on Markovian arrival processes (MAPs), whose correlation structure is determined from the busy-period behavior of the upstream queues. The resulting compact MAPs in connection with sophisticated moment matching techniques allow an efficient decomposition of large queueing networks. Compared with a previous approach, the output approximation of MAP/PH/1(/K) queues - the crucial step in MAP-based decomposition - is refined in such a way that also higher moments of the number of customers in a busy period can be taken into account. Heindl et al (2006) constructed a Markovian arrival process of second order (MAP(2)) and showed numerically how their results can be used to efficiently decompose queueing networks. Kim et al (2005) proposed an improvement to Whitt’s method (called the innovations method) by replacing relations among squared coefficients of variability with approximate regression relationships among in the underlying point pro-

3 A Review of Decomposition Methods for Open Queueing Networks

37

cesses. These relationships allow to add information on correlations between different streams. Kim (2004) combined the variability function and the innovations method in the context of the heavy traffic bottleneck phenomenon (Kim, 2004). Blacio˜glu et al (2008) proposed a tree-parameter renewal approximation to analyze the splitting and superposition of autocorrelated processes based on the work of Jagerman et al (2004). Two parameters capture information on the first and second order statistics of the original process and the third parameter captures the intricate behaviour that a superposition can exhibit.

3.3.2 Multiple Class Generalized Jackson Networks Whitt (1983a) proposed a procedure to aggregate all classes in a single one and utilize the single class model described above. In this way the original multiple class model is reduced to a single aggregate open network. After the analysis of the aggregate class model, the performance measures for each class are estimated individually.In many cases this aggregation step works quite well, but in some cases it does not (Whitt, 1994). Bitran and Tirupati (1988) considered an open queueing network with multiple customer classes, deterministic routing and generally distributed arrivals and service times. They pointed out that the splitting operation in the original Whitt’s procedure may not perform well due to the existence of interference among classes. Their approximation is based on the two–class case, by aggregating all classes except the one of interest into one where the aggregate arrivals of class 2 is assumed to follow a Poisson processes. Their procedure provides dramatic improvements in accuracy in some cases (Whitt, 1994). As an extension to the approximations by Bitran and Tirupati (1988) and Whitt (1994) developed methods for approximately characterizing the departure process of each customer class from a multi-class single-server queue ∑(GIi /GIi )/1 with a non-Poisson renewal arrival process and a non-exponential service-time distribution for each class, unlimited waiting space and the FCFS service discipline. The results are used for improving parametric-decomposition approximations for analyzing non-Markov open queueing networks with multiple classes. The effect of class-dependent service times is also considered there. Whitt used different approaches : an extension of Bitran and Tirupati’s formula (based on batch poisson and batch deterministic processes) and a heuristic hybrid approximation based on the results for the limiting case where a server is continuously busy. Caldentey (2001) presented an approximation method to compute the squared coefficient of variation of the departure stream from a multiclass queueing system generalizing the results of Bitran and Tirupati (1988) and Whitt (1994). Kim (2005) considered a multiclass deterministic routing queueing network with highly variable arrivals. He pointed out that the previous procedures of Bitran and Tirupati (1988) and Whitt (1994), may not be accurate under high variability as-

38

Boualem Rabta

sumption. He proposed refinements to those results based on Whitt’s variability functions.

3.4 Other Classes of Networks 3.4.1 Infinite Server Networks Harrison and Lemoine (1981) considered networks of queues with an infinite number of servers at each station. They pointed out that independent motions of customers in the system, which are characteristic of infinite-server networks, lead in a simple way to time-dependent distributions of state, and thence to steady-state distributions. Moreover, these steady-state distributions often exhibit an invariance with regard to distributions of service in the network. Massey and Whitt (1993) considered a network of infinite-server queues with nonstationary Poisson input. As a motivating application, they cited wireless (or mobile cellular) telecommunications systems.Their model appears as a highly idealized model, which initially ignores resource constraints. The different queues represent cells. Call originations are modeled as a nonhomogeneous Poisson process, with the nonhomogeneity capturing the important time-of-day effect.

3.4.2 Batch Movement Networks In real life, many applications feature simultaneous job transitions. For example, in manufacturing, parts are often processed and transported in batches. Batch queuing networks have been considered by Kelly (1979) and subsequently Whittle (1986) and Pollett (1987). Miyazawa and Taylor (1997) proposed a class of batch arrival batch service continuous-time open queueing networks with batch movements. A requested number of customers is simultaneously served at a node, and transferred to another node as, possibly, a batch of different size, if there are sufficient customers there; the node is emptied otherwise. Their model assumes a Markovian setting for the arrival process, service times and routing, where batch sizes are generally distributed. The authors introduced an extra batches arrival process while nodes are empty and showed that the stationary distribution of the queue length has a geometric product form over the nodes if and only if certain conditions are satisfied for the extra arrivals and under a stability condition. The correspondence between batch–movement queueing networks and single– movement queueing networks has also been discussed in Coleman et al (1997) for class of networks having product–form solutions.

3 A Review of Decomposition Methods for Open Queueing Networks

39

Meng and Heragu (2004) proposed an extention to the classical decomposition algorithm Whitt (1983a) to handle transfer batch size change between stations. They consider only deterministic routing where transfer batch sizes are also deterministic. Acknowledgements This work is supported by the SEVENTH FRAMEWORK PROGRAMME - THE PEOPLE PROGRAMME - Marie Curie Industry-Academia Partnerships and Pathways Project (No. 217891) ”Keeping jobs in Europe”

References Albin SL (1982) On poisson approximations for superposition arrival processes in queues. Management Science 28(2):126–137 Albin SL (1984) Approximating a point process by a renewal process, ii: Superposition arrival processes to queues. Operations Research 32(5):1133–1162 Baldwin R, Davis IV N, Midkiff S, Kobza J (2003) Queueing network analysis: concepts, terminology, and methods. Journal of Systems and Software 66(2):99– 117 Barbour A (1976) Networks of queues and the method of stages. Advances in Applied Probability 8(3):584–591 Baskett F, Chandy K, Muntz R, Palacios F (1975) Open, closed and mixed networks of queues with different classes of customers. Journal of the ACM 22(2):248–260, DOI http://doi.acm.org/10.1145/321879.321887 Bitran G, Morabito R (1996) Open queueing networks : Optimization and performance evaluation models for discrete manufacturing systems. Production and Operations Management 51(2–4):163–193 Bitran G, Tirupati D (1988) Multiproduct queueing networks with deterministic routing: Decomposition approach and the notion of interference. Management Science 34(1):75–100 Bitran GR, Dasu S (1992) A review of open queueing network models of manufacturing systems. Queueing Systems 12(1-2):95–134 Blacio˜glu B, Jagerman D, Altiok T (2008) Merging and splitting autocorrelated arrival processes and impact on queueing performance. Performance Evaluation 65(9):653–669 Bocharov P, Naumov V (1986) Matrix-geometric stationary distribution for the PH/PH/1/r, queue. Elektronische Informationsverarbeitung und Kybernetik 22(4):179–186 Bolch G, Greiner S, de Meer H, Trivedi K (2006) Queueing Networks and Markov Chains : Modeling and Performance Evaluation with Computer Science Applications, 2nd edn. Wiley, New York Borovkov AA (1987) Limit theorems for queueing networks. i. Theory Probab Appl 31(3):413–427 Burke P (1956) The output of a queueing system. Operations Research 4(6):699– 704

40

Boualem Rabta

Buzacott JA, Shanthikumar JG (1992) Design of manufacturing systems using queueing models. Queueing Systems 12(1–2):135–213 Buzacott JA, Shanthikumar JG (1993) Stochastic models of manufacturing systems. Prentice-Hall, Englewood Cliffs, NJ Caldentey R (2001) Approximations for multi-class departure processes. Queueing Systems 38(2):205–212 Coleman J, Henderson W, Pearce C, Taylor P (1997) A correspondence between product-form batch-movement queueing networks and single-movement networks. Journal of Applied Probability 34(1):160–175 Dai J (1995) On positive recurrence of multiclass queueing networks: A unified approach via fluid limit models. Annals of Applied Probability 5(1):49–77 Dai J, Nguyen V, Reiman M (1994) Sequential bottleneck decomposition: an approximation method for generalised jackson networks. Operations Research 42(1):119–136 Disney RL (1981) Queueing networks. Proceedings of AMS Symposia in Applied Mathematics 25:53–83 Disney RL, Konig D (1985) Queueing networks: A survey of their random processes. SIAM Review 27(3):335–403 Govil M, Fu M (1999) Queueing theory in manufacturing : A survey. Journal of Manufacturing Systems 18(3):214–240 Gross D, Harris M (1998) Fundamentals of Queueing Theory, 3rd edn. Wiley, New York Harrison J, Lemoine A (1981) A note on networks of infinite-server queues. Journal of Applied Probability 18(2):561–567 Haverkort B (1995) Approximate analysis of networks of PH/PH/1/K, queues: Theory & tool support. In: MMB ’95: Proceedings of the 8th International Conference on Modelling Techniques and Tools for Computer Performance Evaluation, isbn: 3-540-60300-X, Springer-Verlag, London, UK, pp 239–253 Haverkort B (1998) Approximate analysis of networks of PH/PH/1/K, queues with customer losses: Test results. Annals of Operations Research 79(0):271–291 Heindl A (2001) Decomposition of general tandem queueing networks with mmpp input. Performance Evaluation 44(1-4):5–23 Heindl A, Telek M (2002) Output models of MAP/PH/1(/K) queues for an efficient network decomposition. Performance Evaluation 49(1–4):321–339 Heindl A, Mitchell K, van de Liefvoort A (2006) Correlation bounds for secondorder maps with application to queueing network decomposition. Performance Evaluation 63(6):553–577 Jackson JR (1957) Networks of waiting lines. Operations Research 5(4):518–521 Jackson JR (1963) Job shop-like queueing systems. Management Science 10(1):131–142 Jackson RRP (1954) Queueing systems with phase type service. OR 5(4):109–120 Jagerman D, Bolcio˜glu B, Altiok T, Melamed B (2004) Mean waiting time approximations in the g/g/1 queue. Queueing Systems 46(3):481–506 Kelly FP (1975) Networks of queues with customers of different types. Journal of Applied Probability 12(3):542–554

3 A Review of Decomposition Methods for Open Queueing Networks

41

Kelly FP (1976) Networks of queues. Journal of Applied Probability 8(2):416–432 Kelly FP (1979) Reversibility and Stochastic Processes. Wiley, NY Kim S (2004) The heavy-traffic bottleneck phenomenon under splitting and superposition. European Journal of Operational Research 157(3):736–745 Kim S (2005) Approximation of multiclass queueing networks with highly variable arrivals under deterministic routing. Naval Research Logistics 52(5):399–408 Kim S, Muralidharan R, O’Cinneide C (2005) Taking account of correlation between streams in queueing network approximations. Queueing Systems 49(3– 4):261–281 Kleinrock L (1976) Queueing systems, Vol. II : Computer applications. Wiley, New York Koenigsberg E (1982) Twenty five years of cyclic queues and closed queue networks: A review. The Journal of the Operational Research Society 33(7):605–619 Kraemer W, Langenbach-Belz M (1976) Approximate formulae for the delay in the queueing system gi/g/1. Proceedings of the 8th International Teletraffic Congress, Melbourne 235:1–8 Kuehn PJ (1979) Approximate analysis of general queuing networks by decomposition. IEEE Transactions on Communications 27(1):113–126 Lemoine AJ (1977) Networks of queues - a survey of equilibrium analysis. Management Science 24(4):464–481 Massey WA, Whitt W (1993) Networks of infinite-server queues with nonstationary poisson input. Queueing Systems 13(1–3):183–250 Meng G, Heragu SS (2004) Batch size modeling in a multi-item, discrete manufacturing system via an open queuing network. IIE Transactions 36(8):743–753 Miyazawa M, Taylor P (1997) A geometric product-form distribution for a queueing network with non-standard batch arrivals and batch transfers. Advances in Applied Probability 29(2):523–544 Pollett PK (1987) Preserving partial balance in continuous-time markov chains. Advances in Applied Probability 19(2):431–453 Reich E (1957) Waiting times when queues are in tandem. Annals of Mathematical Statistics 28(3):768–773 Reiser M, Kobayashi H (1974) Accuracy of the diffusion approximation for some queuing systems. IBM Journal of Research and Development 18(2):110–124 Sadre R, Haverkort B, Ost A (1999) An efficient and accurate decomposition method for open finite- and infinite-buffer queueing networks. In W Stewart and B Plateau, editors, Proceedings 3rd International Workshop on Numerical Solution of Markov Chains p 120 Segal M, Whitt W (1989) A queueing network analyser for manufacturing. Teletraffic Science for New Cost-Effective Systems, Networks and Services, Proceedings of ITC 12 (ed M Bonatti), North-Holland, Amsterdam pp 1146–1152 Sevcik KC, Levy AI, Tripathi SK, Zahorjan JL (1977) Improving approximations of aggregated queueing network systems. Computer Performance (eds K Chandy and M Reiser), North-Holland pp 1–22

42

Boualem Rabta

Shanthikumar J, Ding S, Zang M (2007) Queueing theory for semiconductor manufacturing systems : A survey and open problems. IEEE Transactions on Automation Science and Engiheering 4(4):513–522 Shanthikumar JG, Buzacott JA (1981) Open queueing network models of dynamic job shops. International Journal of Production Research 19(3):255–266, DOI 10.1080/00207548108956652 Suresh S, Whitt W (1990) The heavy-traffic bottleneck phenomenon in open queueing networks. Operations Research Letters 9(6):355–362 Suri R, Sanders JL, Kamath M (1993) Performance evaluation of production networks, vol 4. Elsevier, North-Holland,Amsterdam Suri R, Diehl GWW, de Treville S, Tomsicek MJ (1995) From can-q to mpx: Evolution of queuing software for manufacturing. Interfaces 25(5):128–150 Takacs L (1963) A single server queue with feedback. BSTJ 42:505–519 Whitt W (1983a) The queueing network analyzer. The Bell System technical journal 62(9):2779–2815 Whitt W (1983b) Performance of the queueing network analyzer. The Bell System technical journal 62(9):2817–2843 Whitt W (1994) Towards better multi-class parametric-decomposition approximations for open queueing networks. Annals of Operations Research 48(3):221–248 Whitt W (1995) Variability functions for parametric-decomposition approximations of queueing networks. Management Science 41(10):1704–1715 Whitt W (1999) Decomposition approximations for time-dependent markovian queueing networks. Operations Research Letters 24(3):97–103, DOI http://dx.doi.org/10.1016/S0167-6377(99)00007-3 Whittle P (1986) Systems in Stochastic Equilibrium. Wiley, London

Part II

Modelling and Simulation

Chapter 4

Parsimonious Modeling and Forecasting of Time Series drifted by Autoregressive Noise Akram M. Chaudhry

Abstract This paper addresses issue of modeling, analysis and forecasting of time series drifted by autoregressive noise and finding its optimal solution by extending a conventional linear growth model with an autoregressive component. This additional component is designed to take care of high frequencies of autoregressive noise drift without influencing the low frequencies of the linear trend and compromising on parsimonious nature of the model. The parameters of this model are then optimally estimated through the self updating recursive equations using Bayesian priors. For identification of autoregressive order of noise and estimation of its coefficients ATS procedure of Akram (2001) is employed. Further, for unknown variance of observations an on-line variance learning and estimation procedure is discussed. To demonstrate practical aspects of the model some examples are given and for generation of short, medium and long term forecasts in one go an appropriate forecast function is given.

4.1 Introduction In many economic, financial and physical phenomena time series drifted by autoregressive noise are observed. For analysis of such series numerous simple to complex models had been proposed by researchers. Most of these models are meant for either short term forecasts or medium term or long term forecasts only. Very few of these models generate three types of forecasts in one go. To obtain all these types of forecasts, usually, three different models are employed using different model settings. These forecasts are then joined or/combined to visualize them in one sequence over short to long term time horizon. To do so some sort of alignment is made by the foreAkram M. Chaudhry, Associate Professor College of Business Administration, University of Bahrain-P.O.Box #32038, Sakhir, Kingdom of Bahrain, Middle East, Contact: (+973) 39171071(Mobile), 17438586(Office), 17642281(Res.) e-mail: [email protected], [email protected], [email protected]

45

46

Akram M. Chaudhry

casters by underestimating or/and overestimating the actual forecasts at the joints. By doing so some accuracy of forecasts is sacrificed resulting in heuristic rather than optimum forecasts. Further, for identification of an order of auto-regression of noise terms parametric techniques, such as, Akaike (1973) and Bohlin (1978) are frequently used. These techniques are known to be cumbersome and some time ambiguous. To overcome these problems, a parsimonious linear growth model having an additional drift component is presented. This additional component that takes care of auto-regression in noise component can be easily modeled and re-parameterized if need arises. Before discussing extended model, let us go through a conventional linear growth model of Harrison and Akram (1983) meant for taking care of low frequencies of trend, bearing white noise.

4.1.1 Conventional Linear Growth Model For analysis and forecasting of time series {yt }t=1,2,,T bearing white noise {δt }t=1,2,,T the conventional linear growth model at time t is locally defined as: Yt = f θt + δt

θt = Gθt−1 + wt Where: f = (1×n) vector of unknown stochastic parameters. θt = (n×l) vector of unknown stochastic parameters. G = (n × n) matrix, called, state or transition matrix, of nnumber of nonzero eigenvalues {λi }i=1,...,n . δt is an observation noise, assumed to be normally distributed with mean zero and some known constant variance. wt = (n × 1) vector of parameter noise, assumed to be normally distributed with mean zero and a constant known variance-covariance matrix W = diag(W 1, . . . ,W n), the components of which are as defined by Harrison and Akram (1983). 4.1.1.1 Example 1 In case of a second order (n = 2) model, the above components of the model, in canonical form, at time t are: f= 10

θt = (θ1 , θ2 ) , where the parameter θ1 is the level of underlying process of time series and θ2 is the growth parameter.

4 Parsimonious Modeling and Forecasting of Time Series

47

G = {gi j }i, j=1,2 is a 2 × 2 transition matrix having non zero eigenvalues {λi }i=1,2 , such that g11 = 1, g12 = 1, g21 = 0, g22 = 2. This matrix assists in transition of low frequency of trend housed in parameter vector from state at time t − 1 to t. W = diag(w1, w2) where for a smoothing coefficient 0 < β < min(λi2 )i=1,2 the expressions of w1 and w2 are: w1 = w2 =

V (1 − β )(λ1 + λ2 )(λ1 λ2 − β ) λ2 β

V (1 − β )(λ1λ2 − β )(λ1 − λ2β )(λ22 − β ) λ2 β 2

where 0 < β < 1 is a smoothing coefficient The parameters θ1 and θ2 of this model are optimally estimated using recursive equations of Harrison and Akram (1983). This second order model is the most commonly used member of the family of linear dynamic system models as in many real life cases it sufficiently represents the low frequencies of the underlying processes of many time series in a parsimonious manner. In this paper, therefore this type of model shall be referred and used for time series drifted by autoregressive (AR) noise of order p.

4.1.2 Comments In practice more specific version having eigenvalues λ1 = 1 and λ2 = 2 is preferred for a linearly growing phenomena. However, for exponential growth λ1 = 1 and λ2 < 1 are used. Exact value of λ2 that depends upon formation and the representation of growth by exponential functions, such as Logistic and Gompertz may be estimated by λ2 estimation procedure of Akram (1992).

4.2 Extended Linear Growth Model for AR(p) Drifts The observations drifted by AR(p) type noise, i.e. , Φ p (B)Et = δt may locally be modeled as: yt = f θt + Et

θt = Gθt−1 + wt Et = [Φ p (B)]−1 δt Where: B is a backward shift operator, such that BEt = Et − 1

48

Akram M. Chaudhry

p Φ p (B) = ∏i=1 (1 − φi B) is invertible. That is 0 < |φi | < 1 for ∀i .

{φi }i=1,···,p are the autoregressive coefficients

δ t ∼ N(0,V ) and Wt ∼ N(0,W ) are as defined earlier. In a compact form, this conventional representation of drifted time series may be parsimoniously parameterized as: Yt = f ∗ θt∗ ∗ θt∗ = J θt−1 + wt∗ ;

wt∗ ∼ N(0,W ∗ )

Where for an AR(p) process: f ∗ = (1, 0, . . . , 0) a (1 × (n + p)) vector of known functions or constants ∗ θt∗ = (θ1∗ , . . . , θ(n+p) ) a ((n + p) × 1) vector of unknown parameters.

W ∗ = diag{(W1∗ ,W2∗ } is of rank (n + p) Where W1∗ = {Wi j }i, j=1,,p such that wi j = V for i, j = p and zero otherwise W2∗ =

diag{W1 ,W2 } V

where w1 and w2 are as defined earlier.

J = diag{Φ p , G} is a {(n + p) × (n + p)} state transition matrix of full rank Where Φ p is a (p × p) matrix of autoregressive coefficients {φi }i=1,,p defined as {φi j }i=1,,p such that

φi j = φi βφ0.5 for i = j = 1, . . . , p and zero otherwise. Where βφ , such that 0 < βφ < 1, is a damping coefficient for highly volatile noise frequencies The order and the values of {φi }i=1,,p are determined by using noise identification and testing procedure of Akram (2001). G, the state transition matrix for low frequencies of underlying processes, is as defined earlier.

4 Parsimonious Modeling and Forecasting of Time Series

49

4.3 Estimation of the Parameters of the Extended Model For data Dt = (yt , Dt−1 ) assuming prior of parameter θ at time t − 1 (θt−1 , |Dt−1 ) ∼ N [mt−1 ;Ct−1 ] the posterior of θ at time t (θt , |Dt ) ∼ N [mt ;Ct ] is determined by providing initial information on f , G, W , m0 and C0 as stated by Harrison and Akram (1983) and Akram (1992) through the following recursive equations. Rt = JCt−1 J + W ∗ −1 At = Rt f ∗ V + f ∗ Rt f ∗ Ct = [I − At f ∗ ] Rt et = yt − f ∗ Jmt−1 mt = Jmt−1 + At [yt − f ∗ Jmt−1 ] where at time t, R is a system matrix, I is an identity matrix, A is an updating or gain vector, et are one step ahead forecast errors and W ∗ , a variance-covariance matrix of parameter noise, is as defined earlier. The dimensions of all these components are assumed to be compatible with their associated vectors and matrices of the recursive updating equations.

4.3.1 Example 2 For time series drifted by AR(2) noise process a linear growth model, in canonical form, at time t is operated by defining: f ∗ = ( 1 0 0 0 ) a (1 × 4) vector

θt ∗ = (θ1∗ , . . . , θ4∗ ) a (4 × 1) vector of unknown parameters. W ∗ = diag{W1∗ ,W2∗ } is a (4 × 4) matrix Where W1∗ = {wi j }i, j=1,2 such that w22 = V and zero otherwise W2∗ =

diag{w1 ,w2 } V

where w1 and w2 are as defined earlier.

50

Akram M. Chaudhry

J = diag{Φ p , G} is a (4 × 4) state transition matrix of full rank Where Φ2 = {φi j }i, j=1,2 such that

φi j = φi {βφ0.5 }

i=j=1,2 and zero otherwise.

4.4 On Line Variance Learning For the above recurrence equations the observation noise variance V is assumed to be known. If unknown then at time t it may be estimated on line using the following variance estimation equations. Xt = βv Xt−1 + (1 − f ∗At )dt2 Nt = βv Nt−1 + 1, where 0 < βv < 1 is a variance damper. Vt =

Xt Nt

Ytˆ = Vt + f ∗ Rt f ∗ dt = min(et2 , ξ Ytˆ) , where ξ is a preset constant, a value of 4 for 95% confidence level and 6 for 99% confidence level. This variance learning system starts generating fairly accurate variance estimates after couple of observations.

4.5 Forecast Function For generating short, medium and long term forecasts in one go the forecast function is: (k)

Ft

= f ∗ J k mt for k ≥ 1 integers.

This function yields optimum short term forecasts and fairly accurate medium to long term forecasts at the same time.

4 Parsimonious Modeling and Forecasting of Time Series

51

4.6 Comments The above model is presented for time series drifted by AR(p) noise process. In practice rarely time series drifted by more than AR(2) process are observed. In many cases, therefore linear growth model with drifted component of AR(2) is required. For more discussion see Akram (1994), Bohlin (1978) and Harrison and Akram (1983). To determine exact order of AR noise many techniques are available. For great ease however, AIC of Akaike (1973) and ATS of Akram (2001) may be employed. Among these two techniques, ATS may be effectively used by the practitioners to estimate the unknown values of autoregressive coefficients {φi }i=1,,p as demonstrated by Akram and Irfan (2007) The above model is parameterized in a canonical form. For application purpose, if desired, may be transformed to a diagonal form by using inverse transformation of Akram (1988). This model, if used in accordance with Akram (1994) is expected to take care of high frequencies of autoregressive noise while keeping low frequencies of underlying process of time series intact. As a result yielding fairly accurate forecasts.

References Akaike H (1973) Information theory and an extension of the maximum likelihood principle. In: Proceedings of the 2nd International Symposium on Inference Theory. BN Petran, F. Cs´aki Eds. Akad´emiai Kiadi, Budapest, Hungary, pp 267–281 Akram M (1988) Recursive transformation matrices for linear dynamic system models. J Computational Stat & Data Analysis 6:119–127 Akram M (1992) Construction of state space models for time series exhibiting exponential growth. In: Computational Statistics, vol 1, Physica Verlag, Heidelberg, Germany, pp 303–308 Akram M (1994) Computational aspects of state space models for time series forecasting. Proceedings of 11th Symposium on Computational Statistics (COMPSTAT-1994), Vienna, Austria, pp 116–117 Akram M (2001) A test statistic for identification of noise processes. Pakistan Journal of Statistics 17(2):103–115 Akram M, Irfan A (2007) Identification of optimum statistical models for time series analysis and forecasting using akaike information criterion and akram test statistic: A comparative study. Proc.of World Congress of Engineers, London, vol 2, pp 956–960 Bohlin T (1978) Maximum-power validation of models without higher-order fitting. Automatica 14:137–146 Harrison P, Akram M (1983) Generalized exponentially weighted regression and parsimonious dynamic linear modelling. Time Series Analysis: Theory and Practice 3:102–139

Chapter 5

Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal (Bejaia’s Harbor) D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

5.1 Introduction Increasing of the traffic at the park with containers of the Bejaia harbor’s and the widening of its physical surface are not directly proportional. This is why the improvement of the productivity of the park and the good functioning of the unloading and loading system requires the specialization of the equipment and the availability of storage area which can receive the unloaded quantity, and having a configuration which will be able to adapt and answer the traffic growth. Accordingly, a first study which aimed to model the unloading process, had been realized in 2003 (Sait et al, 2007). At that time, the park with containers of the EPB (Harbor Company of Bejaia) was of 3000 ETU (Equivalent Twenty Units): 2100 ETU for the full park and 900 ETU for the empty park. The study showed that for an arrival rate of 0.55 ships/day, and a batch size of 72 ETU, the mean number of containers in the full park was of 1241 ETU. While varying the rate of the arrivals (or the batch size), the park full will be saturated for a rate of 1.0368 ships/day (or for a size of 200 ETU). This study was one of the factors that have raised awareness of the EPB to the need of creating a dedicated terminal in the treatment of container, where the birth of BMT (Bejaia Mediterranean Terminal) Company. The company began its commercial activities in July 2005. In order to ensure a good functioning of the container terminal, some performance evaluation studies are established. A first study was realized in 2007 (see Ayache et al, 2007). It had for objective the global modeling of unloading/loading process and had shown that if the number of ships (having a mean size of 170 ETU), which was of 0.83 ships/day, increases to 1.4 ships/day, the full park will undergo a saturation of 94%. Djamil A¨ıssani Laboratory LAMOS, University of B´ejaia, e-mail: lamos bejaia.hotmail.com Smail Adjabi Laboratory LAMOS, University of B´ejaia, e-mail: [email protected]

53

54

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

In this work, we propose another modeling approach which consists to decompose the system into four independent sub-systems: the loading, the unloading, the full stock and the empty stock processes.

5.2 Park with Containers and Motion of the Containers In this section, we present the park with containers of BMT Company and identify the motion of the containers.

5.2.1 The BMT Park with Containers Actually, the terminal is provided with four quays of 500 m, and a park with containers which has a storage capacity of 10300 ETU. The park is divided into four zones: full park, empty park, park with refrigerating containers and a zone of discharge/potting (see Fig. 5.1.a). The park with full containers has a capacity of 8300 ETU and the one with empty containers has a capacity of 900 ETU. In addition, the BMT container terminal offers specialized installations for the refrigerating containers and the dangerous products of a capacity of 600 ETU, as well as a zone of destuffing/packing of a capacity of 500 ETU (see Fig. 5.1.a).

5.2.2 Motions of the Containers The principal motions of the containers at the Bejaia harbor’s are schematized in the Fig. 5.1.b (Ayache et al, 2007). 5.2.2.1 The Unloading Process The unloading process is made up mainly of five steps. 1. The step of anchorage: With the exception of the car-ferries and container ships, any ship arriving at the Bejaia harbor’s is put on standby in the anchorage (roads) for a duration of time which varies from a ship to another, because of the occupation of the quay stations or unavailability of pilot or tug boats. 2. The step of service: The accosting of the ships is ensured by the operational sections of the Harbor Company of Bejaia, such as the sections of piloting and towing.

5 Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal

55

Fig. 5.1 (a):Plan of the terminal. (b): Plan of the model of treatment of the containers

3. Vessel handling: It consists in the unloading of the containers. That is carried out with the two gantries of quay which have carriages being able to raise the containers from the container ships, and to put them on tractors. 4. The transfer: Once the container unloaded on the tractors, it will be transported towards the zone of storage. 5. Storage: The transferred containers are arranged, piled up and stored in the park of containers. 5.2.2.2 The Loading Process The process of loading is the opposite to the process of unloading. 1. Step of anchorage: Same procedure as in the case of unloading. 2. Step of storage: The RTG (Rubber Tyre Gantry) puts the containers on the tracktors . 3. Step of transfer: The trucks transport the containers beside the container ship.

56

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

4. Handling step: The gantry of a quay raises the container to put it on board the ship. 5. Step of service: The operational service of the EPB escorts the ship till the roads to leave the Bejaia harbor’s. 5.2.2.3 The Delivery and Restitution Processes 1. Deliveries: The delivery concerns the full containers or discharged goods. The means used to perform this operation are: RTG, trucks, stacker and forklifts if necessary. 2. Restitution of the containers: At the restitution of the containers, two zones are intended for the storage of the empty container, one for the empty containers of 20 units and the other for the empty containers of 40 units.

5.3 Mathematical Models We present in this section, the mathematical models corresponding to each system. We regarded the containers as being the customers. We impose the following assumptions: • Only one gantry in the quay for a ship. • The service duration of a truck is the sum of three variables: Duration of moving the quay till a full stock, duration of unloading of the truck by the RTG and duration of return of the truck to the quay station. • No difference between the containers. They are measured in ETU. The model of the unloading process can be represented in the diagram 5.2.a and the model of delivery process is given in the diagram 5.2.b. Remarks • The arrival of a ship represents the arrival of a containers group of a random size. • The quay is composed of two stations with size one (the queue size is limited to a random size in the sense of containers). • The treatment by the gantry is made container by container.

5.4 Calculation of the Forecasts The evolution of the number of containers handled in ETU is presented in the graph (Fig.5.3.a). It is noted that in the year 2007, BMT company treated 100000 ETU. Its objective for the year 2008 was to treat 120000 ETU. In March 2008, a calculation of forecast had been carried out. The designed series is the number of containers treated (loaded/unloaded) in ETU. The used data are

5 Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal

57

Fig. 5.2 (a): Diagram of the model of the unloading process. (b): Diagram of the model of the storage process.

monthly collected and are held forth over a period of two years (from January 2006 to March 2008). The method used for calculation of the forecasts is the exponential smoothing method (Blondel, 2002).

Fig. 5.3 (a): Evolution of the number of containers treated in ETU/year, (b): Original series and forecasts of the number of containers to be treated in ETU in the year 2008

The graph (Fig.5.3.b) represents the original series of the number of containers in ETU, as well as the forecasts (from April to December 2008). It is thus noted that the objective that BMT company had fixed at the beginning of the year was likely to be achieved.

58

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

In the same situation, we completed the same work for the year 2009. The objectives of BMT company correspond to the treatment of 130254 ETU over the year. Calculations of the forecasts are presented in table 5.1. Table 5.1 Forecast of the year 2009 in ETU Month Objective of BMT Forecast of the model 1 2 3 4 5 6 7 8 9 10 11 12

10600 10730 10440 10014 10900 10750 10900 10650 10900 10570 11650 12150

10019 10468 10591 10714 10838 10962 11085 11209 11332 11456 11579 11703

Total

130254

131956

5.5 Performance Evaluation of the BMT Terminal First of all, we will carry out a statistical analysis to identify the model of network of queues which correspond to our system.

5.5.1 Statistical Analysis and Identification of Models The results of the preliminary statistical analysis (estimate and adjustment test) on the data collected for the identification of the parameters of the processes are summarized in table 5.2. According to this preliminary analysis, one concludes that the performance evaluation of the terminal of Bejaia is really a complex problem. Indeed, the system is modeled by a network of unspecified queues, because it consists of queues of type G/G/1, M [x] /G/1, with blocking,...Therefore, we cannot use analytical methods (as for the Jackson networks or BCMP) to obtain the characteristics of the system. The models are: 1. Unloading process

5 Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal

59

Table 5.2 Results of the statistical analysis on the collected data Process

Variable

Law

Parameters of the law

Inter-arrivals of the ships to be loaded

Exponential

1/λ = 0.000424

Size of groups to be loaded

Geometric

p = 0.0046

service duration of the gantries of quay

Normal

μ = 3.0044 and σ 2 = 1.5314062

Service duration of the trucks

Normal

μ = 6.5506 and σ 2 = 6.10435884

Inter-arrivals of the ships to be unloaded

Exponential

1/λ = 0.0005389

Size of groups to be unloaded

Geometric

p = 0.0047

Service duration of the gantries of quay

Normal

μ = 3.0044 and σ 2 = 1.5314062

Service duration of the trucks

Normal

μ = 3.0044 and σ 2 = 1.5314062

Size of groups of delivered containers/day

Uniform

Mean=121

Size of groups of restored containers/day

Uniform

Mean=126

Loading

Unloading

Storage

./G/1 ./G/m −→ .. . Arrivals −→ M [X] /./. ./G/1 −→ Departure

−→ Fig. 5.4 (a): Modeling of the unloading process 2. Loading process Arrivals of the customers of the type 1 −→ M/./.

Arrivals of the customers of the type 2 −→ D[X] /./.

./G[X] /m .. .

−→ −→

Fig. 5.4 (b): Modeling of the loading process 3. Delivery process M [X] /G[X] /1 ./G[X] /1 −→ −→ Departure Arrivals −→ Fig. 5.4 (c): Modeling of the delivery process

./G/1 | | −→ Departure | ./G/1

60

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

4. Restitution process Arrivals of the customers of the type 1 −→ M/./.

Arrivals of the customers of the type 2 −→ D[X] /./.

/G[X] /1 −→ Departure

|

Fig. 5.4 (d): Modeling of the restitution process

In the case of the loading and restitution models, the servers will be in service only if there is at least a customer of type 1 in the first queue which will fix the size of the group to be treated. Otherwise, they will remain in a state of idleness even if there are customers of type 2 in the second queue.

5.5.2 Analytical Results Performances evaluation of systems aims to obtain the numerical values of some of their characteristics. These performances are calculated from the sample which enabled us to adjust the arrivals and services laws. The principal performances are summarized in table 3.

Table 5.3 Performances of the processes Process

Performance characteristics Mean number of ships to be loaded/day

Loading

Inter-arrival mean (day) Mean size of groups to be loaded

Value 0.6104633 1.6381 214.5278

Mean number of ships to be unloaded/day 0.7761129 Unloading Inter-arrival mean (day) Mean size of groups to unload Delivery

1.2884722 218.2174

Mean number of delivered containers/day 120.9000

Restitution Mean number of restored containers/day

125.8974

Gantry

Mean duration of service (minutes)

3.0044

Truck

Mean duration of service (minutes)

6.5506

Interpretation: According to the obtained results (table 5.3.):

5 Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal

61

•

It is noticed that there is a mean of 0.6104633 ships which accost to the Bejaia harbor’s in order to be loaded into containers by BMT company and 0.7761129 ships for the unloading. • The ships to be loaded make a request of 214 ETU in mean and the BMT makes the unloading of 218 ETU in mean by each ship. • The mean number of delivered containers each day is n3 = 120.9000 ETU. • The mean number of restored containers each day is n4 = 125.8974 ETU. Because of the complexity of the global model, it is not possible to calculate some essential characteristics analytically. This is why we will call upon the simulation approach.

5.5.3 Simulation We designed a simulator for each model under the Matlab environment. After the validation tests of each simulator, their executions provided the results summarized in table 5.4. Table 5.4 Performances of the processes obtained by simulation Processes Performance characteristics

Loading

Mean number of loaded containers/month

4299.85

Mean number of loaded ships/month

20.0433

Mean number of ships in roads

0.0742

Mean number of ships in the quay

1.3925

Mean number of unloaded containers/month

5385.71

Mean number of unloaded ships/month

24.6808

Unloading Mean number of ships in roads

Storage

Value

0.0533

Mean number of ships in the quay

1.9308

Mean number of full containers in the park

3372.9

Mean number of empty containers in the park 211.1208

Interpretation: The results of simulation show that the total number of containers loaded during one year will be of 51598.20 ETU and the mean number of ships in roads and in the quay are respectively of 0.0742 and 1.39 ships, the total number of loaded ships during one year will be of 240.52 ships. Concerning the unloading process, the total number of containers unloaded during one year will be of 64628.52 ETU, the mean number of ships in roads and in the quay are respectively of 0.0533 and 1.9308, the total number of ships unloaded

62

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

during one year will be of 296.17 and the total number of containers which will be handled for the year 2008 will be of 116226.72 ETU. Concerning the parks of storage, the mean number of containers in the full park will be of 3372.9 ETU and the mean number of containers in the empty park will be of 211.1208 ETU.

5.6 Variation of the Arrival Rate In order to study the behavior of the system in the case of variation of the arrival rate of the ships to be loaded and unloaded, other executions have been carried out. We have increased the number of ships at the loading and the unloading of 30%. The number of ships passes from 0.6104633 to 0.7936 per day for the loading and from 0.7761129 to 1.0089 per day for the unloading. The obtained results are summarized in table 5.5. Table 5.5 Performances of the processes in the case of increase of 30% of the number of ships arriving at the Bejaia harbor’s obtained by simulation Process

Loading

Performance characteristics

Value

Mean number of loaded containers/month

5458.30

Mean number of loaded ships/month

25.4433

Mean number of ships in roads

0.09230

Mean number of ships in the quay

1.40000

Mean number of containers unloaded /month 6958.04 Mean number of ships unloaded/month Unloading Mean number of ships in roads

Storage

31.8858 0.06690

Mean number of ships in the quay

1.90000

Mean number of full containers in the park

4874.20

Mean number of empty containers in the park 154.9814

Interpretation: With an increase of 30% of the rate of ships arriving at the Bejaia harbor’s, we note that the mean number of ships in roads and in the quay will increase a little. This means that the materials available within BMT company are sufficient to face this situation. In other words, an increase of 30% does not generate a congestion of ships in roads or in the quay. On the other hand, the mean number of handled containers will undergo a remarkable increase equivalent to 30000 ETU. This increase will not have any influence on the full stock or the empty stock. Indeed, they will pass respectively from 3372.9 ETU to 4874.2 ETU and from

5 Forecast of the Traffic and Performance Evaluation of the BMT Container Terminal

63

211.1208 to 154.9814 ETU, say from 41% to 59% for the full park and from 24% to 18% for the empty park.

5.7 Conclusion The objective of this work is to analyze the functioning of the park of containers of the BMT company in order to evaluate its performances, then to foresee the behavior of the system in the case of increase of the arrivals flow of the container ships. For this, we divided the system into four independent sub-systems: the “loading”, the “unloading”, the “full stock” and the “empty stock” processes. Each system is modeled by an opened network of queues and a simulation model of the functioning of each system could be established. The goal of each simulator is to reproduce the functioning of the park with containers. The study shows that the park with containers will have the possibility of handling 116226.72 ETU, say 51598.20 ETU at loading and 64628.52 ETU at unloading and a mean number of 3372.9 ETU in the park, for entry rates of 0.6104 ships per day for the loading process and 0.7761 ships per day for the unloading process. After that, a variation of the arrivals rate of the ships was proposed with an aim of estimating its influence on the performances of the system. With an increase of 30% of the number of ships arriving at the Bejaia harbor’s, we note a small increase in the mean number of ships in roads and in the quay. On the other hand, there will be a clear increase in the total number of treated containers which will pass from 116226.72 ETU to 148996.08 ETU including 65499.6 ETU at loading and 83496.48 ETU at unloading. We also note an increase in the mean number of containers in the full park which will pass from 3372.9 to 4874.2 ETU. Regarding the number of ships, it will pass from 240.52 to 305.3 ships at loading and from 296.17 to 382.63 ships at unloading. It would be interesting to achieve this work, by discussing the following items: • An analytical resolution of the problem. • Determination of an optimal management of the machines of the BMT company. • Variation of other parameters.

References Ayache N, Hidja R, A¨ıssani D, S A (2007) Evaluation des performances du parc a` conteneurs de lentreprise bmt. rapport de recherche no. 3/2007. Tech. rep., D´epartement Recherche Op´erationnelle-Universit´e de B´ejaia. Blondel F (2002) Gestion de la production. Dunod Edition, Paris David M, Michaud JC (1983) La pr´evision: approche empirique dune m´ethode statistique,. Masson Edition, Paris

64

D. A¨ıssani, S. Adjabi, M. Cherfaoui, T. Benkhellat and N. Medjkoune

De Werra D, Liebling TM, Heche JF (2003) Recherche Op´erationnelle pour ing´enieurs, Tome 2. Presses Polytechniques et Universitaires Romandes Gross D, Harris CM (1998) Fundamentals of Queuing Theory. Wiley Series in Probability and Statistics Pujolle G, Fdida S (1989) Mod`eles de Syst`emes et de R´eseaux, Tome 2,. Editions Eyrolles Ruegg A (1989) Processus Stochastiques. Presses Polytechniques et Universitaires Romandes Sait R, Zerrougui N, Adjabi S, A¨ıssani D (2007) Evaluation des performances du parc a` conteneurs de lentreprise portuaire de b´ejaia. In: Proceedings of an International Conference Sada07 (Applied Statistics for Development in Africa), Cotounou (Benin).

Chapter 6

A Dynamic Forecasting and Inventory Management Evaluation Approach Johannes Fichtinger, Yvan Nieto and Gerald Reiner

Abstract A common strategy for companies to hedge unpredictable demand and supply variability is to constitute safety stocks as well as safety capacity. However, classical safety stock calculations, often used in practice, assumed demand and lead time to be identical and independent distributed each, which is generally not true when considering empirical data. One cause for this problem can be the misspecification of the demand forecasting model, e. g. if a standard, additive linear regression model is used to describe heteroscedastic demand. While for a stationary demand process the amount of historical data i. e. the number of periods used for estimation of the process variability does not affect the computation, this no longer holds when using empirical data. In this study, we used a two-stage supply chain model to show that in a non-stationary setting the number of observation periods highly influence the supply chain performance in terms of on-hand inventory, fillrate and bullwhip effect. Also, we use the efficiency frontier approach to provide a single performance measure and further analyse our results.

6.1 Introduction Increasing competition leads companies in a lot of industries to pay more attention to customer satisfaction. Being able to fulfill customer orders with the “right” Johannes Fichtinger Institute for Production Management, WU Vienna – Nordbergstraße 15, A-1090 Wien e-mail: [email protected] Yvan Nieto Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected] Gerald Reiner Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected]

65

66

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

service level is crucial for customer satisfaction. Companies have to carefully adapt their delivery time to customer requirements and be prepared to cope with unplanned variation in demand as well as supply to prevent stock outs. In the context of maketo-stock manufacturing strategy, a common solution to hedge unpredictable demand and supply variability is to constitute safety stocks. This approach is widely used in practice and often relies on a classical calculation that integrates both demand and supply lead time means and standard deviations. A critical point to mention here is that the safety stock calculation assumes a stationary demand process such that the two random variables, demand and lead time, are both assumed to be identical and independent distributed each. Unfortunately, considering empirical data, demand process decomposition does not necessarily show these properties and, as a consequence, this calculation leads to volatile results. While for a stationary demand process the amount of historical data i. e. the number of periods used for estimation of the process variability does not affect the computation, this no longer holds when using empirical data. Often ignored, these points may reveal to be critical as it may impact the supply chain dynamics and lead to inappropriate inventory levels as well as service levels. The aim of this work is to present a dynamic two-stage supply chain model of a supplier and a retailer with focus on the retailer. In particular, for the retailer, we consider a periodic review inventory replenishment model, where the demand distribution is not known. Hence, the retailer uses demand forecasting techniques to estimate the demand distribution. For the supplier’s manufacturing process we assume a pure make-to-order production strategy subject to limited capacity, where orders are processed based on a strict first-in, first-out priority rule. Considering that the supply chain evaluation has to be product- and customer-specific we use an empirical reference dataset of a retail chain company to discuss our research question. We show how unstable forecast errors impact supply chain performance through its implication on order-up-to level calculation. Specifically, we build a process simulation model and measure the effect of the number of periods used in demand estimation on the performance of the supply chain. Hence, the independent variable is the number of past periods the retailer considers for calculating the mean and variance of demand. The performance measures, the dependent variables, are average on-hand inventory, the bullwhip effect as the amplification between demand variance and order variance and the fillrate as a service level criterion. Moreover, we consider the effect of manufacturing capacity (upper limit of the throughput rate) on these measures. To reduce the multi-criteria based performance measurement, we use the efficiency frontier approach to provide a single performance measure. Since our aim is to consider many aspects of a supply chain, the relevant literature is vast. Even if we use a simple inventory policy, we refer the interested reader for a comprehensive review on inventory models to Silver and Peterson (1985), Zipkin (2000) and Porteus (2002), and especially for the multi-echelon models to Axs¨ater (2006). The classical optimization approaches in inventory management are focusing on minimization of the total inventory system cost (Liu and Esogbue, 1999). A fundamental problem in this context is the “right” estimation of costs. This problem

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

67

is mentioned also by Metters and Vargas (1999), i. e., classically, different performance measures are converted into one monetary performance measure. Therefore, these authors suggested applying data envelopment analysis to be able to take different performance measures into consideration. In general it has to be mentioned that multi-criteria optimization as well multi-objective decision making problems have been solved in many areas. Surprisingly, till now only a couple of papers have been published in the field of inventory management (see also Maity and Maiti, 2005). One of the performance measures that we consider, the bullwhip effect (Lee et al, 1997a,b; Sterman, 1989), gained significant interest of many researchers. A pointed definition of the bullwhip effect is provided by de Kok et al (2005): “The bullwhip is the metaphor for the phenomenon that variability increases as one moves up a supply chain”. Different approaches to identify the causes of the bullwhip effect have been made so far. Lee et al (1997b, 2004) describe four fundamental causes; demand signalling processing, price variations, rationing games and order batching. While the latter three are not considered in this work, the demand amplification due to the combined effects of demand signalling processing and non-zero lead times are a main focus of this work. In a work on the interface of the forecasting and replenishment system with focus on the bullwhip effect, Chen et al (2000b) use a two stage supply chain model and consider the dependencies between forecasting, lead times and information in the supply chain. In their model, the retailer does not know the distribution of demand and uses a simple moving average estimator for mean and variance of demand. Similar two-stage supply chain model have also been used by e. g. Boute et al (2007) to successfully study dynamic impact of inventory policies. The literature on the efficiency frontier approach for performance/efficiency measurement is vast after the seminal work of Charnes et al (1978). An excellent recent review can be found in Cook and Seiford (2009). Dyson et al (2001) discuss the problems of factor measurement related with percentage values, as e. g. the fillrate in our approach. The remainder of this paper is organized as follows. Section 6.2 introduces the basic supply chain model for a single supplier and a single retailer using demand forecasting. In section 6.3 we present simulation results based on numerical data and empirical examples. Section 6.5 contains further extensions to the current model and conlcluding remarks.

6.2 A Simple Supply Chain Model with Demand Forecasting Consider a simple supply chain consisting of a single retailer and a single manufacturer. The retailer does not know the true distribution of customer demand, so he uses a demand forecasting model to estimate mean and variance of demand. In each period, t, the retailer checks his inventory position and accordingly places an order, qt to the supplier. After the order is placed, the retailer faces random costumer demand, Dt , where any unfulfilled demand is lost. There is a random lead time, L,

68

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

such that an order placed at the beginning of period t arrives at the beginning of period t + lt , where lt denotes the random realization of the order lead time placed in t. We assume that the retailer uses a simple order-up-to policy based on demand forecasting methods using regression analysis. We use aggregated weekly empirical sales data of about 220 periods (approx. 4 years) from 01/2001 to 04/2005 to estimate demand Dt for specific products. The data do not only contain sales information (units sold) but also gross price pt , stock available, number of market outlets Ot , which needs to be considered in an expanding company and a features indicator Ft (binary information to account for the effect of advertisement, e. g. by means of newspaper supplements as flyers and leaflets). To clean the data from these effects and additionally from trend and seasonality, we use a least squares regression model as proposed by Natter et al (2007). Dt = β0 + β1 pt + β2t + β3 sin

2t π 52

+ β4 cos

2t π 52

+ β5Ot + β6 Ft + et

(6.1)

Note that the sales data do not necessarily correspond to the underlying real demand process since demand during stockouts is not recorded. However, an analysis of stockout situations on the real data show that they occur in less than 2% of the selling periods. Therefore, we take the existing sales data as censored information to for demand. We tested the assumptions related with classical linear regression models for the cleaning model (see e. g. Greene, 2008, for a comprehensive discussion). There is no exact linear relationship for the independent variables in the regression (full rank assumption), and the independent variables are found to be exogenous. On the Contrary, the assumption of homoscedasticity and nonautocorrelation was not fulfilled for many products. An earlier study on the same data by Arikan et al (2007) shows that for many products a nonlinear relationship between price and demand such as (6.2) Dt = a · pt−b · et for a > 0, b > 1 could be found better explaining the pricing effect. As a consequence, for such products estimating an additive demand model as (6.1) leads to a decreasing variance of the error term, ε , in price. Hence, Var(ε |pt ) is not independent of price anymore. These in practical demand forecasting and replenishment problems inevitably occurring effects destroy the common stationarity assumption on the demand error term and, hence, are the focus of the subsequent analysis. Supplier

Retailer

Orders

x Production capacity Delivery

x Base stock policy x Target fillrate x Forecast accuracy x Observation periods

Fig. 6.1 Two stage supply chain model

Demand

Customer x Demand characteristics

Sales

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

69

As shown in Fig. 6.1, similar to the model of Chen et al (2000a) the retailer follows a classical infinite horizon base stock policy using weekly replenishments, where the order-up-to point St is estimated based on the expected demand for the actual period, μt , and an estimate for the standard deviation of the (1 + L) periods demand forecast error, σˆ t1+L as St = (1 + λt )μt + zt σˆt1+L ,

(6.3)

where the safety factor, zt , is chosen to meet a certain target fillrate, FR, service measure. In particular, since any unsatisfied customer demand is lost, zt is found such that it satisfies Rμt 1 − FR , (6.4) G(zt ) = · σt FR where G(·) denotes the standard normal loss function. The supply lead time the retailer faces is stochastic, where the corresponding random variable, L, has mean λt and standard deviation υt . It is well-known that for the case of fixed order costs, an (s, S) policy is optimal, however, we do not consider fixed order costs as we are interested in the effect of forecasting and the order-up-to level on the performance measures. Note that the order-up-to point in (7.1) is calculated based on the standard deviation of the (1 + λt ) period forecast error σt1+L and its estimator σˆ t1+L rather then the standard deviation of the demand over (1 + λt ) periods. As Chen et al (2000a) point out very clearly, using σˆt1+L captures the demand error uncertainty plus the uncertainty due to the fact that dt+1 must be estimated by μt+1 . Finally, defining an integer nt = max{n : n ≤ λt , n ∈ Z} helps to express the actual demand error observation, et1+L , as nt

et1+L = dt − μt + ∑ (dt+i − μt+i ) + (λt − nt ) (dt+nt +1 − μt+nt +1 ).

(6.5)

i=1

Based on the random variable of the demand error, ε 1+L , in (1 + L) periods, the estimator in period t of the standard deviation of the past demand errors can be calculated as σˆt1+L = Var(ε 1+L ) + υt2(μt λt )2 . (6.6) For the supplier’s manufacturing process we assume a pure make-to-order production strategy, where orders are processed based on a strict first-in, first-out basis. While the period length is one week for the retailer, the supplier is assumed to deliver at the end of the day the order is completely produced. We consider production of the supplier taking place on at most five days a week, hence, the supply lead time, L, can take values, l, such that 5l ∈ Z, so that l ∈ {0.2, 0, 4, 0.6, . . .}. The supplier has a fixed capacity C available solely for the retailer under consideration. For this very reason the retailer faces lead time variation, but due to missing information sharing with the supplier the retailer does not consider the supply as capacitated, and uses uncapacitated stochastic lead time models for replenishment. The lead time observation li for an order placed in i can be defined as

70

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

li = pi + wi ,

(6.7)

where pi = min{pi : pi qi ≥ C, 5pi ∈ Z} denotes the time necessary to complete the order qi and wi is the time during which the order was backlogged. Observe that the presented replenishment models assume the distribution of the supply lead time, L, to be independent and identical. However, due to the capacitated supplier, lead time is neither identical distributed since higher order quantities lead to stochastically longer lead times, nor is it independent due to the strict first-in first-out scheduling rule used by the supplier. The performance of the supply chain is evaluated against different measures: the ¯ the bullwhip effect as the amplification average on-hand inventory at the retailer, I, of the customer demand variance to the order demand variance, BW , and the average fillrate observed at the retailer, FR. All of these measures need to consider the available capacity at the supplier, C. While the on-hand inventory can be seen as a proxy for inventory related cost such as e. g. holding cost, the service level refers to the retailer’s “quality” of customer service, and the available capacity indicates the flexibility of the supplier within the supply chain. It is well-known from inventory theory that tradeoffs between these measures exits, e. g. the higher the average on-hand inventory, the (potentially) higher the service level and vice versa. Let xt be the quantity of goods received from the supplier at time t, then the retailer’s on-hand inventory, It , is It = max(It−1 + xt − dt , 0) for t ≥ 0,

(6.8)

and the corresponding lost sales vt are vt = − min(It−1 + xt − dt , 0) for t ≥ 0.

(6.9)

The observed average fillrate at the retailer can be calculated as FR = 1 −

T vt ∑t=1 , T D ∑t=1 t

(6.10)

and, finally, the bullwhip effect is measured as BW =

cout , cin

(6.11)

where cin and cout denote the coefficient of variation of the demand and the retailer’s order quantity , respectively (Fransoo and Wouters, 2000). Both coefficients can be easily calculated as SD(D) SD(q) , and cout = . (6.12) cin = D¯ q¯ Efficiency measurement in organizations has been the focus of numerous research activities, especially after the appearance of the seminal work of Charnes et al (1978) on the efficiency of decision making units.

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

71

To be able to measure the efficiency of the forecasting system and the inventory policy, we consider a set of n simulation runs j = {1 . . . n}. The performance of each run is calculated as the efficiency of the supply chain by the ratio of the weighted fillrate in per cent, FR j , to the average on-hand inventory, I¯j , uFR j . vI¯j

(6.13)

This performance/input ratio is basis for calculation of the standard engineering ratio of productivity and hence, the basis of the subsequent analysis. For our analysis, we use the extension of Banker et al (1984) of using a variable returns to scale (VRS) model such that the efficincy is defined as u FR j − u0 . v I¯j

(6.14)

Due to unknown multipliers u and v , Charnes et al (1978) and Banker et al (1984) in their extension proposed to solve the following linear programming problem. In particular, for the observation j = 0 under consideration, solve e∗0 = max u FR0 s.t. v I0 = 1 u FR j − u0 − v I j ≤ 0 j = 1, . . . , n u ≥ δ , v ≥ δ , u0 unrestricted.

(6.15)

Fig. 6.2 shows a geometrical representation of the efficiency frontier problem. By solving this input oriented model in (6.15) for each of the observations, it refers to projecting every observation to the left on the frontier. As example, in the case of the observation Obs 1, the projection is represented by the point B. The efficiency of Obs 1 is calculated as the ratio of the distances B0/A0.

6.3 Model Verification and Validation Based on our approach presented before, we will provide examples and validation for the analysis. First, we will consider artificialy generated datasets in order to validate the model against theory. Artificially generated samples were based on a normal distribution of the form N (250, 50) and used to evaluate the performance of the supply chain under 120 scenarios. Each scenario is related to a specific capacity at the supplier as well as a specific number of observation periods. Specifically, we use 5 distinct capacities, i.e. 1.1, 1.25, 1.5, 1.8 and 2.5 times the average demand of the dataset, and 24 different numbers of observation periods, i.e. each second periods starting from 4 to 50. Figure 6.3 presents the results obtained for the 3 performance measures of interests, i. e. the average on-hand inventory at the retailer, the bullwhip effect and the fillrate. It can be noticed that results are generally in

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

Servicelevel: Fillrate 94 95 96

97

72

B

●● ● ● ● ●● ● ● ● ● ●● ● ●

A ● Obs 1●

●

●

● ●

●

93

●

175

●

180

185 OHI

190

Fig. 6.2 Calculation of the efficiency of a single observation based on the efficiency frontier

accordance with theory. Specifically, higher capacity lead to higher performance in terms of inventory and fillrate, as benefiting from shorter lead time. Bullwhip is here neglectable, however, it illustrates, as expected from theory, that no noticable impact from the number of observation periods on the performance measures is to be reported for this stationary setting. Nevertheless, it appears that a minimum of around 18 periods of observation is necessary to permit stable estimations. A last comment is on the discrepency between the observed fillrate and the target of 98% that we assume to be related to the non-normality of the lead time. This explanation is convicing, knowing that tests involving normally distributed lead time leads to fillrate in the range of 97.2% to 98.3 (for all capacities and with sufficient number of observation periods). Bullwhip

C = 1.1

Fillrate 97.0

1.10

OHI

96.0

1.06

180

C = 1.25

20

30 40 Nr Periods

50

20

30 40 Nr Periods

50

94.0

C = 2.5

95.0

1.02

C = 1.8

0.98

160

170

C = 1.5

20

30 40 Nr Periods

50

Fig. 6.3 OHI, Bullwhip and Fillrate of a stationary demand product, plotted as a function of the number of periods used for estimating the standard deviation of the demand error term. When using around 15 periods or more, the effect of the number of periods vanishes

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

73

20 40 Nr Periods

C=1.8

20 40 Nr Periods

0.970 0.980 0.990 1.000

20 40 Nr Periods

C=1.5

0.970 0.980 0.990 1.000

20 40 Nr Periods

C=1.25

0.970 0.980 0.990 1.000

C=1.1

0.970 0.980 0.990 1.000

0.970 0.980 0.990 1.000

Next, we used the performance results as input for the efficiency analysis and results are presented in Fig. 6.4. It can be observed that, if any, only marginal improvements are possible by increasing the number of observation periods. Nevertheless, it is worth to mention that the optimal is proper to each capacity setting and that, for now, no comparison is possible between the different efficiency analysis. Based on the later, we consider that our model match theory expectation and is therefore verified and validated for further analysis. C=2.5

20 40 Nr Periods

Fig. 6.4 Fillrate efficiency for stationary demand products. Using 20 periods or more leads to an negligible efficiency gap smaller 1%

6.4 Empirical Data Analysis In particular, we will use empirical data presenting non stationarities to illustrate our position. We will consider two datasets, one presenting a seasonal pattern (Fig.6.4a) and a second including a single strong peak in demand (Fig.6.4b). Both datasets present therefore nonstationarities wich could have different impact on the performance of the supply chain regarding the number of observation periods considered. Simulation was performed using the same scenarios structure we presented earlier (see section 3.1) and results are presented in Fig.6.6. It can be observed that for the seasonal data , results tend to reach stability once a sufficient number of periods is available (Fig.6.6a). However, it is worth to mention that first the number of observation periods required is higher in this setting, which can be argued in the sense that, by being more structured and more volatile, valide estimations will necessitate more information, i. e. more observation. Second, fillrate remains more volatile, even for high number of observation periods and the reason for it is assumed to be linked with intrinsic dynamic of the model. In this context, order sizes are more variable, which can lead to stronger bias in lead time distribution and impact fillrate. In the case of the second dataset, interesting results in terms of performance can be observed (Fig.6.6b). In this case, the hierarchy related to capacity is much unclear considering fillrate. The number of observation periods strongly impact the service performance and make optimal setting difficult to identify.

74

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

Product B

300

400

500

600

700

100 200 300 400 500 600 700

800

Product A

0

50

100

150

200

0

50

100

week

150

200

week

Fig. 6.5 Demand plots

Bullwhip

Fillrate

160

96.0

1.2 1.1 1.0

C = 1.5 C = 1.8

95.0

170

C = 1.25

C = 2.5

C = 1.1

210

C = 1.8

30 40 Nr Periods

50

30 40 Nr Periods

50

Fillrate

92

0.90

C = 2.5

20

93

1.05 0.95

230

C = 1.5

20

50

1.00

250

C = 1.25

30 40 Nr Periods

Bullwhip

1.10

270

OHI

20

94

50

91

30 40 Nr Periods

90

20

94.0

180

C = 1.1

97.0

190

OHI

20

30 40 Nr Periods

50

20

30 40 Nr Periods

Fig. 6.6 OHI, Bullwhip and Fillrate of non-stationary demand products A (up) and B (down)

50

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

75

The results from the efficiency analysis presented in Fig.6.7 confirmed the previous observations, i.e. the impact of the number of observation periods is limited in the case of the seasonal dataset (Fig.6.7a). However, the dataset including a strong peak in demand leads to erratic results regarding number of observation periods. In this last case, the highest number of periods did not lead to optimal anymore and the choice of the observation range can have strong impact on performance and this independently from the elected strategy.

0.98 0.90

20 40 Nr Periods

20 40 Nr Periods

0.98

C=2.5

20 40 Nr Periods

0.90

0.94

0.98

C=1.8

0.94 20 40 Nr Periods

C=2.5

0.94

0.98 0.90

0.94

0.98 0.94

C=1.5 0.98

20 40 Nr Periods

C=1.8

0.90

20 40 Nr Periods

20 40 Nr Periods

0.94

0.94

0.98

C=1.25

0.90

0.90

0.94

0.98

C=1.1

20 40 Nr Periods

0.90

20 40 Nr Periods

C=1.5

0.90

0.94

0.98

C=1.25

0.90

0.90

0.94

0.98

C=1.1

20 40 Nr Periods

Fig. 6.7 Fillrate efficiency for non-stationary demand products A (up) and B (down)

6.5 Conclusion We considered a dynamic two-stage supply chain model with focus on the retailer to identify the possible impact of the number of observation periods used to calculate the order-up-to level using an efficiency frontier approach. Based on this, we showed for the stationary demand case that as long as the number of periods is sufficiently large (here around 18 periods), it has no noticeable effect on the performance of the supply chain. However, considering non-stationary demand caused e. g. by a misspecification of the price dependency of demand in the demand forecasting model, the number of observation periods can lead to divergent results and considerably affect efficiency. Based on our results, we demonstrate that the impact of non stationarities when using classical safety stock calculation is highly influenced by the number of observation periods considered. In addition, as it is not possible to know ex-ante which

76

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

periods number is optimal (neither do we know if such evaluation is possible), the danger of non stationarity should be highlighted. Here we discuss a few potential extensions to the current model. First, as mentioned in section 6.2, the retailer has no information about the cause of the lead time variability because of lacking information sharing in the supply chain. However, the benefits of information sharing for both of the supply chain partners are well-known as described in Chen (2003). In particular, in the current model, the retailer does not know about and hence, does not consider the production capacity at the supplier. On the contrary, sharing the available capacity beween the supplier and the retailer allows the retailer to apply replenishment models intergrating capacitated suppliers. We refer the interested reader to Federgruen and Zipkin (1986a,b). They showed that under certain reasonable assumptions it is optimal to follow a base-stock policy when possible and to order capacity when the prescribed quantity would exceed the capacity. By turning the problem into a cost minimization problem, cost improvements due to explicit consideration of the capacitated supplier can be evaluated. Hence, the value of information sharing in this framework becomes measureable. A second extension can be considering supplier lead time as stochastic as in Ciarallo et al (1994). They show in their model that under stochastic lead times an order-up-to policy is still optimal. Using stochastic supplier capacity, a supplier with equal mean but increasing variance in the lead time is considered less reliable. By comparing the performance measures at the retailer the value of a more reliable supplier can be made explicit. Thirdly, to fully support decision making, the efficiency frontier approch could be extended. Assuming that extra capacity has a cost, flexibility at the supplier should be integrated in the analysis to contribute to the evaluation. Also, each input could be linked to limits and weights related to the specificity of the supply chain under study in order to provide more realistic evaluation. Finally, consideration of supply chain contracts (see e. g. Cachon, 2003, for a comprehensive discussion) might help gain additional insights in the value of information sharing, the value of capacity and reliability of the supplier and the impact of the forecasting and replenishment model used by the retailer.

References Arikan E, Fichtinger J, Jammernegg W (2007) Evaluation and extension of single period combined inventory and pricing models. In: Proceedings of the 14th International Annual EurOMA Conference, Ankara, Turkey Axs¨ater S (2006) Inventory Control. Springer Verlag Banker RD, Charnes A, Cooper WW (1984) Some models for estimating technical and scale inefficiencies in data envelopment analysis. Management Science 30(9):1078 – 1092 Boute RN, Disney SM, Lambrecht MR, Van Houdt B (2007) An integrated production and inventory model to dampen upstream demand variability in the supply

6 A Dynamic Forecasting and Inventory Management Evaluation Approach

77

chain. European Journal of Operational Research 178(1):121–142 Cachon GP (2003) Supply chain coordination with contracts. In: de Kok A, Graves S (eds) Supply Chain Management: Design, Coordination and Operation, Handbooks in Operations Research and Management Science, vol 11, pp 227–339 Charnes A, Cooper W, Rhodes E (1978) Measuring the efficiency of decision making units. European Journal of Operational Research 2(6):429–444 Chen F (2003) Information sharing and supply chain coordination. In: de Kok T, Graves S (eds) Supply Chain Management: Design, Coordination, and Operation, Handbooks in Operations Research and Management Science, vol 11, Elsevier Chen F, Ryan J, Simchi-Levi D (2000a) The impact of exponential smoothing forecasts on the bullwhip effect. Naval Research Logistics 47(4):271–286 Chen YF, Drezner Z, Ryan JK, Simchi-Levi D (2000b) Quantifying the bullwhip effect in a simple supply chain: The impact of forecasting, lead times, and information. Management Science 46(3):436–443 Ciarallo F, Akella R, Morton T (1994) A periodic review, production planning model with uncertain capacity and uncertain demand-optimality of extended myopic policies. Management Science 40(3):320–332 Cook WD, Seiford LM (2009) Data envelopment analysis (DEA) - thirty years on. European Journal of Operational Research 192(1):1–17 de Kok T, Janssen F, van Doremalen J, van Wachem E, Clerkx M, Peeters W (2005) Philips electronics synchronizes its supply chain to end the bullwhip effect. Interfaces 35(1):37–48 Dyson RG, Allen R, Camanho AS, Podinovski VV, Sarrico CS, Shale EA (2001) Pitfalls and protocols in dea. European Journal of Operational Research 132(2):245–259 Federgruen A, Zipkin P (1986a) An inventory model with limited production capacity and uncertain demands I. the average-cost criterion. Mathematics of Operations Research 11(2):193–207 Federgruen A, Zipkin P (1986b) An inventory model with limited production capacity and uncertain demands II. the discounted-cost criterion. Mathematics of Operations Research 11(2):208–215 Fransoo JC, Wouters MJ (2000) Measuring the bullwhip effect in the supply chain. Supply Chain Management: An International Journal 5(2):78–89 Greene WH (2008) Econometric Analysis, 6th edn. Pearson Prentice Hall, Upper Saddle River, New Jersey, US Lee HL, Padmanabhan V, Whang S (1997a) The bullwhip effect in supply chains. Sloan management review 38(3):93–102 Lee HL, Padmanabhan V, Whang S (1997b) Information distortion in a supply chain: The bullwhip effect. Management Science 43(4):543–558 Lee HL, Padmanabhan V, Whang S (2004) Comments on information distortion in a supply chain: The bullwhip effect. Management Science 50(12):1887–1893 Liu B, Esogbue A (1999) Decision criteria and optimal inventory processes. Kluwer Academic Publishers

78

Johannes Fichtinger, Yvan Nieto and Gerald Reiner

Maity K, Maiti M (2005) Numerical approach of multi-objective optimal control problem in imprecise environment. Fuzzy Optimization and Decision Making 4(4):313–330 Metters R, Vargas V (1999) A comparison of production scheduling policies on costs, service level, and schedule changes. Production and Operations Management 8(1):76–91 Natter M, Reutterer T, Mild A, Taudes A (2007) Practice prize report – an assortment-wide decision-support system for dynamic pricing and promotion planning in DIY retailing. Marketing Science 26(4):576–583 Porteus EL (2002) Foundations of Stochastic Inventory Theory. Stanford University Press Silver E, Peterson R (1985) Decision systems for inventory management and production planning. Wiley, New York et al. Sterman JD (1989) Modeling managerial behavior: Misperceptions of feedback in a dynamic decision making experiment. Management Science 35(3):321–339 Zipkin PH (2000) Foundations of Inventory Management. Shelstad, Jeffrey J.

Chapter 7

Performance Evaluation of Process Strategies Focussing on Lead Time Reduction Illustrated with an Existing Polymer Supply Chain Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

Abstract The ability to fulfil customer orders is crucial for companies which have to operate in agile supply chains. They have to be prepared to respond to changing demand without jeopardizing service level, i. e. delivery performance is the market winner (Christopher and Towill, 2000; Lee, 2002). In this context, lead time reduction (average as well as variability) is of key interest since it allows increasing responsiveness without enlarging inventories. In front of these possible levers (e. g. Chandra and Kumar (2000), the question arises of the dynamic assessment of potential process improvements for a specific supply chain and moreover a combination of potential process improvements related to an overall strategy (responsive, agile, etc.). Using process simulation, we demonstrate how the coordinated application of strategic supply chain methods improves performance measures of both intra- (lead time) and interorganizational (service level) targets.

7.1 Introduction The intention of this study is to analyse and assess the effects of shortening lead time, i. e., average as well as variability, on the performance of the entire supply chain (delivery service, delivery time, cost, etc.). There are a great number of different strategic/tactical supply chain approaches (Chandra and Kumar, 2000; Mentzer Dominik Gl¨aßer Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected] Yvan Nieto Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected] Gerald Reiner Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected]

79

80

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

et al, 2001) that make it possible to improve the supply chain processes by means of, e. g., demand forecast (see Winklhofer et al (1996) for a review), capacity management, organizational relations, better communication, reduction of supply chain echelons and adapted inventory management. Also, the possibility of moving the customer order decoupling point has been recognized (Olhager, 2003), opening the door to postponement strategies, etc.. In front of these possible levers, the question arises of the dynamic assessment of potential process improvements for a specific supply chain and moreover a combination of potential process improvements related to an overall strategy. Supply chain evaluation is of primarily importance in order to support decision making. We will demonstrate that these theoretical concepts as well as the related restrictions have to be modified under consideration of “real” processes. Therefore, the question arises if these concepts are robust enough to improve also “real” processes. The investigations will be carried out on the basis of quantitative models using empirical data (Bertrand and Fransoo, 2009). Basically, with the quantitative examination of empirical data a model will be developed which reproduces causal correlations between the control variables and the performance variables. Furthermore, Bertrand and Fransoo (2002) pointed out that this methodology offers a great opportunity to further advance the theory. According to Davis et al (2007) the choice of simulation technology is an important decision when it comes to achieving the research objective. Thus, simulation models will be developed, e. g., discrete event simulation (Sanchez et al, 1996), since the possibility of understanding the supply chain as a whole and analyzing and assessing different strategic/tactical action alternatives offer a considerable benefit. This is why we have opted to use ARENA for developing the simulations models (Kelton et al, 2003). First, Section 2 discusses the effects of optimised lead time as regards supply chain performance. Furthermore, the importance of supply chain evaluation is emphasised. Then, in Section 3, we set out our research approach with the help of a polymer processing supply chain. Finally, Section 4 provides concluding remarks plus a look at other research possibilities.

7.2 Theoretical Background Cutting lead time is understood by experienced managers but is seldom awarded sufficient importance. Thus, it is that Little’s Law (Hopp and Spearman, 1996) says that cutting delivery time also cuts work in process. However, strategic advantages such as improving service level or cost advantages, can also be achieved. If one considers the classic formula for calculating the safety stock that is used as part of a reorder point replenishment policy, (7.1) Is = z μτ υ 2 + λτ στ2 it is evident that delivery time directly affects this (Silver et al, 1998). Here, μτ stands for the delivery time mean and στ2 for the delivery time variance. The other

7 Performance Evaluation of Process Strategies Focussing on Lead Time Reduction

81

parameters stand for demand mean (λτ ) and demand variance (υ 2 ) as well as the safety factor z, which represents a trade-off between service level and stock keeping costs, and Is for safety stock. Therefore there is a lot of interest in reducing the variance as well as average delivery time. On the one hand, this result in reduced safety stock that is reflected by lower stock keeping costs. On the other hand, this in no way worsens the service level, e. g. the number of stock outs. Thus the operational objective of a supply chain, i. e. increased customer satisfaction and lower costs at the same time, becomes more realistic. This can even turn out to be a strategic competitive advantage. A decisive element is the customer order decoupling point (CODP) (Mason-Jones et al, 2000). It is the point where the forecast-driven standard production, mostly serial production of standard components (PUSH), and the demanddriven production, i. e. commissioned production in response to customer orders or other requirement indicators (PULL) meet. Physically, the decoupling point in the supply chain is the ultimate inventory where components not yet relating to any order (Mason-Jones and Towill, 1999). The further downstream in the supply chain the decoupling point is, the less the quantities taken from the inventories agree with real demand at the point of sale (POS). Owing to the fact that most supply chain partners do not see real customer demand, they tend to be forecast-driven and not demanddriven (Christopher and Towill, 2000), which also enforces the so-called “bullwhip effect” (increasing fluctuations in order quantities and inventory up-stream in the supply chain whilst end customer demand remains constant (Lee et al, 2004)). To increase competitive advantage, Olhager (2003) determines that companies can either keep the CODP at its current position and reduce delivery lead time or maintain the delivery lead time and move the CODP upstream in order to reduce or clear stocks. Strategically positioning the CODP particularly depends on the production to delivery lead time (P/D) ratio and on relative demand volatility (RDV) (standard deviation of demand relative to the average demand). In this way, for example, a make to order (MTO) strategy can only be achieved if the P/D ratio is less than 1 (Olhager, 2003). This is because when production lead time is greater than the delivery lead time of a customer order, customer service of course suffers (Jammernegg and Reiner, 2007). On the other hand, it is not advisable to apply a make to stock (MTS) strategy (lead time is zero) if the RDV is very high because this results in huge inventories if customer service is to be maintained, and this of course results in high inventory costs. If, in this case, the P/D ratio is greater than 1, then some components would have to be produced for stock, which leads to an assembly to order (ATO) or an MTS strategy. The importance of lead time is also emphasised by Cachon and Fisher (2000), in that they assert that reducing lead time or batch size can affect supply chain performance more than information sharing. Likewise, in the 12 Mason-Jones et al (2000) for simplifying material flow cutting lead time is an important point.

82

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

7.2.1 Supply Chain Evaluation Evaluation of real supply chain processes is always challenging since a valid estimation can only be obtained through a detailed, specific process analysis. Improvements of a specific supply chain process can never be 100% applied (copied) to another setting. Nevertheless, they can be used as best practice indicating improvement potentials to another company / supply chain. This analysis must be product-specific as well as company-specific and the performance measures have to be selected carefully and in accordance with the specificity of the system under study (Reiner and Trcka, 2004). An important step in defining suitable performance measures is determining market qualifiers and market winners, which determine the alignment and therefore different metrics for leanness and agility of supply chain performance (Mason-Jones et al, 2000; Naylor et al, 1999). When drawing up the analysis and assessment model a product-specific supply chain design method should be selected in order to achieve results that are close to reality. This method provides for the fact that a supply chain always has to be designed in a product-specific and customerspecific way (Fisher, 1997) and that the alignment of the supply chain with regard to its leanness, agility or a combination of both (Lee 2002, Christopher and Towill 2000) plays a decisive role. If a supply chain already exists in reality, then the necessary data for the specified performance measures can be obtained by i. e. analysing existing IT systems as well as interviewing the supply chain partners. However, if alternative supply chain strategies have to be analysed in terms of their performance, then data is never available. In this case, missing values can be calculated, estimated or obtained by simulation. But calculation is often impossible and a general estimation is too imprecise (Jammernegg and Reiner, 2007). Dynamic stochastic computer simulations can provide not only average values for performance measures but also give information about their probabilistic distribution (Kelton et al, 2003) because of the use of random variables (Jammernegg and Reiner, 2007). Random variables, which simulate risks, are essential to reliable evaluations because, according to Hopp and Spearman (1996), risks negatively affect supply chain performance. To enable precise evaluation, the model must include all important process-related matters.

7.3 Illustration of the Supply Chain To illustrate the ”real” improvement potential of theoretical lead time reduction approaches, we analysed empirical data from a supply chain in the polymer as well as furniture industry. The supply chain is characterized by three levels, i. e. a supplier, a manufacturer and a sales office, and ends with a market-leading OEM as unique customer. In this case, delivery performance is the market winner and on-time delivery is therefore crucial to maintain customer loyalty. Due to the tremendous variety of products offered by the manufacturer (more than 50000), the analysis had to be limited to key articles. The selection of the product was performed using ABC-XYZ

7 Performance Evaluation of Process Strategies Focussing on Lead Time Reduction

83

analysis. This classification is an extension of the ABC analysis (Vollmann et al, 2004), because it not only takes the value into consideration, but also the variability of the demand (Sch¨onsleben, 2004). We opt for the most sold product, which has a coefficient of variation (standard deviation/mean) greater than 2/3. Thereby, this product represents the AY category (Reiner and Trcka, 2004).

7.3.1 Process Design The manufacturer, located in Western Europe, delivers goods to a sales office located in Eastern Europe. In turn, the sales office supplies the four OEM production plants (C1, , C4) belonging to the customer and also located in the eastern part of Europe. The entire procedure is set out in Fig. 1, with the sales office as well as the production sites arranged as in reality. In more details, the sales office uses its inventory to fulfil the customer orders. As soon as the inventory level at the sales office decreases down to a reorder point a stock replenishment order is placed at the manufacturer. The manufacturer must then supply the goods and send them to the sales office as fast as possible. No delivery time is specified. It is to be borne in mind that the manufacturer’s finished goods inventory merely serves as a buffer store for transport purposes (batching) and is thus not able to deal with any significant demand fluctuations because the manufacturing strategy is make to order for the manufacturer.

WLR WD RU U VS GH DQ 2U 7U Q

Fig. 7.1 The initial process

The sales distribution process is to be regarded as based on the classic push principle (make to stock). In a dynamic environment where there is uncertainty about demand and fluctuations in demand this make to stock strategy may lead to great

84

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

problems. Fig. 2 shows the stock movements at the sales office over a year. The diagram shows that there is an increase of stock outs during the first half of the year, and this has a negative effect on customer satisfaction. The problems associated with in this setting are manifold. (1) Owing to the irregular pattern of customer order placing, it is difficult for the sales office to produce a forecast for the future. (2) Furthermore, available information of the sales office is not sent promptly to the manufacturer. (3) There is a lack of transparency, the manufacturer is not aware of actual customer demand. Therefore, he is not able to discern whether or not there is a genuine customer order behind the stock replenishment order placed by the sales office. This frequently leads to unfavourable prioritisation of the production orders and this, in turn, sometimes results in long and varying delivery periods. (4) There is no classical replenishment policy used by the sales office, so that decisions concerning reorder points and order quantity are mostly made on a one-off basis by the sales office staff. 6WRFN0RYHPHQW

4XDQWLW\

'D\V

Fig. 7.2 Stock movements at the sales office

As already mentioned, when choosing performance measures, a special focus is on the total lead time as well as on costs and customer satisfaction (number of stock outs). The period between receipt of the order at the manufacturer’s sales office and the point when the goods are dispatched to the customer from the sales office inventory is described as the total lead time. Reducing lead time has financial implications as well as strategic effects (Gunasekaran et al, 2001). Table 1 sets out initial situation performance measures. Long lead times and large numbers of stock outs are particularly apparent. Stock keeping costs are correspondingly low owing to the many stock outs. In the following the representation of the initial situation based on actual historical data is called initial scenario.

7.3.2 A Simulation Study For our simulation model, we use discrete event simulation (ARENA). We apply the model to asses the performance of different supply chain settings as well as to

7 Performance Evaluation of Process Strategies Focussing on Lead Time Reduction

85

evaluate design alternatives. For each scenarios tested, replications were carried out in order to asses the variability and the robustness provided by each strategy (Reiner and Trcka, 2004). One simulation run is for a period of 365 days. Quantitative models based on empirical data are largely dependent on the data they integrate as well as on process design descriptions. These are necessary for making sure that the way the model works comes as close as possible to actual observations and processes. In order to obtain a coherent data base free of organisational barriers, the data triangulation approach was chosen (Croom, 2009). In particular, we looked at the existing IT systems at the plant, the sales office and the organisational administration departments. Based on direct data access we ensured that data could be directly acquired from the source using the database query. The model design was adapted in line with the product-specific supply chain method based on analyses and observations of reality, e. g. participant observations and questioning responsible supply chain managers. Product specification, according to Mason-Jones et al (2000), yielded that the market winner is the level of readiness to deliver, whereas quality, costs and lead time are the market qualifiers. This indicates an agile supply chain environment. For model validation, the initial scenario was simulated and the result data were compared with the real data. The comparison showed that the results of the model reflect the reality. Finally the completed model design including the simulation results were again confirmed through responsible supply chain managers and participant observations.

7.3.3 Scenario 1 - Forecast Based Inventory Management Based on interviews we figured out that a 4-week rolling forecast from customer could be provided, which will constitute the core alternative of our first scenario. The rolling forecast represents actual order entry with optional manual adoptions from the customer. In addition, and in order to support the impact of the forecast, an (s,S) inventory policy will be applied at the sales office, with a safety stock calculated as in eq. 1 with a target cycle service level of 95%. The order quantity also takes the manufacturer’s batch size into account. All applied distributions for stochastic input variables (e. g. delivery time between manufacturer and sales offices incl. production time, transport cost) were worked out on the basis of real data, taking account of chi-square and Kolmogorov-Smirnov goodness-of-fit hypothesis tests. In addition, all distributions have been validated in a graphical evaluation. As it has not yet been possible to estimate the precision of the customer’s forecast, it was assumed in the simulation that the actual order can deviate 20% from the forecast per period. The results of the scenario 1 are presented in Table 1.

86

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

7.3.4 Scenario 2 - Improvements Along the Supply Chain Scenario 2 focuses on shortening the supply chain by closing the inventory at the sales office and by direct customer delivery from the manufacturer. Now, the manufacturer’s order policy envisages always having sufficient articles in stock for the next two weeks (based on average demand per week). Therefore, in order to enable this strategy, a forecast is necessary and the 4-weeks rolling forecast from scenario 1 was conserved. By doing so, the manufacturer becomes aware of the actual customer requirement leading to an upstream move of the CODP. It is worth to mention that the sales office stays responsible for customer relation, contract extension and contract monitoring. In addition, this strategy results in new transport costs. New transport prices were estimated from interviews with the carrier and were factored into the simulation. Fig. 3 shows the entire process.

& 5

&5 0

0

Fig. 7.3 Improved process of scenario 2

7.3.5 Simulation Results The performance measures are set out in Table 21.2 and are related to an entire year. Based on the described improvements in scenario 1, we are able to reduce the number of stock outs, which is a direct indicator of customer satisfaction. Lead times and costs can not be reduced. Owing to the delivery time (mean and the variance) between sales office and manufacturer, it is necessary to keep large stocks, which in turn has a negative effect on stock keeping costs and the profit margin. This case

7 Performance Evaluation of Process Strategies Focussing on Lead Time Reduction

87

would also mean building work to extend the sales office inventory to handle the high stock of inventory. As it is not possible to find an improved solution according to all of our performance dimensions (lead time, customer satisfaction and cost), we decided to consider the entire supply chain in scenario 2. By cutting the flow of products through shortening the supply chain and postponing the CODP, it was possible to achieve a marked reduction in total lead time in scenario 2. Compared to the initial scenario, these activities also had a positive effect on customer satisfaction because it is possible to react much faster to customer requirements. Also stock keeping costs are reduced as sales office stores are no longer required and production is carried out on forecasts provided by the customer. In order to be complete, it has to be mentioned that this strategy would only be possible by extending the manufacturer inventory capacity. Nevertheless, we assume that this would be a realistic investment as the costs of building an extension to the manufacturer’s stores would easily be compensate by the saving on the transport costs side within one year.

7.4 Conclusion In this paper we analysed and assessed two different possibilities for supply chain improvements. We regarded their effects on lead time and it was possible to show financial and strategic enhancements. Our approach was illustrated by a polymer supply chain with a major OEM as end customer. For each of the alternatives, the performance was measured using lead time, finished articles inventory stocks as well as costs, number of stock outs and transport costs; where the number of stock outs constitutes a decisive index for customer satisfaction. The threshold number of stock outs should be less than 10 days per year. We were able to confirm the positive impact of lead time reduction on supply chain performance, i. e. the simultaneous reduction of inventory and increase of customer satisfaction. We managed to identify this specific dynamic behaviour by quantifying the benefits earned through each alternative. Further on, we confirmed the importance of considering the supply chain as a whole when assessing improvement alternatives. Our results demonstrate that the benefits of certain alternatives can only be realised if improvements are aligned along the supply chain partners, e. g. inventory management is based on the customer forecast and linked to the production planning. We believe these results to be interesting for both academics and practitioner as they contribute to better understanding the dynamics of the supply chain and the importance of the entire supply chain-specific evaluation of improvements. One of our next research activities will be to implement the most suitable alternative, in order to be able to draw further conclusions about the model (see also Mitroff et al (1974) and to ascertain an appropriate forecast algorithm based on historical data to support the customer forecast.

88

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

Table 7.1 Results of initial scenario and simulation runs Initial scenario Scenario 1 1 Total lead time (order entry at 43,14 manufacturer up to delivery of goods at the sales office in days)

Scenario 2

54,31 min 7,67 max 112,67 σ 16,60

14,62 min 2,68 max 38,67 σ 4,02

20,28 min 0.0 max 48,80 σ 7,47

12,93 min 0.0 max 38,67 σ 7,14

3 Delivery time between sales 35,44 office and manufacturer in days

35,11 min 1 max 79 σ 14,93

Omitted min 1 max 87,44 σ 14,35

4 Production lead time in days

No detailed consideration. Included in row 3

No detailed consideration. Included in row 3

1,73 min 0.82 max 11.1 σ 1.4

5 Stock outs in days

57

0,76 min 0.0 max 6 σ 1.5

1,75 min 0.0 max 6,04 σ 2.55

6 Transportation costs manufacturer → sales office in Euros

150000

158964 min 150800 max 163800

Omitted

7 Transportation costs sales office → customer 1-4 in Euros

33000

31895 min 29300 max 34430

Omitted

8 Transportation costs manufacturer → customer 1-4 in Euros

Omitted

Omitted

119600 min 113100 max 126100

9 Inventory costs sales office in Euros

7091

18753 min 9964 max 21779

Omitted

10 Inventory costs manufacturer in Euros

1774

1818 min 1778 max 1883

8904 min 7686 max 9196

2 Period of storage at the sales office (scenario 1) respectively the manufacturer (scenario 2) in days

7,8

Acknowledgements Partial funding for this research has been provided by the project “Matching supply and demand – an integrated dynamic analysis of supply chain flexibility enablers” supported by the Swiss National Science Foundation.

7 Performance Evaluation of Process Strategies Focussing on Lead Time Reduction

89

References Bertrand J, Fransoo J (2002) Operations management research methodologies using quantitative modeling. International Journal of Operations and Production Management 22(2):241–264 Bertrand J, Fransoo J (2009) Researching Operations Management, 1st edn, Routledge, New York, chap Modelling and Simulation Cachon GP, Fisher ML (2000) Supply Chain Inventory Management and the Value of Shared Information. Management Science 46(8):1032–1048 Chandra C, Kumar S (2000) Supply chain management in theory and practice: A passing fad or a fundamental change? Industrial Management and Data Systems 100(3):100–13 Christopher M, Towill D (2000) Supply chain migration from lean and functional to agile and customised. Supply Chain Management: An International Journal 5(4):206–13 Croom S (2009) Researching Operations Management. 1st edn, Routledge, New York, chap Introduction to Research Methodology in Operations Management Davis J, Eisenhardt K, Bingham C (2007) Developing theory through simulation methods. The Academy of Management Review (AMR) 32(2):480–499 Fisher M (1997) What is the right supply chain for your product? A simple framework can help you figure out the answer. Harvard Bus Rev 75(2):105–116 Gunasekaran A, Patel C, Tirtiroglu E (2001) Performance measures and metrics in a supply chain environment. International Journal of Operations and Production Management 21(1/2):71–87 Hopp WJ, Spearman ML (1996) Factory Physics: Foundations of Manufacturing Management. Irvin Inc., Chicago Jammernegg W, Reiner G (2007) Performance improvement of supply chain processes by coordinated inventory and capacity management. International Journal of Production Economics 108(1-2):183–190 Kelton W, Sadowski R, Sturrock D (2003) Simulation with ARENA, 3rd edn. McGraw-Hill Science/Engineering/Math Lee H (2002) Aligning supply chain strategies with product uncertainties. California Management Review 44(3):105–119 Lee HL, Padmanabhan V, Whang S (2004) Comments on Information Distortion in a Supply Chain: The Bullwhip Effect. Management Science 50(12):1887–1893 Mason-Jones R, Towill D (1999) Using the information decoupling point to improve supply chain performance. International Journal of Logistics Management 10(2):13–26 Mason-Jones R, Naylor B, Towill D (2000) Lean, agile or leagile? Matching your supply chain to the marketplace. International Journal of Production Research 38(17):4061–4070 Mentzer J, DeWitt W, Keebler J, Min S, Nix N, Smith C, Zacharia Z (2001) Defining supply chain management. Journal of Business logistics 22(2):1–26

90

Dominik Gl¨aßer, Yvan Nieto and Gerald Reiner

Mitroff I, Betz F, Pondy L, Sagasti F (1974) On managing science in the systems age: Two schemas for the study of science as a whole systems phenomenon. Interfaces 4(3):46–58 Naylor J, Naim M, Berry D (1999) Leagility: Integrating the lean and agile manufacturing paradigms in the total supply chain. International Journal of Production Economics 62(1-2):107–118 Olhager J (2003) Strategic positioning of the order penetration point. International Journal of Production Economics 85(3):319–329 Reiner G, Trcka M (2004) Customized supply chain design: Problems and alternatives for a production company in the food industry. A simulation based analysis. International Journal of Production Economics 89(2):217–229 Sanchez S, Sanchez P, Ramberg J, Moeeni F (1996) Effective engineering design through simulation. International transactions in Operational research 3(2):169– 185 Sch¨onsleben P (2004) Integral logistics management: Planning & control of comprehensive supply chains. CRC Press Silver E, Pyke D, Peterson R (1998) Inventory management and production planning and scheduling, 3rd edn. Wiley New York Vollmann T, Berry W, Whybark D, Jacobs F (2004) Manufacturing planning and control systems for supply chain management, 5th edn. McGraw-Hill Winklhofer H, Diamantopoulos A, Witt S (1996) Forecasting practice: A review of the empirical literature and an agenda for future research. International Journal of Forecasting 12(2):193–221

Chapter 8

A Framework for Economic and Environmental Sustainability and Resilience of Supply Chains Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

Abstract Traditionally supply chain management decisions are based on the economic performance which is expressed by financial and non-financial measures, i.e. costs and customer service. From this perspective, in the last decades, several logistics trends, i.e. outsourcing, offshoring and centralization, emerged. Recently, studies have shown that the focus on the cost aspect is no longer sufficient. Due to internal and external drivers (e.g. customer pressure, regulations, etc.) environmental criteria become more and more important for the decision-making of individual enterprises. Furthermore, the risk which is related to the increased transportation distances resulting from these strategies is often not taken into account or underestimated. These shifts in priorities of companies force them to search for new logistics strategies that are at the same time cost-efficient, environmentally friendly and reliable. Based on this integrated perspective new logistics trends, like on- and nearshoring, flexible supply base or flexible transportation, have come up recently and will gain more importance in the near future. Relying on a flexible supply base a company can benefit from low costs in an offshore facility and simultaneously be able to respond quickly to demand fluctuations and react to delivery delays and disruptions by serving the market also from an onshore site. A single-period dual sourcing model is presented to show the effects of emission costs on the offshore, onshore and total order quantity.

Heidrun Rosiˇc Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria, e-mail: [email protected] Gerhard Bauer Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria, e-mail: [email protected] Werner Jammernegg Vienna University of Economics and Business, Nordbergstraße 15, 1090 Vienna, Austria, e-mail: [email protected]

91

92

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

8.1 Introduction Traditionally supply chain management decisions are based on the economic performance which is expressed by financial and non-financial measures, i.e. costs and customer service. From this perspective, in the last decades, different logistics trends, i.e. outsourcing, offshoring and centralization, have emerged. Even though these trends seem to be rather “old” they are still prevailing in today’s businesses. Recently, a study conducted in Austria has shown that 41% of the interviewed companies still intend to offshore some of their production activities in the following two years. Furthermore, 35.4% of them plan to move their production sites to Asia; especially China is a prominent destination for offshoring. The low cost of the production factors (personal, material, etc.) are the key drivers for their decisions (Breinbauer et al, 2008). A European-wide study carried out by Fraunhofer ISI concerning offshoring showed similar results. Between 25% and 50% of the surveyed enterprises moved parts of their production abroad in the years 2002 and 2003 (Dachs et al, 2006). Further examples can be found. For instance, the Austria-based Knill Group, which is active in the field of infrastructure, supplying systems and applications for energy and data transmission, built new production facilities in India and China within the past 36 months in order to take advantage of lower wages in Asia (Breinbauer et al, 2008). NXP, a leading semiconductor company is headquartered in Europe and employs more than 33,500 employees. The company pursued a strong offshoring strategy and now more than 60% of its production activities are located in Asia, 5% in America; only 33% have remained in Europe. Also, AT&S, a large Austrian manufacturer of printed circuit boards, continues its offshoring strategy. In January 1999, AT&S started operating in India by acquiring the largest Indian printed circuit board manufacturer and now it will build a second facility located nearby. The investments for this project will amount to 37 million Euros and production activities shall start in the third quarter of 2009. Besides, AT&S operates facilities in China and Korea. In section 2 prevalent logistics trends are presented focusing on a cost perspective, thereby showing the trade-offs that exist between the different cost components. The trends presented, i.e. outsourcing, offshoring and centralization, usually lead to lower production (procurement) costs in the case of offshoring and outsourcing or lower inventory costs in the case of physical centralization. But, in general, they result in an increase of transportation distances, therefore making supply chains longer and/or more complex. Often in the evaluation of these strategies side effects of increased transportation distances are not taken into account adequately. Therefore, in section 3 in addition to the economic criteria, “soft” factors, like lead time, delivery reliability, flexibility, etc. and the environmental impact are included. Based on this integrated perspective consisting of costs, risks and environment new logistics trends are highlighted. One of these new logistics trends is then analyzed in more detail, namely flexible supply base with the specific variant dual sourcing. In section 4 a transport-focused framework for dual sourcing (off- and onshore supply source)

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

93

and in section 5 a single-period model for dual sourcing including emission costs are presented.

8.2 Prevalent Logistics Trends: Cost Perspective The subject of supply chain management is the organization of the transformation of raw materials into final products by a well defined network of manufacturing, warehousing and transportation processes. Most of the necessary activities are determined by the design of the supply chain. In network design, for instance, it is decided where manufacturing activities take place, which location performs a certain activity, where procurement and/or distribution centers are located and how the transportation is handled between the different stages. Traditionally, these decisions are based on the economic performance which is expressed by financial and non-financial measures, i.e. costs and customer service. Often these measures are conflicting, like optimizing the total landed cost of all involved transformation processes and satisfying customer requirements. A first trade-off is between logistics cost and customer service: high levels of product availability can be achieved with high inventory and thus high inventory cost; short delivery times are possible with additional resources for manufacturing and/or transportation related with an increase of the respective cost. Moreover, a second trade-off must consider the costs of resources (manufacturing, transportation, storage facilities). In order to stay competitive in the market, an enterprise chooses the strategy which is most efficient, generates lowest total landed cost (facilities, inventory and transportation) and satisfies customer requirements. Different trends, i.e. outsourcing and offshoring of production activities and physical centralization, have emerged due to a focus on cost reduction. Outsourcing of production activities means to subcontract a process to a third-party in order to concentrate investments and time on the core competencies of a company. Outsourced processes may be done more efficiently and cheaper by a third party which gains economies of scale. Further, the fixed cost of production can be reduced (Chopra and Meindl, 2006). Offshoring is defined as locating activities abroad with varying degree of the geographical distance between the original and the new location depending on the author, e.g., “outside a country’s boundaries”, “outside the first world”, “outside of the continent”. In this paper, offshoring does not include each transfer of manufacturing facilities outside a country’s boundaries, but the term offshoring only applies to those relocations made to a far-distant country. With respect to the term “far-distant country” it has to be kept in mind that the actual geographical distance is relevant and not the legal boundaries of a country. The main driver for offshoring is to lower operational costs due to lower wages of the workforce abroad or lower raw material costs. Further reasons for offshoring are gaining market access, following a key customer or productivity increases. Typical offshore regions are situated in Asia because there a company can take advantage of significantly lower labor costs. The dislocation of production activities from West-

94

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

ern Europe to Eastern Europe is often called nearshoring as the distance and the cultural differences are less (Ferreira and Prokopets, 2009). Physical centralization means that the number of production, procurement and/or distribution sites is reduced to a single one, this means “consolidating operations in a single location” (Van Mieghem, 2008). The main goal of centralization is to pool risk, reduce inventory and exploit economies of scale (Chopra and Meindl, 2006). These trends mainly lead to a reduction of total landed cost due to lower production (procurement) cost in the case of offshoring and outsourcing or lower inventory cost due to risk pooling in the case of physical centralization. But as a negative sideeffect supply chains are longer and/or more complex (Tang, 2006). Due to the increased length of supply chains more transportation activities are necessary leading to an increase of the respective costs. In this paper we will especially pay attention to the effect of transportation activity within a supply chain.

8.3 New Logistics Trends: Integrated Perspective The presented logistics trends have proven to be optimal for industrial companies under economic considerations. Recently studies have shown that the focus on the cost aspect of a certain strategy is no longer sufficient. Environmental criteria become more and more important for the decision-making of individual enterprises. Walker et al (2008) differ between internal (organizational factors, efficiency improvements) and external drivers (regulation, customers, competition and society) which may induce the consideration of environmental aspects in supply chain decision-making. Especially carbon dioxide (CO2) emissions heavily accelerate the greenhouse effect; 60% of this effect is caused by CO2. This is a reason why governmental institutions (UN, EU, etc.) often focus their regulations on CO2-reduction (Kyoto protocol, EU emission trading scheme, etc.). Furthermore, the risk which is related to these strategies is often not taken into account or underestimated. There are various types of risks that exist especially in the case of offshoring. Currency risk and political risk depend on the economic and political stability within a country. Intellectual property risk and competitive risk should also not be ignored (Van Mieghem, 2008). Ferreira and Prokopets (2009) conclude from the “2008 Archstone/SCRM Survey of Manufacturers” (in-depth survey of 39 senior executives from US and European-based manufacturers) that executives also start to recognize aspects of offshoring, such as “quality problems, longer supply chains, lack of visibility, piracy and intellectual capital theft”. Due to these additional aspects the cost savings of offshoring which represent between 25% and 40% on average start to diminish. In addition, an offshoring strategy negatively affects the flexibility and responsiveness of a supply chain as shipments have to be made in large lots (e.g. containersize) and the delivery time is very long (e.g. up to several months). Besides, the customization of products to individual customer needs is more difficult. Furthermore, the cost components are about to change; 40% of the manufacturing enterprises have

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

95

experienced an increase of 25% or more in direct costs of offshoring (materials, components, logistics and transportation) over the last three years. Nearly 90% of them expect costs to rise by more than 10% in the next 12 months. This is due to increasing labor costs in important offshore countries, like China (2005-2008: wages + 44%), an increase in transportation charges for sea freight (2005-2008: freight charges + 135%) and a non-favorable development of foreign currencies (Ferreira and Prokopets, 2009). Furthermore, Simchi-Levi et al (2008) point out that even though the oil price has decreased recently it is likely that it will increase again above $100 a barrel in the year 2009. Offshoring, outsourcing and centralization result in supply chains which are costefficient, but as a negative side-effect they are longer and/or more complex (Tang, 2006). Due to the increased length of supply chains more transportation activities are necessary; even though some of the transport can be shifted to more environmental friendly modes, such as sea transport, in total, these trends have a negative impact on the environment. Similar conclusions can be drawn with respect to the risk dimension. The more extended a supply chain is, the more risk it has to bear and the more difficult it is to guarantee a certain delivery reliability. It can be concluded that in the future the existing trends have to be reconsidered; environmental criteria and the risk dimension will become more important. Further, the cost structure is expected to change. These shifts in priorities of companies as well as the shifts in the cost components force companies to search for new logistics strategies that are at the same time cost-efficient, environmental friendly and reliable. Supply chain risks as well as environmental aspects should be considered, besides economic criteria, in the performance evaluation of a supply chain. Based on this integrated perspective new logistics trends have come up recently and will gain more importance in the near future. Through network redesign, i.e. by moving production activities back or closer to the market through near-, onshoring or decentralization, the transportation distances can be reduced. The study of Ferreira and Prokopets (2009) shows that 30% of the companies surveyed have already reversed their offshoring decision; 59% are willing to change their strategy with respect to offshoring. This means that either offshored activities are relocated or that managers will show an increased awareness in future offshoring decisions. For instance, a company from the apparel industry which produces casual wear, sportswear and underwear in two manufacturing sites in the US considered redesigning its network in order to reduce its CO2-emissions. For inbound transportation rail and trucks were used whereas on the outbound side the company completely relied on road transportation. The moving of some production activity to a low-cost site in Mexico and the introducing of new distribution sites was evaluated considering cost and CO2-emissions. The optimization only with respect to cost led to a moving of production activity to Mexico and the installation of two additional distribution centers; total costs reduction (costs for warehouse and production sites, transportation and inventory) amounted to 16%, nearly US$ 8 million in absolute figures, and CO2-emissions could be lowered by 5%. Then, a reduction of CO2-emissions by 25% was introduced as constraint. Now, nearly no production activity was dislocated to Mexico, thereby producing closer to the

96

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

market and reducing transportation distances. This new network design resulted in a small increase of total costs compared to the optimal solution, but the total costs are still more than 10% smaller than in the initial situation and the CO2-emissions could be reduced by a quarter (Simchi-Levi, 2008). Concerning supply chain risks, it has to be pointed out that offshoring, outsourcing and centralization typically move production away from the market which reduces the responsiveness and flexibility of a supply chain. This has to be considered together with the possible cost reductions of a certain strategy (Allon and Van Mieghem, 2009). Further, Tang (2006) points out that supply chains have to become robust which means that a supply chain is able to fulfill customer requirements even though a disruption of the supply chain has occurred. This disruption can be of different kind, either a short one due to congestion or accidents or a long one which can be the result of a natural disaster or a terrorist attack destroying one node or arc in the supply chain. By using a flexible supply base a company can benefit from low costs in an offshore facility and simultaneously be able to respond quickly to demand fluctuations by serving the market also from an onshore site and react to delivery delays and disruptions. In this way, the amount of long-distant transport can be reduced, therefore mitigating transportation risks. For instance, Hewlett Packard uses an offshore facility to produce the base volume and employs also an onshore facility to quickly react to disruptions and demand fluctuations (Tang, 2006). Furthermore, flexible transportation helps to improve the performance of a supply chain by a change of transport mode, multi-modal transportation or the use of multiple routes. The use of a single mode is mainly due to cost consideration and the aim to reduce complexity in supply chains but this increases the vulnerability of the supply chains. By using multi-modal transportation the supply chain is able to obtain more flexibility and therefore can handle disruptions easier. Especially in the case of congestion an alternative route could increase the time- as well as costeffectiveness. For instance, LKW Walter decided to change the mode on the link north-eastern Spain to southern Italy. Road transportation was replaced by a multimodal solution (sea/truck). Thereby, 1,211 km per shipment (1,523 km on the road vs. 312 km short sea/trucking), in total over 1.2 million km per year, could be saved (ECR, 2008). Nike operates a distribution center in Belgium that serves the European market. 96% of the freight to the location is transported by inland waterways. Thereby, 10,000 truck loads could be saved and also on the distribution side Nike very much relies on waterways; only the direct delivery to the customers is carried out by truck (Seebauer, 2008). Improvements in transportation efficiency can be achieved through better vehicle utilization, the reduction of empty trips as well as less frequent shipments with larger lot sizes. This leads to a reduction of the number of transports. Thus costs, CO2-emmissions and fossil fuel consumption can be reduced significantly. S.C. Johnson & Son Inc., a household and personal-care products maker, for instance, was able to cut fuel use by 630,000 liters by improving truckload utilization (Simchi-Levi et al, 2008). By maximizing full truck load and supplying the market from the closest location, PepsiCo, on average, saved 1.5 million km and 1,200 t

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

97

CO2-emissions (ECR, 2008). The British drugstore chain Boots, for instance, could avoid empty runs by using route planning. Thereby, 2.2 million kilometers on the road could be eliminated which resulted in a reduction of 1,750 t CO2-emissions. In combination with the use of larger containers, increased utilization of the containers and reduced amount of air transportation Boots achieved a reduction of 3,000 t CO2 (-29%) between 2004 and 2007. These improvements were only possible due to the tight collaboration between Boots and its logistics service provider Maersk Logistics (Seebauer, 2008). According to Simchi-Levi et al (2008) logistics service providers will be employed more often in order to increase efficiency. They are able to consolidate the shipments from a large number of customers and therewith can reduce the number of empty trips. Again Boots was able to save approximately 120,000 km as well as 92 t of CO2-emissions per year by sharing transportation with another company in the UK. Further examples in this context can be found in the ECR Sustainable Transport Project (ECR, 2008). Table 8.1 gives an overview of the presented new logistics trends. In the following sections we use the flexible supply base - one of the presented new logistics trends - to develop a transport-focused framework and a stylized model for dual sourcing.

8.4 Transport-Focused Framework for Dual Sourcing In the previous section it was exemplarily shown that a flexible supply base can help to improve the performance of a supply chain from an integrated perspective including economic, risk and environmental criteria. In the following we will focus on a certain type of this strategy, i.e. dual sourcing depending on a cheap but inflexible and slow offshore supply source and on an expensive but flexible and fast onshore supply source. The onshore supply source can help to improve the performance of a supply chain with respect to risks in two cases, to bridge delivery delays and/or disruptions or to fulfill demand exceeding the offshore order quantity. Table 8.2 gives an overview of the external conditions that have an impact on a company’s policy and the decisions to be taken. Environmental regulations, like the emission trading scheme of the EU, impose restrictions on companies and therefore influence the policies they choose. The emission trading scheme of the EU (EU ETS) was implemented in order to reach the goals stated in the Kyoto protocol. It is a cap-and-trade system of allowances for emitting CO2 and other greenhouse gases whereby each allowance certifies the right to emit one ton of CO2. Only certain industries are included in this regulation up-tonow. These industries are heavy energy consuming industries, like refineries, power generation with fossil resources, metal production and processing, pulp and paper, etc. Today, 11,000 sites that produce around 50% of the EU’s total CO2-emissions are covered by the EU ETS. A certain number of emission allowances are allocated to the companies free of charge. Those companies that produce fewer emissions than the number of allowances owned can sell them, whereas those producing more have

98

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

Table 8.1 Overview of new logistics trends: Integrated perspective Logistics trends

Characteristics

Relevance on in- Case study tegrated perspective

Network redesign

Nearshoring, Reduced Onshoring, transportation Decentralization distances and number of transports

Using regional distribution centers, a company from the metal manufacturing industry was able to reduce the average distance to customer by 46%.In the apparel industry, the decision to produce at an onshore facility reduced CO2emissions by 25%.

Flexible supply base

Using multiple supply sources (offshore and onshore)

Reduced number of long-distant transports and mitigation of transportation risks

Hewlett Packard uses an offshore facility to produce the base volume and employs also an onshore facility to quickly react to disruptions and demand fluctuations.

Flexible transportation

Change of transport mode, Multi-modal transportation, Multiple routes

Reduced CO2-emissions and dependence on fossil fuels, Reaction to occurrence of risk events

LKW Walter saved 1,211 km per shipment by changing the mode (1,523 km on the road vs. 312 km short sea/trucking), in total over 1.2 million km per year.

Transportation efficiency

Vehicle routing and loading, Consolidated shipments

Reduced number of empty trips, Improved, vehicle utilization

By maximizing full truck load, PepsiCo, on average, saved 1.5 million km and 1,200 t CO2-emissions. A manufacturer of household and personal-care products cut fuel use by 630,000 litres by combining multiple customer orders.

Table 8.2 Transport-focused framework for dual sourcing External conditions

Environmental regulations (emission trading scheme) Transportation network including transportation risks

Policies

Dual sourcing (off- and onshore supply source)

Decisions

Offshore order quantity Emission allowances

to buy additional allowances, get credits by engaging in emission-saving projects or have to pay a penalty. The aim is to reduce the number of allowances constantly, so as to decrease the total CO2-emissions within the EU (-21% until 2020). In 2006, half of the greenhouse gases in the EU are caused by industry; the second largest “polluter” is transportation accounting for nearly 20%. The EU is already planning to increase the number of companies and sectors which have to comply with the trad-

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

99

ing scheme, e.g. include civil aviation by 2013 (EC, 2008). So, it has to be expected that the whole transport sector will be confronted with more severe regulations or the inclusion into the EU ETS in the near future. External conditions are also determined by the transportation network including respective risks. According to Rodrigues et al (2008) transportation risks are related to the carrier who executes the transport and to external factors. The carrier is a source of risk with respect to his fleet capacity, network planning, scheduling and routing and information system as well as his financial conditions and reliability. As external risk factors, transport macroeconomics (oil price, availability of drivers, etc.), infrastructure conditions (congestion, construction, etc.) and future government policies have to be mentioned. Further, severe shocks, like terrorist attacks, natural disasters or industrial action, might have a strong impact on the transportation network. Whereas the probability of such event is very low, the impacts can be detrimental. Based on this, they state that with the increasing degree of outsourcing and the higher geographical spread of supply chains the transportation risks increase (Rodrigues et al, 2008). The paper by Allon and Van Mieghem (2009) about global dual sourcing shows that it is almost impossible to derive the optimal sourcing policy for a responsive near-shore source and a low-cost offshore source even if the criterion is just cost minimization. By including an environmental criterion, thus, it seems reasonable to develop a simple model for dual sourcing with onshore reactive capacity to be able to analyze the consequences for the offshore order quantity.

8.5 Single-Period Dual Sourcing Model Including Emission Costs In the seminal newsvendor model a possibility to reduce the mismatch cost of understocking or overstocking is to allow for a second order opportunity. In the simplest version it is assumed that at the beginning of the selling season the demand of a product is known exactly or that the second production facility can immediately produce any requested quantity (see e.g.,Warburton and Stratton, 2005 or Cachon and Terwiesch, 2009, chapter 12). In the considered single-period dual sourcing model a product can either be sourced from an offshore production facility and from an onshore production plant whereby the capacity of the onshore supply source is unlimited and can deliver immediately. The two suppliers can be internal or external to the company. Because of the long procurement lead time the offshore order quantity of the product is based on the random demand X characterized by the distribution function F. The company, e.g. a retailer, sells the product at the unit selling price p. The purchase price per unit from the offshore supplier is denoted by co f f , that from the onshore supplier is con . Leftover inventory at the end of the regular selling season can be sold at a unit salvage value z. It is assumed that p > con > co f f > z holds. Then the profit P depends on the offshore order quantity q and on the realized demand x:

100

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

P(q, x) =

px − co f f q + z(q − x)+ x ≤ q px − co f f q − con(x − q)+ x > q

The optimal offshore order quantity q∗ is derived by maximizing the expected profit E(P(q, X)). Using the framework of the classical newsvendor model the optimality condition is given by (see, e.g. Cachon and Terwiesch, 2009, section 12.4): F(q∗ ) =

con − co f f con − z

The unit purchase price from the offshore supplier is composed of the product price per unit c and the emission cost factor ϕ ; the unit purchase price from the onshore supplier is obtained by adding a domestic premium (d · c) to the offshore product price per unit. This premium mainly is caused by higher labor costs that have to be paid in the onshore production facility (Warburton and Stratton, 2005). The two cost parameters are defined as : co f f = (1 + ϕ )c, con = (1 + d)c. The offshore supply source is only used if it is overall cheaper than the onshore supply source, which is the case as long as ϕ < d. As soon as d ≥ ϕ the product quantity is exclusively procured from the onshore source on order. The factor ϕ represents the emission costs per product unit, whereby it is assumed that costs for emission allowances only arise for long-distant transportation from the offshore location. The emission costs per unit sourced from the offshore supplier depend on the selected transportation route and transportation mode. For the different modes average emission factors per kilometer exist. Multiplying these emission factors with the distance the vehicle has to travel, the CO2-emissions for one trip can be calculated. The emission costs, then, are derived from the buying price of an emission allowance, traded under the EU ETS. It is reasonable to assume that the emission cost factor ϕ is independent of the order quantity q if the transport is carried out by a logistics service provider. The company, e.g. retailer, then has to reserve a fixed transport capacity which determines the factor ϕ . If part of that reserved capacity is not used by the company, the logistics service provider can sell it to other customers and therefore usually achieve high vehicle utilization. A numerical example with the following cost and price parameters is presented in order to show the impact of emission costs on the quantity decisions: selling price p = 20, product price per unit c = 10, salvage value z = 5 and domestic premium d = 0.2. The emission costs factor ϕ is varied in order to show the impact of increasing environmental costs on the optimal decision. Demand is assumed to be normally distributed with a mean μ of 1,000 units whereby two different standard deviations (σ1 = 150, σ2 = 300) are used in order to show the impact of variability. Taking a normally distributed demand is justified if the coefficient of variation (σ /μ ) is small enough (Warburton and Stratton, 2005). The offshore order quantity depends on the relative cost advantage that can be achieved through offshore sourcing. The lower the offshore cost is the more the

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

101

retailer will procure from the offshore source. The onshore supply source is only employed in order to fulfill the demand that exceeds the offshore order quantity, i.e. expected lost sales. Therefore, with the onshore supply source a service level of 100% can be guaranteed. But it should not be forgotten that this comes at a high domestic premium. Nevertheless, the dual sourcing strategy often outperforms a pure offshoring strategy with respect to expected profit (see, e.g. Cachon and Terwiesch, 2009). With increasing emission costs (ϕ · c) the company sources less from offshore as the cost advantage is reduced. The offshore quantity decreases nearly linearly with increasing ϕ until a certain point after which it decreases sharply. The total order quantity (off- and onshore quantity) also decreases depending on ϕ . This is due to the following fact: The fewer units are procured through the offshore supply source the lower is the expected leftover inventory (I). The whole expected lost sales quantity (qon) is then fulfilled from the onshore supply source and this decision is taken under complete certainty. Overall, the total order quantity converges to the mean demand because q∗ + qon = E(X) + I Higher demand uncertainty, i.e. a higher coefficient of variation of demand, implies that the onshore supply source is used more. The numerical results for the two different demand distributions with the above price and cost parameters are graphically shown in Fig.8.1 and Fig.8.2. The emission costs factor is varied in the range 0 ≤ d < ϕ .

Fig. 8.1 Off-, onshore and total order quantity depending on the emission cost factor ϕ for normally distributed demand with μ = 1, 000, σ1 = 150 and d = 0.2

102

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

Fig. 8.2 Off-, onshore and total order quantity depending on the emission cost factor ϕ for normally distributed demand with μ = 1, 000, σ1 = 3000 and d = 0.2

The presented model is based on limiting assumptions with respect to the existing environmental regulations concerning emission allowances. Under the existing EU ETS companies receive allowances free of charge. Therefore, in contrast to the model presented, emission costs do not arise for each unit ordered, but only if a certain threshold is exceeded. For a more general model with a positive emission limit and the opportunity to buy additional emission allowances or to sell unused ones we refer to Rosiˇc and Jammernegg (2009).

8.6 Summary Prevalent logistics trends, i.e. outsourcing, offshoring, and centralization are presented from a cost perspective. These strategies are chosen with the objective to reduce total landed costs (e.g. reduction of labor costs through offshoring or inventory costs through centralization). But as direct consequence transportation distances increase; supply chains become longer and/or more complex. This has negative impacts on the risk a supply chain has to face (e.g. congestion on transportation links) and on the environment (e.g. CO2-emissions). An integrated perspective is presented and new logistics trends which perform better with respect to transportation risks and the environment are illustrated by several case studies. Further, we use one of the presented trends - flexible supply base - to develop a transport-focused framework for dual sourcing. Dual sourcing means that a company relies on a cheap but slow offshore supply source and on an expensive but fast and unlimited onshore supply source. The external conditions which influence the policies of an individual

8 A Framework for Economic and Environmental Sustainability and Resilience of SC

103

company are environmental regulations focusing on the emission trading scheme for CO2-allowances of the EU and the transportation network including the respective risks. Then, a single-period dual sourcing model is presented. The objective is to maximize expected profit and it has to be decided how much to order from the offshore source; the onshore source is used in order to fulfill the demand exceeding the offshore order quantity. The costs for emission allowances are included in the offshore purchase price as for the quantity procured from the offshore supply source more transportation activity is necessary. It is shown that with increasing emission costs the offshore order quantity decreases, whereas the onshore order quantity increases and the total order quantity converges to the expected demand. The presented model is based on limiting assumptions with respect to the existing environmental regulations for emission allowances. For a more general model with a positive emission limit and the opportunity to buy additional emission allowances or to sell unused ones we refer to Rosiˇc and Jammernegg (2009).

References Allon G, Van Mieghem J (2009) Global Dual Sourcing: Tailored Base Surge Allocation to Near and Offshore Production. Tech. rep., Working Paper, Kellogg School of Management, Northwestern University Breinbauer A, Haslehner F, Wala T (2008) Internationale Produktionsver¨ lagerung Osterreichischer Industrieunternehmer Ergebnisse einer empirischen Untersuchung. Tech. rep., FH des bfi Wien, URL http://www.fhvie.ac.at/files/2008 Studie Produktionsverlagerungen.pdf Cachon G, Terwiesch C (2009) Matching supply with demand: An introduction to operations management, 2nd edn. McGraw-Hill, Boston Chopra S, Meindl P (2006) Supply chain management, 3rd edn. Pearson Prentice Hall, New Jersey Dachs B, Ebersberger B, Kinkel S, Waser B (2006) Offshoring of production A European perspective. URL http://www.systemsresearch.ac.at/%20getdownload.php?id=154 EC (2008) EU action against climate change The EU Emissions Trading System. European Commission. URL http://ec.europa.eu/environment/climat/pdf/brochures/ets en.pdf ECR (2008) ECR Sustainable Transport Project Case Studies. URL http://www.ecrnet.org/05-projects/transport/Combined%20Case%20studies v1%208 220508 pro.pdf Ferreira J, Prokopets L (2009) Does offshoring still make sense? Supply Chain Management Review 13(1):20–27 Rodrigues V, Stantchev D, Potter A, Naim M, Whiteing A (2008) Establishing a transport operation focused uncertainty model for the supply chain. International Journal of Physical Distribution & Logistics Management 38(5):388–411

104

Heidrun Rosiˇc, Gerhard Bauer and Werner Jammernegg

Rosiˇc H, Jammernegg W (2009) The environmental sustainability of quick response concepts. Working paper, Department of Information Systems and Operations, Vienna University of Economics and Business ¨ Seebauer P (2008) Supply Chain unter der Oko-Lupe. Logistik heute 2008 (10):54– 55 Simchi-Levi D (2008) Green and supply chain strategies in a volatile world. Fachkoferenz: Gr¨une Supply Chains, Frankfurt/Main, Germany Simchi-Levi D, Nelson D, Mulani N, Wright J (2008) Crude calculations. URL http://online.wsj.com/article/SB122160061166044841.html Tang C (2006) Robust strategies for mitigating supply chain disruptions. International Journal of Logistics: Research and Applications 9(1):33–45 Van Mieghem J (2008) Operations Strategy: Principles and Practice. Dynamic Ideas, Charlestown Walker H, Di Sisto L, McBain D (2008) Drivers and barriers to environmental supply chain management practices: Lessons from the public and private sectors. Journal of Purchasing and Supply Management 14(1):69–85 Warburton R, Stratton R (2005) The optimal quantity of quick response manufacturing for an onshore and offshore sourcing model. International Journal of Logistics, Research and Applications 8(2):125–141

Chapter 9

An Integrative Approach To Inventory Control Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

Abstract Inventory control systems consist of three types of methods: forecasting, safety stock sizing and order timing and sizing. These are all part of the interpretation of a planning environment to generate replenishment orders, and may consequently affect the performance of a system. It is therefore essential to integrate these aspects into a complete inventory control process, to be able to evaluate different methods for certain environments as well as for predicting the overall performance of a system. In this research a framework of an integrated inventory control process has been developed, covering all relations from planning environment to performance measures. Based on this framework a simulation model has been constructed; the objective is to show how integrated inventory control systems perform in comparison to theoretical predictions as well as to show the benefits of using an integrated inventory control process when evaluating the appropriateness of inventory control solutions. Results indicate that only simple applications (for instance without forecasts or seasonality) correspond to theoretical cost and service level calculations, while more complex models (forecasts and changing demand patterns) show the need for tight synchronization between forecasts and reordering methods. As the framework describes all relations that affect performance, it simplifies the construction of simulation models and makes them accurate. Another benefit of the framework is that it may be used to transfer simulation models to real-world applications, or vice versa, without loss of functionality.

Philip Hedenstierna Logistics Research Group, University of Sk¨ovde, 541 28 Sk¨ovde, Sweden Per Hilletofth, Corresponding author Logistic Research Group, University of Sk¨ovde, 541 28 Sk¨ovde, Sweden, Tel.: +46 (0)500 44 85 88; Fax: +46 (0)500 44 87 99, e-mail: [email protected] Olli-Pekka Hilmola Lappeenranta Univ. of Tech., Kouvola Unit, Prikaatintie 9, 45100 Kouvola, Finland

105

106

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

9.1 Introduction The purpose of inventory control is to ensure service to processes or customers in a cost-efficient manner, which means that the cost of materials acquisition is balanced with the cost of holding inventory (Axs¨ater, 1991). This is done by interpreting data describing the planning environment, i.e. the parameters that may affect the decision, to generate replenishment order times and quantities (Mattsson, 2004). The performance of an inventory control system may then be measured by the service and the total cost caused, when applied in a certain environment. Inventory control methods may be classified by whether they determine ordering timing, quantity, or both (Mattsson and Jonsson, 2003). For systems that determine only one aspect, such as the reorder point system or the periodic ordering system, the undetermined aspect must be calculated beforehand, typically using the economic order quantity or the corresponding economic inventory cycle time (Waters, 2003). The parameters that inventory control systems are typically including such things as demand forecasts, projected lead times, holding rates and ordering costs. Of these, the forecast is of special concern, as it is not a part of the planning environment, but a product thereof. This makes forecasting, given our definition of inventory control, an integral part of the inventory control system. To maintain service when there is forecast and lead time variability, safety stock is used, which is based on variability and on the uncertain times relating to the used inventory control model (Axs¨ater, 2006). As the safety stock incurs a holding cost, it may be argued that it should be part of the timing/sizing decision; however, it is usually excluded as optimization in most cases only gives insignificant cost savings (Axs¨ater, 2006). However, in larger distribution systems reduction of safety stocks is one primary driver of centralization of warehouses to one location (e.g. demand pooling, square root law, see Zinn et al 1989 and Das and Tyagi 1999) these could concern feeding warehouses of factories (Rantala and Hilmola, 2005, e.g.) as well as retail related distribution operations (Leknes and Carr, 2004, e.g.). We have now discussed three interdependent areas that are usually treated individually. They are all part of the interpretation of the planning environment to generate replenishment orders, and may consequently affect the performance of the system. The current approach to inventory control does generally not consider it as a single system, but as separate methods (e.g. Axs¨ater 2006, Waters 2003, Mattsson and Jonsson 2003 and Vollmann 2005). An exception is Higgins (1976), who describes inventory control as a process (see Fig.9.1), but does not detail how information flows through the model, nor does he isolate the functions of forecasting, safety stock sizing and inventory control. Looking at Higgins model, it is easy to realize that corruption of data occurring in an operation or between operations will cause the incorrect data to affect subsequent operations (Ashby, 1957). When theoretical models are applied to scenarios that follow the assumptions of the models, this is not an issue; but when a model is applied to a scenario it is not designed for, data corruption ensues. Applied to inventory control, this may mean that a simple forecast or a simple inventory control method is applied in an environment that does not reflect the method’s assumptions

9 An Integrative Approach To Inventory Control

107

(The popularity of using simple methods, like the reorder point method is shown in Jonsson and Mattsson 2006 and Ghobbar and Friend 2004). When a method’s assumptions are unmet, its performance may be difficult to predict. The scenario of using theoretically improper methods is not unlikely, as businesses may want to utilize inventory control methods that are simple to manage, such as the reorder point system, even when the planning environment would require a dynamic lotsizing method such as the Wagner-Whitin, Part-period or the Silver-Meal algorithm (Axs¨ater, 2006). In the same fashion, simple forecasting methods may be applied to complex demand patterns to simplify the implementation and management of the forecasts.

Fig. 9.1 Higgins inventory control process model (Higgins, 1976)

To understand how a method will respond to an environment it was not designed for, it is necessary to understand the entire process, from planning environment to measurement of results. As it may be difficult to predict how a system based on a required type of input will react to unsuitable data, a model of the system may help to give insight into the system’s performance. In his law of requisite variety, Ashby (1957) states that a regulatory system must have at least as much variety as its input to fully control the outcome; applied to the inventory control process, this means that all aspects of a system must be modeled to get an accurate result. Inventory control systems consist of three types of methods: forecasting, safety stock sizing as well as order timing and sizing (Axs¨ater, 2006). Though there are many individual methods, only one method of each type may be used in an inventory control system for a single stock-keeping unit. In this research a framework of an integrated inventory control process has been developed, covering all relations from planning environment to performance measures. The design of the framework was based on a literature review of inventory control theory and on the authors’ experience of the area. Based on this framework a simulation model has been constructed considering demand aggregation, forecasting, and safety stock calculations, as well as reordering methods. The research objective of this study is to provide an increased understanding of the following research questions: (i) ‘How may the process of inventory control be described in a framework that allows for any combination of methods to be used?’, and (ii) ‘Is there any benefit of integrated inventory control when deciding appropriate inventory control solutions?’.

108

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

The remainder of this paper is structured as follows: First, Section 9.2 integrates existing theory to describe a framework for designing inventory control models. Section 9.3 introduces empirical data from a company, whose planning environment was interpreted in Section 10.2.2 to develop a simulation model based on the framework. Section 9.5 describes the results of the simulations. Thereafter, Section 23.5 discusses the implications of the results, while Section 18.5 describes the conclusions that can be drawn from the study.

9.2 Framework for Integrated Inventory Control The design of the framework is based on observing how inventory control methods operate, what input they require and what output they provide. An underlying assumption for inventory control systems is that there for any given time t, is an inventory level LL, which is reduced by demand D and increased by replenishment R. Another assumption is that time is divided into buckets as described by Pidd (1988), for continuous systems buckets are infinitesimal, and that for each bucket the lowest inventory level, which is sufficient to evaluate the effects of inventory control, is governed by Formula 1. The relationship between these factors has been deduced from the rules that material requirements planning is built on (Vollmann, 2005). LLt = LLt−1 + Rt−1 − Dt

(9.1)

where LLt = lowest inventory level at time t,Rt−1 == replenishment quantity occurring before t and Dt = demand during t. Formula 1 dictates how transactions of any system placed in the framework will operate. It considers replenishment to occur between time buckets, meaning that it is sufficient to monitor the lowest inventory level to manage inventory transactions. Information such as service levels, inventory position and the highest stock level may be calculated from the lowest inventory level. The formula governs the inventory transactions of any inventory control system, and must be represented in any inventory control application. All other parts of an application may vary, either depending on the planning environment in which an inventory control system is used, or on the design of the system. Fig. 9.2 shows the framework, which starts with the planning environment and ends with a measurement of the system’s performance. The planning environment comprises the characteristics of all aspects that may affect the timing/sizing decisions (Mattsson, 2004). For each time unit, the environment, which determines the distribution of demand, generates momentary demand that is passed on to a forecasting method, to an inventory control method and to actual transactions. The type of demand, which is dictated by the planning environment, tells whether a backlog can be implemented or not, and what function that may represent it (Waters, 2003). Forecasting is affected by past demand information and the planning environment (Axs¨ater, 1991). The former is used to do time series

9 An Integrative Approach To Inventory Control

109

analysis, which is common practice in inventory control, while the latter may concern other input, such as information that may improve forecasting or data needed for causal forecasting. The environment may also tell of changes in the demand pattern, which may necessitate adjusting forecasting parameters or changing the forecasting method. It is necessary to consider the aggregation level of data, as longer term forecasts will have a low coefficient of variation, at the cost of losing forecast responsiveness (Vollmann, 2005). Forecast data is necessary for inventory control methods (forecasted mean values), and for safety stock sizing (forecast variability) (Waters, 2003). Safety stock sizing is a method buffering against deviations from the expected mean value of the forecast (Waters, 2003). The assumption of safety stock sizing is that all forecasts are correct estimations of the future mean value of demand; any deviations from the forecast are attributed as demand variability. This effectively means that an ill-performing forecast simply detects higher demand variability than a good forecast. The sizing is also affected by the planning environment, as the uncertain time determines the need for safety stock, as lead times also may have variability, and as the environment will determine to what extent customers accept shortages, or low service (Axs¨ater, 1991). Inventory control methods rely on forecasts, on safety stock sizing and on the planning environment. The safety stock is used as a cushion to maintain service, while forecasts and data from the planning environment, which are ordering costs, holding costs and lead times, are used to determine when and/or how much to order (Vollmann, 2005). The actual balancing of supply, which comes from the replenishment of the inventory, and demand, which is sales or lost sales, takes place as inventory transactions. Measuring these transactions gives an understanding of how well an inventory control system performs for the given planning environment (Waters, 2003).

9.3 Empirical Data Data was collected from a local timber yard, currently not using an inventory control policy. Existing functionality for the reorder point method and for the periodic order quantity method allowed for these methods to be deployed at low cost. The issue was whether the methods could cope with the fluctuations in demand, as trend and seasonal components were assumed to exist. Based on an analysis of sales data, the demand for timber was found to be seasonal, but with no trend component. This information was used to generate a demand function, based on the normal distribution. The purpose of the demand function was to allow the simulation model to run several times (200 independent simulations of three consecutive years were run, with random seeds for each simulated day). Real demand, as well as a sample of simulated demand, is shown in Fig. 9.3. Demand characteristics are shown in Table 9.1 and other parameters pertaining to the planning environment are shown in Table 9.2. As transport costs were considered to be semi-fixed rather than variable, the reordering cost is valid for the reorder

110

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

Fig. 9.2 Framework describing relations within inventory control systems

quantity used. Increasing the order quantity was not a cost-effective option. Stock out costs were not considered, as the consequences of stock outs are hard to measure; not only are sales lost, there is also the possibility of competitors winning the sale, and of losing customers, as they cannot find what they need. Lead times were considered as fixed, as no information on delivery timeliness was available. The expected fill rate (fraction of demand serviceable from stock) for the reorder point method was 99%, while the it for the periodic order quantity method would be 98% and for the lot-for-lot method would be 96% (calculated

9 An Integrative Approach To Inventory Control

111

Fig. 9.3 Simulated and actual monthly demand Table 9.1 Demand characteristics Demand Year Month Day

μ σ

56316 4693 156 710 205 37

Table 9.2 Environmental parameters Parameter Lead time (days)

Value 7

Reordering cost (SEK) 3200 Holding rate (%) 20% Unit cost (SEK) 4.49 Order quantity (ROP) Order interval (POQ) Order interval (L4L)

6500 45 7

112

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

using the loss function, based on the standard deviation, as described by Axs¨ater (2006).

9.4 Simulation Model To test how the framework could be applied to a real-world scenario, a simulation model was constructed to evaluate inventory control solutions considered by a local timber yard. To support the inventory control systems, some simply managed forecasting systems were chosen. The complete selection of methods, put into the context of the framework is shown in Fig. 9.4.

Fig. 9.4 Methods placed in the framework

All methods were verified against theory by testing, if the method implementations gave the values that theory dictates. For the reorder point method, the reorder point was raised by days of forecasted demand, to prevent undershoot, as described in (Mattsson, 2007). Several forecast methods were considered, and the actual choice of forecast for this case was based on the mean absolute deviation.

9 An Integrative Approach To Inventory Control

113

Bias was calculated to see whether a forecasting method followed the mean of demand. The seasonally adjusted moving average (Waters, 2003) was chosen as the preferred method, as it proved to be nearly as accurate as Holt-Winters (Axs¨ater, 1991), while not requiring as careful calibration. Table 9.3 Summary of forecast errors Type

EXP MA HOLT H-W S-MA

MAD 1281 1189 1267 179 263 Bias(%) 0.9% 0.3% 0.5% 0.1% -0.5%

Forecasts were monthly, and predicted the demand for the following month. The forecast value was multiplied by 1.5 to reflect an economic inventory cycle time of 45 days. This simplification was done to see how the system would react to systematic design errors.

9.5 Simulation Results To investigate how seasonality affects the performance of simple planning methods, variations of the presented demand pattern were simulated. In the first, demand was constant (Case 1); in the second, variance was added (Case 2), while in the third, seasonality (±20%) was introduced (Case 3). For the different cases a moving average forecast was used, adjusted for seasonality. Ignoring safety stock altogether, the measures used for the simulations were fill rate and inventory cycle service level. It should be noted that lot-for-lot has a different inventory cycle from the other reordering mechanisms, meaning that its service levels cannot be directly compared to them. Fig. 9.5 shows the results of the simulation runs. What can be seen is that inventory cycle service is higher for the periodic order quantity method than for the reorder point method. However, the fill rate shows that the reorder point method is better at meeting demand. Together, the measures tell that the periodic order quantity method has fewer, but greater stock-outs than the reorder point method. Seasonality affected the fill rate of the periodic order method the worst, while the reorder point method was hardly affected. The final situation added the seasonal pattern of the original data, and considered another inventory control solution, namely lot-for-lot. The result is shown in Figure 9.6, which is the fill rate and the holding cost incurred, compared to the theoretical fill rate, should no seasonality or forecast error be present. The periodic order quantity system shows a considerably worse fill rate than the reorder point system. The lot-for-lot system has a much lower holding cost than the other two methods due to more frequent orders; however, safety stock accounts for a larger fraction of the total holding, than for the other methods, leading to higher safety stock costs per stored unit. Observing the graphs, the lot-for-lot method is the only method where the mea-

114

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

Fig. 9.5 Cycle service levels and fill rates

sured cost/service relationship does not intersect with the theoretical function. For the methods with longer order cycles, the measured cost/service relationship shows a much flatter curve than expected, indicating that these functions will require more safety stock than expected to improve fill rates.

9.6 Discussion Completed simulations indicate that the fill rate of the periodic order quantity method suffers when the seasonal demand pattern is introduced, while the reorder point method can maintain the same fill rate as if no seasonality were present. This is a result of the nature of the two methods, where variability affecting the reorder point method will affect the time of ordering, while the periodic order quantity method, with fixed ordering times, cannot regulate order timing to prevent stock outs. Instead, it must let the inventory level take the full effect of any variability. Conversely, the effect of variability on the reorder point method is that the resulting order interval may not be economic. When comparing the methods used in the simulation using the framework, we find that the reorder point method is superior both concerning holding costs, and fill rate. What system is preferable depends on whether suppliers can manage to deliver varying quantities (up to three times the average, both for periodic ordering, and for lot-for-lot) or at varying intervals.

9 An Integrative Approach To Inventory Control

Fig. 9.6 Total inventory costs and fill rates

115

116

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

The difference between measured and theoretical fill rate demonstrated by the periodic order quantity shows how inventory control methods not designed for a certain planning environment can be affected. The use of a monthly forecast not representing the next inventory cycle may also have contributed to the low fill rate. The simulation based on the framework helped give insight into how the inventory control process would react to the planning environment. It showed that a large safety stock would be required if the periodic order quantity were to be used, as the periodic order quantity method undershot performance predictions much more than the reorder point method. If applied over multiple products, the framework can tell if consolidation using the sensitive periodic order quantity system is less costly than the reorder point system. Given that the periodic order quantity system has a 100% uncertain time (Axs¨ater, 1991), it may be used as a benchmark in simulations, as variability and problems caused by poor design of the process always are reflected in the fill rate. In the future we plan to enlarge our simulation experiments by incorporating different kind of demand types (continuous and discontinuous) as well as new methods used in forecasting and ordering. Recent research has shown that autoregressive forecasting methods outperform others in situations where demand is fluctuating widely and follows a “life-cycle” pattern (Datta et al, 2009). Similarly, purchasing order method research argues that not a single ordering method should be used (so basically it is not a question, which method is the best one, but which one best suits the environment), but usually a combination of different purchasing methods should be incorporated in ERP systems during the entire life-cycle of a product (Hilmola et al, 2009). However, if volumes are low, then even economic order quantities/reorder point systems, and periodic order policies should be abandoned; a lot for lot policy might produce best results in these situations (Hilmola et al, 2009). Thus, much depends from the operations strategy (order or stock based system), and from the amount of time, which customers are willing to wait for a delivery to reach their facilities (Hilletofth, 2008).

9.7 Conclusions Treating inventory control and forecasting as separate activities, while not acknowledging how forecasting and its application affect inventory control may lead to incorrect assessments of a system’s performance in a certain planning environment. Approaching inventory control as a process, starting with a planning environment and ending with a measurement of the system’s performance shows that all activities are related, and that the end result may be affected by the activities or by the way they are connected. This paper uses a simulation model to show how the use of forecasts and complexity in demand patterns affects the performance of the reorder point system and the periodic order quantity system. Simulations show that performance generally is worse than expected, and that periodic ordering consistently shows a greater susceptibility both to variability and to design errors, due to its inability to

9 An Integrative Approach To Inventory Control

117

buffer against these by changing the reordering interval. This weakness also appears in lot-for-lot systems, as they are based on periodic ordering.

References Ashby WR (1957) An Introduction to Cybernetics. London: Chapman & Hall, London, UK Axs¨ater S (1991) Lagerstyrning. Studentlitteratur, Lund, Sweden Axs¨ater S (2006) Inventory Control. Springer Verlag, New York, USA Das C, Tyagi R (1999) Effect of correlated demands on safety stock centralization: Patterns of correlation versus degree of centralization. Journal of Business Logistics 20:205–214 Datta S, Granger CWJ, Graham DP, Sagar N, Doody P, Slone R, Hilmola O (2009) Forecasting and risk analysis in supply chain management GARCH Proof of Concept. ESDCEE, School of Engineering, URL http://dspace.mit.edu/bitstream/handle/1721.1/43943/GARCH%20Proof%20of%20Conept%20 %20Datta Granger Graham Sagar Doody Slone Hilmola %202008 December.pdf?sequence=1 Ghobbar A, Friend C (2004) The material requirements planning system for aircraft maintenance and inventory control: A note. Journal of Air Transport Management 10(3):217–221 Higgins J (1976) Information systems for planning and control: Concepts and cases. Edward Arnold, London, UK Hilletofth P (2008) Differentiated Supply Chain Strategy Response to a Fragmented and Complex Market. PhD thesis, Chalmers University of Technology, Department of Technology Management and Economics, Division of Logistics and Transportation, Gteborg, Sweden Hilmola OP, Ma H, Datta S (2009) A portfolio approach for purchasing systems: Impact of switching point. Massachusetts Institute of Technology, no. ESD-WP2008-07 in Working Paper Series, URL http://esd.mit.edu/WPS/2008/esd-wp2008-07.pdf Jonsson P, Mattsson S (2006) A longitudinal study of material planning applications in manufacturing companies. International Journal of Operations & Production Management 26(9):971–995 Leknes H, Carr C (2004) Globalisation, international configurations and strategic implications: The case of retailing. Long Range Planning 37(1):29–49 Mattsson S (2004) Logistikens termer och begrepp. PLAN, Stockholm, Sweden Mattsson S (2007) Inventory control in environments with short lead times. International Journal of Physical Distribution & Logistics Management 37(2):115–130 Mattsson S, Jonsson P (2003) Produktionslogistik. Studentlitteratur Pidd M (1988) Computer simulation in management science. John Wiley & Sons, Hoboken, USA

118

Philip Hedenstierna, Per Hilletofth and Olli-Pekka Hilmola

Rantala L, Hilmola O (2005) From manual to automated purchasing Case: Middlesized telecom electronics manufacturing unit. Industrial Management & Data Systems 105(8):1053 – 1069 Vollmann T (2005) Manufacturing planning and control systems for supply chain management. McGraw-Hill Waters D (2003) Inventory Control and Management. Wiley, Hoboken, USA Zinn W, Levy M, Bowersox D (1989) Measuring the effect of inventory centralization/decentralization on aggregate safety stock: The ’square root law’ revisited. Journal of Business Logistics 10(1):1–13

Chapter 10

Rapid Modeling of Express Line Systems for Improving Waiting Processes No´emi Kall´o and Tam´as Koltai

Abstract In time-based competition, one of the main management objectives in services is to decrease customers’ waiting. Accordingly, search for designs of queuing systems which reduce waiting has become a major concern of managers. A frequently used solution is the application of express lines. The operation of express line systems can be optimized based on different objective functions. The minimization of average waiting time and the reduction of the variance of waiting times are the classical objectives for operation managers. According to perception management, however, the perceived waiting times and satisfaction generated by waiting should be considered as well. To analyze the effects of different management objectives on the operation of express line systems, a numerical and a simulation model were developed. The study of a superstore shows that the rapid numerical model and the time-consuming simulation model provide the same result when the parameter values ensuring optimal operation must be determined. Consequently, in these problems, simulation can be substituted efficiently by rapid modeling.

10.1 Introduction Companies, which are successful in cost- and quality-based competitions, look for other factors that can help them to gain further competitive advantage. Therefore, time-based competition spreads among leading companies. Time has turned into No´emi Kall´o Department of Management and Corporate Economics, Budapest University of Technology and Economics-Hungary, 1111 Budapest, M¨uegyetem rkp. 9. T. e´ p. IV. em. e-mail: [email protected] Tam´as Koltai Department of Management and Corporate Economics, Budapest University of Technology and Economics-Hungary, 1111 Budapest, M¨uegyetem rkp. 9. T. e´ p. IV. em. e-mail: [email protected]

119

120

No´emi Kall´o and Tam´as Koltai

a strategic resource and, as a consequence, its importance has become equivalent to the significance of money, productivity, and innovation (Stalk, 1988). That is, competitiveness nowadays requires balancing quality, cost, and time. In time-based competition environment, one of the main service management objectives is reducing waiting times. The simplest way for decreasing customer waiting is using additional servers (Hillier and Lieberman, 1995). This kind of waiting time reduction is, however, quite expensive. Consequently, search for the best configuration of waiting lines and service facilities has become a major concern of service managers (Hill et al, 2002). A queuing system configuration frequently used to reduce waiting times of some customers is the application of express checkouts. When express checkouts are applied, two customer groups are created: the first group consists of customers buying few items while all other customers belong to the second group. Customers buying more items than a certain number have to use the regular checkouts. Costumers buying less items than or equal quantity to this certain number can join the express lines. The number of items that controls line-type selection is called limit value. Our analyses revealed that one of the parameters which influence the waiting process when express lines are applied is the limit value. With different limit values, different waiting characteristics can be achieved. A suitable limit value can minimize a particular waiting characteristic. However, instead of the specific measures, the waiting process as a whole should be optimized. An important management objective connected to express line systems is to determine a limit value which optimizes the performance of the queuing system. This article presents tools to determine this value by reviewing some operational issues related to express line systems. The article is structured as follows. First, the tools developed for analyzing express line systems are presented. An analytical model (based on the results of queuing theory) and a simulation model (working according to the real process of a queuing system) are used for the analyses. For checking their validity, the real data of a do-it-yourself superstore was used. Next, different management objectives related to the service-profit chain, stretching from waiting minimization to satisfaction maximization, are discussed. Later, the results based on the case of the superstore are presented to compare the effects of the different management objectives on the performance of the system. Finally, the main conclusions are summarized.

10.2 Tools for Analyzing Express Line Systems Express line systems, like most queuing problems, can be modeled both analytically and empirically. Analytical models are based on the results of queuing theory. Generally some existing analytical models are used to approximate the operation of the queuing system. These models are quite simple to use; however, in the case of complex queuing systems, they give only rough estimation of the real operation. For analyzing these problems, new analytical models must be developed or simulation models can be used (Hillier and Lieberman, 1995). Simulation modeling requires

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes

121

more time and resource; however, quite special characteristics of queuing processes can be modeled in this way. For our analyses, an analytical and a simulation model were created as well.

10.2.1 Analytical Approach Queuing systems with express lines have several special characteristics which make their analytical modeling difficult. The most important specialty is that express lines are generally used in supermarkets where many service facilities are located and each has its own separate waiting line. Analyzing this kind of queuing systems with the models of queuing theory presents difficulties because there is no existing analytical model which can properly describe such a system. In this case, two analytical models can be used as approximations: one consisting of many service facilities with a common waiting line and another containing many independent queuing systems each having one service facility with its own separate queue. If analytical formulae have to be used for the whole queuing system - containing k checkouts and k waiting lines -, the following two approaches can be used: One-common-line approach. For this, the queuing system can be modeled as if all checkouts had one common queue. For this, a G/G/k model can be applied or, according to the system characteristics, a special type of it (for example M/G/k or M/M/k). If there are E express and R regular checkouts, then a model with k=E and another with k=R are required. Modeling the checkout system as a queuing system with one common line for all checkouts is an optimistic approach. It underestimates the average waiting time by assuming optimally efficient queue selection of customers which minimizes their waiting times. That is, it supposes that customers always choose the queue in which their waiting time will be shortest, and if their waiting line moves too slowly, they jockey among the queues. In some cases, however, customers cannot behave in the most efficient way. If there are idle checkouts but jockeying to these lines is difficult, or jockeying does not provide considerable time savings, then customers do not change line. Consequently, the one-common-line approach provides a best-case estimate of the operation of the queuing system. Independent-queuing-system approach. In this case, k independent G/G/1 models are applied or, according to the system characteristics, other special models (for example M/G/1 or M/M/1). If there are E express and R regular checkouts, then E+R models are required. Modeling the checkouts of a supermarket as independent queuing systems gives a pessimistic approach of waiting since it overestimates the average waiting time. Waiting lines are, however, generally not independent from each other, which can help to reduce the average waiting time. First, most of the arriving customers try to join the shortest queue. Second, some customers jockey from slowly moving lines to fast moving ones. If, for example, some checkouts become idle, customers waiting in line try to jockey to the idle checkouts. That is, queues are not independent from

122

No´emi Kall´o and Tam´as Koltai

each other. As a conclusion, the independent-queuing-system approach provides a worst-case estimate of the operation. It depends on the system characteristics, which of the presented approaches gives more accurate approximation of actual operation. If customers estimate the workloads of servers before selecting waiting line, the application of the one-commonline approach gives a better approximation (Rothkopf and Rech, 1987). For some reasons, however, customers do not always selects the queue with the minimal workload. For example, when workloads cannot be observed, the best decision cannot be made. In this case, the independent-queuing-system approach is more suitable. To create a numerical model for analyzing express line systems the approaches discussed above were used as best-case and worst-case estimations of waiting time. In Fig.10.1, a part of the numerical model can be seen - with the data of a do-ityourself superstore (Sec. 2.3).

Fig. 10.1 Numerical model

Introducing express lines into a queuing system requires the consideration of several operational issues. One of the main questions related to develop an express line system is to determine the limit value which optimizes the operation of the system. To make this decision, the effect of all possible limit values must be examined. As the limit value determines which customers use the express and which the regular checkouts, for this analysis, characteristics of customer groups generated with all possible limit values must be determined. These main characteristics are the arrival

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes

123

rates, the average service times and variances of service times. Before introducing express checkouts, this information is unavailable, that is, it must be determined by using the data of the existing system. For building the analytical - and the simulation - models only such information and data were used which can be determined without the actual introduction of the express lines. The data used can be obtained by observing and measuring the operation of the existing queuing system without express lines. Therefore, decisions about the implementation of express lines can be made in advance and the possible effects on customer waiting can be forecasted. For determining the service characteristics of different customers the relationship between the number of items bought and the service times must be analyzed. By using this relationship, the average service time and the variance of service times can be determined for customers buying a certain amount. With the help of the distribution function of the number of items bought, the average arrival and service rates, and the variances of services times can be calculated for all possible customer groups as well (for details see Koltai et al, 2008). The model works in the following way. Based on the main characteristics of the existing queuing systems (in italics), the special characteristics of the express line systems with different limit values are determined. With these parameters, using the formulae of M/G/1 and M/G/k queuing models, the average waiting times also can be calculated (typed boldface). Knowing all possible waiting times the smallest one must be selected (framed). The minimal average waiting time, eventually, determines the optimal limit value. Analyses with different parameter values showed that the waiting time as a function of the limit parameter has a distinct minimum. That is, an optimal limit value can be determined for every express line system (Fig.10.2).

Fig. 10.2 Waiting time as a function of the limit parameter

124

No´emi Kall´o and Tam´as Koltai

10.2.2 Simulation Model Express line systems have several special characteristics, and only a few of them can be taken into consideration with analytical models. For example, the managerial regulation which controls the use of checkouts can be build into analytical models. If there are more than one checkout accessible for customers, their choice among them cannot be considered. The analytical model appropriate for describing express line systems assume either uniform customer distribution among accessible waiting lines (independent-system approach) or do not deal with line selection at all (onecommon-queue approach). A simulation model, considering several customer behavioral issues, was built for studying the operation of express line systems. The block diagram of the simulation model created with Arena, simulation software of Rockwell Automation, can be seen on Fig.10.3

Fig. 10.3 Simulation model

With the first, create block, customers are generated according to a stochastic distribution. The assign block, based on a formerly defined distribution function, determines the number of items bought by each customer. With this quantity, knowing their stochastic relationship, the service time of each customer is also calculated. The branch block creates two customer groups: one of them can use the express checkouts, the other are directed to the regular lines. Customers entitled to use express checkouts buy no more items than the limit value. Customers in each group have to make a decision about which line to choose. Rules forming the base of this decision can be given in the pickq blocks. Next, the customer joins the selected queue, waits until the server will be free and can be seized. At this point, the waiting process in queue ends. The waiting time is recorded by a tally block. The following branch block is needed for data collection and statistical analyses. The customer’s route continues along the solid lines while waiting time data of the same customer group are combined in tally blocks (along the dashed lines). As the service needs a specific amount of time, the customer is delayed. When service ends, the server

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes

125

is released and made free for the next customer. At this point the sojourn time ends and it is recorded by a tally as well. After combining the different waiting time data, the customer can leave the system at the dispose block.

10.2.3 The Case Study of a Do-It-Yourself Superstore For the analyses, the real data of a do-it-yourself superstore is used. In this store, generally five checkouts operate. Using the data provided by the checkout information system, the arrival rates for the different days and for the different parts of the days was estimated. For all periods, the Poisson arrival process is acceptable according to Kolmogorov-Smirnov tests. Based on R´enyi’s limiting distribution theorem and its generalizations, the arrival processes of the two customer groups can also be approximated with Poisson processes (R´enyi, 1956; Sz´antai, 1971a,b). The density function of the number of items bought by customers is also provided by the checkout information system. For describing it, a truncated geometric distribution with a mean of 3.089 is found acceptable by a chi-square test. As the service time of customers cannot be obtained from any information systems, it was measured manually. The relationship between the number of items bought and the service time was analyzed with regression analysis. A 0.777 correlation coefficient supported the assumption of linearity. According to the results of linear regression, service time has two parts. On the average, the part independent of the number of items bought lasts 0.5463 minute, reading a bar code needs 0.1622 minute. With linear regression, the standard deviation of these parameters and the service times of customers buying different amounts were determined as well (for details see Koltai et al, 2008). Results presented in this article are valid for a midday traffic intensity with an arrival rate most characteristic for the store (λ =95 customers/hour). According to the geometric distribution, customers buy generally only few items. Therefore, two of the 5 working checkouts was considered to be express servers (S=5, E=2). In the store in question, express lines have not been used yet. Therefore, the real queuing system could not be used to validate the simulation model. Consequently, the analytical models were used for checking the validity of results. The fundamental simplifications applied in analytical models were introduced to the simulation model. In the M/G/k simulation model, there is a common line for customers entitled to use express checkouts and another one for customers buying many items. In the M/G/1 simulation model, there are independent arrival processes for all of the checkouts and their own waiting lines. The analytical and simulation results gained by the same type of models are quite close to each other; accordingly, they can be considered valid (Table 10.1).

126

No´emi Kall´o and Tam´as Koltai

Table 10.1 Analytical and simulation results Limit value Model

L=1

L=2

L=3

L=4

Analytical M/G/k Analytical M/G/1 Simulation M/G/k Simulation M/G/1

0.0492 0.3727 0.0481 0.3155

0.0417 0.2755 0.0431 0.2414

0.0811 0.3289 0.0838 0.3244

0.1563 0.4803 0.158 0.4882

10.3 Objective Functions for Operating Express Line Systems In time-based competition, the most important management objective in services is to decrease customer waiting. According to the classical objective of operations management, generally the average waiting time is minimized by service managers. With the help of queuing theory, measures related to the average waiting can easily be determined - and further on minimized. Our analyses showed, however, that as an effect of applying express checkouts, depending on the limit value, the average wait fluctuates in a high degree (Fig.10.2). With an erroneously determined limit value, not only the average waiting time but waits in express lines can be higher than in the original system. The application of express checkouts, as other specialization of servers, helps to make the services more standard. They do not necessarily reduce average waiting time but decrease the fluctuation in the length of services and, consequently, reduce the standard deviation of waiting time as well. It can happen that customers in an express line have to wait to the same amount as they would wait in a regular line but do not have to worry about that a customer buying huge amount will lengthen their waiting. Customers, being risk-averse, prefer (within limits) longer waiting with smaller fluctuations. The variance of waiting times can be determined with the statistical data of simulation or, in simple cases, with the help of queuing theory. The importance of reducing variance of waiting time attracts attention to the significance of human characteristics in service systems. According to the intention of perception management, instead of minimizing the average of objective waiting times the average of subjective waiting times should be reduced. The perceived waiting time is known only by the customer who passed through it. As people do not percept time linearly, there can be significant differences between the two values. The human time perception, according to the psychophysical literature, can be approximated by power functions (Stevens, 1957). To calculate perceived waiting times, after determining the values of some parameters, only the actual waiting times must be known. These values, however, can easily be determined by applying a suitable transformation. The evaluation of waiting depends on several factors. The same perceived waiting times can generate satisfaction to different degrees in different people. One of the main factors which determine customer satisfaction related to waiting is the value of the service for which people are waiting (Maister, 1985). Customers in express

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes

127

lines buy only few items, that is, they receive service of lower value. Accordingly, their satisfaction will be lower even if they must wait the same time as customers in the regular lines. The relationship between waiting and satisfaction can be described with a suitable utility function. For the calculations, as a simplification of the expected utility model, a mean-variance (or a two-moment) decision model can be used (Levy and Markowitz, 1979; Meyer, 1987). In this way, the transformation of waiting time into customer satisfaction, after determining the parameter values characteristic for customers, can be performed based on measures which can easily be determined analytically or empirically.

10.4 Optimization of the Waiting Process Actual waiting, perceived waiting, and satisfaction related to waiting constitute a simplified service-profit chain. Improving any of them will result in a higher service level. The ultimate management objective should be the maximization of customer satisfaction. To determine the customers’ satisfaction related to waiting, however, requires thorough knowledge about the customers and, consequently, timeconsuming and expensive analyses. The cost of this analysis may exceed the benefits which can be obtained by a system operated according to a more sophisticated objective function. Based on our analyses, it can be concluded, however, that there are no significant differences in the operation of an express line system whether it is operated according any of the first three objective functions. The numerical data for a case which is most characteristic for the superstore analyzed are given in Table 10.2. Table 10.2 The operation of the system with different objective functions and limit values Limit value Objective

L=1

Average waiting time 0.3403 Standard deviation of waiting times 0.7374 Average perceived waiting times 0.3288

L=2

L=3

L=4

0.257 0.5535 0.255

0.3125 0.5757 0.31

0.4614 0.7444 0.4491

In Table 10.2, the optimal objective function values are typed boldface. It can be seen that the same optimal limit value (L=2) is obtained independently of which objective function is used. This result has two consequences. First, managers trying to optimize the operation of their queuing systems can use, aside from satisfaction maximization, any of the possible objective functions and will get the same result (optimal limit value). Moreover in this way they will optimize (or at least improve) all of the measures mentioned.

128

No´emi Kall´o and Tam´as Koltai

Second, as average waiting time can be optimized easily and fast with analytical models, there is no need for using a more time-consuming and hardly manageable simulation model. It must be mentioned that there are situations when the different objective functions determine different optimal limit values. Our analyses showed, however, that these limit values are numbers next to each other and, in these cases, the waiting measures are nearly equal independently of the applied limit values. Therefore, even if the different objective functions give different optimal solutions for the limit value, the different limits result only slight differences in the waiting measures.

10.5 Conclusions The application of express lines is a widely used management tool for waiting process improvement. One of the main parameters of express line systems is the limit value which controls checkout type selection. Its value must be selected carefully because introducing express lines with an improper limit value can increase customer waiting significantly. Therefore, determining the optimal limit value, which minimizes average waiting time, is one of the most important tasks of managers operating express lines. For determining optimal limit value, special tools are required. Our analyses show, however, that simple analytical models are accurate enough for practical applications. They give only rough approximation of operation and they are appropriate for analyzing only simple waiting measures and management objectives; however, they can be used to determine the optimal limit value. Using analytical models, the time, money and knowledge needed for developing and running simulation models can be saved. That is, analitical models provide an effective rapid modelling tool for service managers. It also must be mentioned, that beside limit value there is another parameter which can be used by managers for influencing waiting time without cost consequences. This is the ratio of express and regular checkouts (when the total number of checkouts is constant). With this parameter, if optimal limit values are used, waiting time cannot significantly decreased, therefore it is recommended to use this parameter to assure constant limit value when total number of checkouts is changed for some reasons. The waiting time decreasing effect of express lines are limited. Notwithstanding express lines are popular among customers. Therefore, to reveal all consequences of applying express lines, their effects on waiting distribution among the different customer groups and, accordingly, on satisfaction related to waiting times must be analyzed as well. These are topics of our further researches.

10 Rapid Modeling of Express Line Systems for Improving Waiting Processes

129

References Hill A, Collier D, Froehle C, Goodale J, Metters R, Verma R (2002) Research opportunities in service process design. Journal of Operations Management 20(2):189– 202 Hillier F, Lieberman G (1995) Introduction to operations research. McGraw-Hill Koltai T, Kall´o N, Lakatos L (2008) Optimization of express line performance: numerical examination and management considerations. Optimization and Engineering pp 1–20 Levy H, Markowitz H (1979) Approximating expected utility by a function of mean and variance. The American Economic Review 69(3):308–317 Maister D (1985) The psychology of waiting lines. In: Cziepel J, Solomon M, Surprenant C (eds) The Service Encounter, Lexington Books Meyer J (1987) Two-moment decision models and expected utility maximization. The American Economic Review 77(3):421–430 R´enyi A (1956) A Poisson folyamat egy jellemz´ese (A possible characterization of the Poisson process). MTA Mat Kut Int K¨ozl 1:519-527 Rothkopf M, Rech P (1987) Perspectives on queues: combining queues is not always beneficial. Operations Research 35(6):906–909 Stalk G (1988) Time–the next source of competitive advantage. Harvard Business Review 66(July-August):41–51 Stevens S (1957) On the psychological law. Psychological Review 64(3):153–181 Sz´antai T (1971a) On limiting distributions for the sums of random number of random variables concerning the rarefaction of recurrent process. Studia Scientiarum Mathematicarum Hungarica 6:443–452 Sz´antai T (1971b) On an invariance problem related to different rarefactions of recurrent processes. Studia Scientiarum Mathematicarum Hungarica 6:453–456

Chapter 11

Integrating Kanban Control with Advance Demand Information: Insights from an Analytical Model Ananth Krishnamurthy and Deng Ge

Abstract This paper investigates the benefits of integration of advance demand information (ADI) with the Kanban Control System (KCS). ADI shared by customers is integrated into production release policies thereby enabling simultaneous improvements in service levels and reductions in inventory and costs. Under Markovian assumptions, an exact analysis of the production system is carried out. Through numerical studies, the system performance is compared to those obtained from classical Kanban control system and base stock system with ADI.

11.1 Introduction Recent advances in information technology have led to the belief that sharing advance demand information (ADI) to manufacturers will allow customers to receive better service from their manufacturing suppliers. Manufacturers also expect that this ADI can be effectively integrated into their production inventory control systems (PICS) to reduce lead times and inventories. This paper investigates the effect of integrating ADI into Kanban Control Systems (KCS). Using analytical models, we quantify the improvements obtained in system performance when the KCS is integrated with ADI.

Ananth Krishnamurthy University of Wisconsin-Madison, Department of Industrial and Systems Engineering, 1513 University Avenue, Madison, WI 53706, USA, e-mail: [email protected] Deng Ge University of Wisconsin-Madison, Department of Industrial and Systems Engineering, 1513 University Avenue, Madison, WI 53706, USA, e-mail: [email protected]

131

132

Ananth Krishnamurthy and Deng Ge

The effect of ADI on PICS has been the focus of several studies. Survey articles such as Uzsoy and Martin-Vega (1990) provide an overview of the prior research on kanban controlled systems. A number of researchers like PhiIipoom et al (1987) and Di Mascolo and Frein (1996) have studied various aspects of the design of a classical kanban controlled system. Other researchers have proposed and analyzed the performance of variations of the KCS. For instance, Dallery and Liberopoulos (2000) introduce the Extended Kanban Control System (EKCS). They show the EKCS is a combination of the classical KCS and the Base Stock (BS) system. They also show that the EKCS provides the flexibility to decouple design decisions related to production capacity and base stock levels. Buzacott and Shanthikumar (1993) introduces the Production Authorization Control (PAC) system that incorporates advance order information from customers. Karaesmen et al (2002) analyze a discrete-time make-to-stock queue and investigate the structure of the optimal policy and associated base stock levels. Liberopoulos and Koukoumialos (2005) analyze a system operating under KCS with ADI and conduct simulation experiments to investigate tradeoffs between base stock levels, number of kanbans and manufacturing lead times. The analytical model discussed here is a first step towards models that could provide understanding how system performance can be improved further by integration of the ADI with the kanban controlled system. The model presented in this paper is for a single-stage system. We compare system performance with that obtained under the classical KCS and BS system with ADI. Based on the Markov chain analysis, we show that the integration of the KCS with ADI results in superior system performance as the integration combines the best features of KCS and based stock systems with ADI. The remainder of the paper is organized as follows. Section 11.2 describes the operation of a system operating under the KCS with ADI, followed by Section 11.3 that describes the detailed Markov chain analysis for the system. Section 11.4 compares the performance of the different systems, and Section 11.5 summarizes the insights.

11.2 Kanban Control System with ADI This section describes the queuing network model of the KCS with ADI using the general framework provided in Liberopoulos and Tsikis (2003). The operational characteristics of the system are described in terms of movement of activated orders, products, and free kanbans in the network. The model is composed of a singlestage manufacturing station (MFG), fork/join synchronization stations (FJ1 , FJ2 ) and order delay station (OD). Figure 11.1 shows a schematic of the system. We assume that customer orders arrive at the system according to a Poisson process with rate λ . However, each customer places their order LTD time units in advance of the due date. We call LTD the demand lead time and let τd = E[LTD ] (the case of no ADI corresponds to the case where LTD = 0). Note that the demand lead time

11 Kanban Control with ADI

133

is customer specified and it is different from the planning lead time (LTS ) that the manufacturing system uses for planning order releases for production. Note that if sufficient ADI is available, the system might be able to meet customer demand with less finished goods inventory than that required in a system operating under the KCS without ADI. For instance, if E[LTD ] > LTS it is possible that the system operates in a make-to-order mode with minimal inventory. This paper focuses on the more interesting case wherein E[LTD ] < LTS . Consequently, orders received from customers are immediately activated. However, they may not be released into the manufacturing system immediately, as they might wait in buffer BD1 for a free kanban to be available in queue FK. When a free kanban is available in FK, an activated order in BD1 and a free kanban are matched together and released into the manufacturing stage MFG which consists of a single exponential server with mean service time μs−1 . After completing service, the product queues in the finished goods buffer FG. At buffer BD2 , LTD time units after an order is placed, the customer arrives demanding a product. If a unit is available in finished goods, the demand is immediately satisfied. The kanban attached to the order is released and routed back to FK where it is available to release another activated order into production.

K Kanbans FK

MFG

FG (Z) Satisfied Demands BD2

BD1

FJ1

Demands

τd

FJ2 OD

External Orders Fig. 11.1 Queueing network model of the KCS with ADI

We assume that (i) the number of kanbans, K in the system is fixed; (ii) demands that are not satisfied immediately get back-ordered; (iii) the system maintains a target base stock level, Z of finished products in FG. The factors affecting system performance are the demand and planning information, target base stock levels (Z), the number of kanbans (K), and the characteristics of the demand and manufacturing processes. The service times at the manufacturing station and the inter-arrival times of demands and orders are assumed to be independent. Since orders arrive at rate λ , and the service rate of the manufacturing station is μs , we assume that the system utilization is ρ = μλs ≤ 1.

134

Ananth Krishnamurthy and Deng Ge

To analyze the dynamics of the system, we define the following performance measures at time t: C(t)= the number of kanbans/parts available at the manufacturing stage, F(t)= the number of free kanbans available at FK, P(t)= the number of pending orders waiting for free kanbans at BD1 , I(t)= the number of finished items in FG, W (t) =the number of waiting orders in OD, and B(t)= the number of backorders in BD2 . The dynamics of the KCS with ADI implies that the following flow conservation equations hold at any time t: F(t) + C(t) + I(t) = K P(t) + C(t) + I(t) = Z + W (t) + B(t)

(11.1) (11.2)

The main performance measures of interest are the (i) average work in process, E[C], (ii) average finished goods inventory, E[I], (iii) the probability of backorder, PB , (iv) the average number of backorders, E[B], and (v) the overall average total cost E[TC].

11.3 Markov Chain Analysis In this section, we analyze the Markov chain for the KCS with ADI. To develop the Markov chain analysis, we assume that the demand lead time LTD has an exponential distribution. Let X1 (t) = F(t) − P(t), X2 (t) = I(t) − B(t), t ≥ 0, then the system performance measures defined in Section 11.2 can be uniquely determined by states (X1 (t), X2 (t)) as follows: F(t) = X1+ (t), P(t) = (−X1 )+ (t), I(t) = X2+ (t), B(t) = (−X2 )+ (t), t ≥ 0 (11.3) C(t) = K − X1+(t) − X2+ (t), W (t) = K − Z − X1 (t) − (−X2)+ (t), t ≥ 0

(11.4)

where Xi+ (t) = max{Xi (t), 0},

i = 1, 2. Note that the size of Markov chain is infinite when no limits are imposed on the number of pending orders. To solve a finite Markov chain, we assume that the number of pending orders at BD1 is at most K0 , where K0 < ∞. Then, we have the following important properties for states: −K0 ≤ X1 (t) ≤ K − Z, t ≥ 0

(11.5)

−(K − Z + K0 ) ≤ X2 (t) ≤ K, t ≥ 0

(11.6)

The Markov chain for the system is developed as shown in Fig. 11.3. The state space can be partitioned into six areas based on the number of finished goods/backorders. Let Ni be the number of states in area i, where i ∈ (1, 2, 3, 4, 5, 6). Then we have the number of states in each area as follows, N1 = K0 + 1, N2 = 1 2 (2K0 + K − Z + 2)(T − 1), N3 = K0 + K − Z + 1, N4 = (K0 + K − Z + 1)(Z − 1), N5 = K0 + K − Z + 1, N6 = 12 (K0 + K − Z + 1)(K0 + K − Z). This implies that total number of states is N = ∑6i=1 Ni .

11 Kanban Control with ADI

135

Let π (x1 , x2 ) be the limiting probability i.e. P {limt→∞ [X1 (t), X2 (t)]} = π (x1 , x2 ), where −K0 ≤ x1 ≤ K −Z, −(K0 +K −Z) ≤ x2 ≤ K. Let T = K −Z, we can write the Chapman-Kolmogorov equations for each of the six areas of the Markov chain. As an example the equations for Area 6 where −(T + K0 ) ≤ x2 ≤ −1, −K0 ≤ x1 ≤ T − 1 are given below: For x2 = −(T + K0 ), x1 = x2 + T :

μπ (x1 , x2 ) = (T − x1 + x2 + 1)τd−1 π (x1 , x2 + 1)

Area 1

0, K Tα

(T − 1)α

Tα

μ

λ

···

λ

−K0 , K

(T + K0 )α

μ −1, K − 1

μ (T + 1)α

λ

··· ···

μ

λ

−K0 , K − 1

(T + K0 )α

μ

μ

... · · · ... ... · · · ... .. .. · · · .. .. . . . . 2α μ (T + 2)α μ μ (T + 1)α ··· λ λ λ λ T − 1, Z + 1 0, Z + 1 −1, Z + 1 −K0 , Z + 1 ··· ··· α 2α (T + 1)α (T + K )α μ Tα μ μ μ · · · Area 3 λ λ λ T, Z T − 1, Z −K0 , Z 0, Z ··· −1, Z ··· μ α μ μ · · · μ T α μ (T + 1)α (T + K )α μ λ λ λ λ λ T, Z − 1 1, Z − 1 0, Z − 1 −K0 , Z − 1 ··· ··· (T − 1)α · · · μ Tα (T + K )α · · · μ μ μ ... · · · ... ... ... ... Area 4 ··· ··· ··· ··· Area 2

...

0, K − 1

λ

−1, K

(T + 1)α

μ λ

1, K − 1

λ

(11.7)

0

0

0

α

μ T, 1

Area 5

λ μ

2α

μ

T − 1, 1 α

μ

T, 0 μ

Area 6

λ

λ

··· 2α · · ·

λ

μ

(T + 1)α

λ

0, 1 λ

Tα

μ

μ

μ

(T + 2)α

−1, 1

λ μ

(T + 1)α

λ −1, 0 0, 0 ··· μ μ α ··· Tα λ λ 0, −1 −1, −1 ··· T − 1, −1 (T − 1)α · · · Tα μ ... · · · ... ··· T − 1, 0

μ

α

λ

···

(T + K0 )α

λ

λ

μ

· · · −K0, 0 · · · μ (T + K )α · · · λ −K0, −1 (T + K − 1)α ··· ... · · · ... 0

μ

···

−(K0 − 1), −(T + K0 − 1)

−K0 , 1

0

2α

μ λ

−K0 , −(T + K0 − 1)

μ

α −K0 , −(T + K0 )

Fig. 11.2 Markov chain transition diagram for the EKCS with ADI, where α = τd−1

For x1 = x2 + T : (λ + μ )π (x1 , x2 ) = μπ (x1 − 1, x2 − 1) + (T − x1 + x2 + 1)τd−1 π (x1 , x2 + 1) (11.8)

136

Ananth Krishnamurthy and Deng Ge

For x1 = −K0 : (T − x1 + x2 )τd−1 + μ }π (x1, x2 ) = λ π (x1 − 1, x2 )+ (T − x1 + x2 + 1)τd−1 π (x1 , x2 + 1) (11.9) For −K0 < x1 < x2 + T : {λ + (T − x1 + x2 )τd−1 + μ }π (x1, x2 ) = λ π (x1 − 1, x2) + μπ (x1, x2 − 1) + (T − x1 + x2 + 1)τd−1 π (x1 , x2 + 1)(11.10) These balance equations can be solved to obtain the key performance measures. However, the expressions for the performance measures of KCS with ADI are not closed form. Let Pb , E[I], E[B], E[C] be the probability of being backordered, and the expectation of I(t), B(t),C(t), respectively. Then, if ρ = μλs and τd = E[LTD ], we have: Pb =

∑

π (x1 , x2 )

(11.11)

∑

x2 π (x1 , x2 )

(11.12)

∑

|x2 |π (x1 , x2 )

(11.13)

(x1 ,x2 ):x2 <0

E[I] =

(x1 ,x2 ):x2 >0

E[B] =

(x1 ,x2 ):x2 <0

E[C] =

∑+

(x1 ,x2 ):K−x1 −x+ 2 >0

+ (K − x+ 1 − x2 )π (x1 , x2 )

(11.14)

11.4 System Comparison Since a system operating under KCS with ADI combines features of both the classical KCS and BS system with ADI, we compare the performance of all three policies assuming that the manufacturing system has same configuration and the parameters characterizing the ADI and demand arrival processes are the same. Note that analytical expressions have already been established for the performance measures of KCS and BS with ADI systems by Dallery and Liberopoulos (2000) and Karaesmen et al (2002) respectively. Table 11.1 shows expressions of performance measures for these two systems. To compare system performance under all three control policies, we introduce the expected total cost defined in Equation 11.15, where hw , h f and b are cost rates for average work in process, finished goods and backorders, respectively. E[TC] = hw E[C] + h f E[I] + bE[B]

(11.15)

11 Kanban Control with ADI

137

Table 11.1 Analytical Expressions for Performance Measures Measures

BS with Deterministic ADI

Classical KCS

Pb

ρ K+1

E[I]

e−μτd (1−ρ ) ρ Z+1 Z + λ τd − 1−ρ ρ 1 − ρ Z e−μτd (1−ρ )

E[B]

ρ K+1 1−ρ

e−μτd (1−ρ ) ρ1−ρ

E[C]

ρ (1−ρ )

ρ (1−ρ K ) (1−ρ )

K − 1−ρ ρ (1 − ρ K ) Z+1

11.4.1 Design of Experiments This section presents the design of experiments used for comparing the performance of the KCS with ADI with the classical KCS and BS system with ADI. In these experiments, the service time of the manufacturing station is assumed to have an exponential distribution with mean μs−1 = 1. The experiments are conducted by varying K = (5, 10, 20, 30), Z = (0, K/2, K) and λ = (0.5, 0.6, 0.7, 0.8, 0.9), respectively. We assume that the average demand lead time LTD (τd ) is set as τd = 0.9τs , where τs , the average flow time (the average time from order activation at BD1 till the delivery of a finished product to FG is estimated by τs μs 1−λ . Here we set K0 large enough so that the underlying Markov chain is finite, and yet no more than 0.1% of the orders that arrive are rejected from the system.

11.4.2 Effect of Base Stock Levels on the Performance Measures In this section, we discuss the effect of base stock level Z on the performance measures for the three different policies. The experiment was carried on λ ∈ {0.5, 0.6, 0.7, 0.8, 0.9} and K ∈ {5, 10, 20, 30}. For each given (λ , K), Z ranges from 0 to K. We compare E[B], E[I] and E[TC] for KCS with ADI, BS with ADI and the classical kanban system (KCS). Figure 11.3 plots the trade-offs obtained. In particular, Figures 11.3 i-a and i-b show that the average finished goods of the system operating under KCS with ADI is less than that of system operating under BS with ADI or the classical Kanban system. i.e., E[I] ≤ min(E[Ik ], E[Ibsa ]). This implies that KCS with ADI provides a better control over inventory than the base stock system with ADI or the classical KCS. Figures 11.3 ii-a and ii-b show that as Z increases, the average number of backorders decreases for both the KCS with ADI and BS with ADI, but is constant for the KCS. This is because both KCS with ADI and BS with ADI use a target stock level Z to reduce the backorders. The KCS does not set a base stock level, and hence the number of backorders in the system is constant for a given number of kanbans,

138

Ananth Krishnamurthy and Deng Ge

K. We also notice that the average number backorders of the system operating under KCS with ADI lies between those of BS with ADI and classical KCS. Figures 11.3 iii-a and iii-b show the tradeoffs with respect to total cost. We notice that for a system operating under KCS with ADI, the E[TC] function is neither convex nor concave over Z. However, for the BS system with ADI, the expected total cost is convex over Z. As expected, for the KCS the cost is constant, for a given K and λ . For low values of λ (or system load), the KCS with ADI behaves similar to the BS with ADI, but for high values of λ (or system load), the KCS with ADI achieves lower cost than BS with ADI for all values of Z.

Fig. 11.3 Effect of Z on Performance Measures

11.4.3 Effect of Number of Kanbans on the Performance Measures In this section, we study the effect of number of kanbans on the performance measures for the KCS with ADI, BS with ADI and the classical KCS. The target base stock level Z is set as Z ∗ , which is the optimal base stock level for BS with ADI hf ln h +b

system, where Z ∗ = varied from Z ∗ to 30.

f

lnρ

ρ) + μτdln(1− (Buzacott and Shanthikumar, 1993) and K is ρ

11 Kanban Control with ADI

139

Figure 11.4 plots the performance tradeoffs. In Figures 11.4 i-a and i-b, we see that for a system operating under the KCS, E[I] increases almost linearly as K increases, but for the KCS with ADI, E[I] increases initially with the increase in K, but is then bounded by the target stock level Z ∗ . This is due to the structure of the KCS system with ADI: the excess of kanbans queue up as free kanbans waiting for activated orders. This prevents release of additional kanbans into production limiting the built up of excess finished goods inventory. Figures 11.4 ii-a and ii-b show that initially E[B] decreases with increase in K, but then approaches a constant as E[I] approaches the target base stock level. The reason is similar to that given above: when E[I] approaches the target stock level, increase in K does not reduce backorders, as the additional kanbans queue up as free kanbans instead of being used to further reduce backorders. In Figures 11.4 iii-a and iii-b, we see that for a system operating under KCS, the expected total system cost, E[TC] is convex, but for a system operating under KCS with ADI, the expected total cost is neither convex nor concave. The optimal number of kanbans for the KCS with ADI appears to be close to the optimal kanban setting for the classical KCS. For either low or high λ (system load), the KCS with ADI always performs better than classical KCS.

Fig. 11.4 Effect of K on Performance Measures

140

Ananth Krishnamurthy and Deng Ge

11.4.4 Effect of Pair (K, Z) on Total Cost This section demonstrates the impact of control pair (K, Z) on the overall performance. We vary K from 1 to 30 and Z from 0 to K. For each λ ∈ {0.7, 0.8, 0.9}, we consider all 495 combinations of (K, Z) and study its impact on the total cost. Figure 11.5 shows the case of λ = 0.9. As we have seen in Fig. 11.3 iii-b and Fig. 11.4 iii-b, E[TC] does not demonstrate convexity or concavity over the control pair, and E[TC] has local minimums.

Fig. 11.5 Effect of (K, Z) on Total Cost

11.5 Conclusions and Ongoing Work This paper provides analysis of single-stage single-class production-inventory Kanban control system with ADI. We develop an analytical model for a system operating undet the KCS with ADI and compare the performance to system operating under the BS system with ADI and the classical KCS. Our results show that the KCS with ADI helps to reduce the inventory levels and backorders beyond that possible in the KCS or BS with ADI for the same system parameters. However, cost function for the KCS with ADI is neither convex nor concave, so determining optimal system parameters for a system operating under KCS with ADI is a challenge. Our ongoing

11 Kanban Control with ADI

141

work is aimed at developing detailed closed form approximations for key performance measures and optimizing overall system performance over the controllable parameters.

References Buzacott J, Shanthikumar J (1993) Stochastic models of manufacturing systems. Prentice Hall, New Jersey Dallery Y, Liberopoulos G (2000) Extended kanban control system: Combining kanban and base stock. IIE Transactions 32(4):369 – 386 Di Mascolo M, Frein Y (1996) An analytical method for performance evaluation of Kanban... Operations Research 44(1):50 – Frein Y, di Mascolo M, Dallery Y (1995) On the design of generalized kanban control systems. International Journal of Operations & Production Management 15(9):158 – 184 Karaesmen F, Buzacott JA, Dallery Y (2002) Integrating advance order information in make-to-stock production systems. IIE Transactions 34(8):649 – 662 Liberopoulos G, Koukoumialos S (2005) Tradeoffs between base stock levels, numbers of kanbans, and planned supply lead times in production/inventory systems with advance demand information. International Journal of Production Economics 96(2):213 – 232 Liberopoulos G, Tsikis I (2003) Unified modelling framework of multistage production-inventory control policies with lot sizing and advance demand information. In: Shanthikumar J, Yao D, Zijm W (eds) Stochastic Modeling and Optimization of Manufacturing Systems and Supply Chain, Kluwer Academic Publishers, pp 271–297 PhiIipoom PR, Rees LP, Tailor III BW, Huang PY (1987) An investigation of the factors influencing the number of Kanbans required in the implementation of the JIT technique with Kanbans. International Journal of Production Research 25(3):457–472 Uzsoy R, Martin-Vega L (1990) Modelling Kanban-Based Demand-Pull Systems: A Survey and Critique. Manufacturing Review 3(3):155–160

Chapter 12

Rapid Modelling in Manufacturing System Design Using Domain Specific Simulators Doug Love and Peter Ball

Abstract Simulation is an important tool for evaluating manufacturing system designs in the face of uncertainties like demand variation, supply variation, breakdowns and absenteeism. The simulation model building and experimentation stages can be long when compared to the time available for the overall manufacturing system design leading to potential benefits of simulation being limited. Thus the simulation process may be iterative for a single model/design but rarely iterates across multiple models for new design options. In order to maximise the value of simulation and to improve the design outcome the model building time needs to be minimised to keep pace with the manufacturing system design process. This paper argues that problem specific interfaces are needed for simulators to allow rapid and intuitive model creation. This paper reviews two case studies that illustrate an approach using domain-specific simulators combined with specialist software that manipulates the design data into the form required by the modelling system. The preprocessor-based simulators were developed to avoid the user having to specify any of the simulation logic which speeds up model building considerably. This paper contributes to the rapid modelling field by showing how domain specific, data driven simulators can enhance the manufacturing system design process.

12.1 Simulation in Manufacturing System Design Many of the key performance aspects of a manufacturing system are related to the effect of stochastic events on its operation and although mathematical modelling can Doug Love Aston Business School, Aston University, Birmingham, B4 7ET, U.K., e-mail: [email protected] Peter Ball Department of Manufacturing, Cranfield University, Cranfield, Bedford, MK43 0AL, U.K., e-mail: [email protected]

143

144

Doug Love and Peter Ball

help with some of these it is simulation that provides the most flexible and powerful means of estimating their impact. Reliable estimates of lead times, work in progress levels, delivery performance, resource utilization etc. all depend on proper representation of such sources of uncertainty. Determination of the robustness of the design requires study of external and internal sources of uncertainty, for example changes in volume and product mix are external to the system whilst breakdowns or scrap are internal factors. Smith (2003) reviews the literature on the use of simulation in manufacturing and lists many examples of its use in the design of manufacturing systems. However the review finds few papers that are concerned with role of simulation in a comprehensive manufacturing system design (MSD) process such as that proposed by Parnaby (1979). Kamrani et al (1998) presented a simplistic three stage methodology for cell design in which simulation was the third phase. Other examples of simulation being discussed in the context of the manufacturing system design process include Paquet and Lin (2003) who introduce ergonomic considerations and AlDurgham and Barghash (2008) who propose a framework for manufacturing simulation that covers some aspects of the design problem but is presented from a more general perspective. Conventionally simulation has been linked with the ’dynamic design’ stage of the manufacturing system design process which follows the concept and detail design phases in which steady-state conditions are assumed (Love, 1996). During these earlier stages average losses or utilization factors are assumed to cover internal uncertainties and average conditions said to apply to demand and product mix. Only at the dynamic design stage are these factors studied (and represented) in more depth so that reliable estimates of many of the manufacturing system’s key performance metrics will only be revealed at this late stage. Ideally the evaluation of dynamic performance of the manufacturing system should be included in every stage in the design process but this means that the simulation model would need to change as the engineers develop their view of the manufacturing system design. Lewis (1995) proposed a manufacturing system design methodology that incorporated just such a synchronized approach but it was never fully implemented. He suggested that the simulation model should be used throughout the system design and through the all the iterations of its development. The feasibility of such an approach clearly depends on ability of the modeller to complete the simulation re-build loop inside the time available for each stage in the system design process. If that cannot be done then inevitably the simulation will be left until the system design has stabilized to the point where major changes in the simulation model would not be needed - that is why the simulation is often built toward the end of the design project once the detail design phase is complete. Of course it means that any serious deficiencies in the design that emerge from the dynamic analysis may require expensive revision of the system architecture that could have been accommodated more easily at an earlier stage. Manufacturing system redesign is normally initiated when there is a compelling business need and that need is usually time-sensitive so that there is considerable pressure to complete the project as soon as possible. This pressure means that design team are unlikely to

Title Suppressed Due to Excessive Length

145

favour extending the project time scale even if the extra time spent on a simulation study would result in a higher quality and more robust design. Clearly if the time required to perform the simulation analysis could be significantly reduced then it would alter the trade-off between design quality and project duration in favour of the use of simulation.

12.2 Rapid Modelling in Manufacturing Systems Design During the early stages in MSD the architecture of the system may change substantially, for example the cell and related part families may be redefined completely, so simulation support through this phase implies an ability to completely rebuild the simulation model quickly. As the architecture is developed a series of models will be required to test out very different alternatives. Differences will not merely relate to number and distribution of resources but may require more fundamental revisions to reflect changes to cell families, material flow paths, work and skill patterns and machine tool capabilities. This means that the time to build a complete model from scratch is a key determinant of whether simulation can be used to support this early phase in the project. Building a model from scratch always takes a significant period of time, especially if the model is complex Cochran et al (1995) suggest that over 45% of simulation projects take more than three months and nearly 30% require over three man-months of effort. We have not been able to identify a more recent study that assessed the impact of the technical enhancements seen since that time or was focused specifically on manufacturing design projects. Speeding up model building has long been a desirable objective for simulation system developers, for example see Love and Bridge (1988). It is clear that whilst improvements have been made, the position is still seen as one in which scope exists for further improvement. For example Pedgen’s review (see Pegden, 2005) of future developments in simulation states that: “If we want to close the application gap we need to make significant improvements in the model building process to support the fast-paced decision making environment decision making environment of the future”. In response to this pressure software systems have improved considerably, notably in relation to the use of graphics and reusable elements; for examples of this trend see the Simul8, Witness and Arena systems amongst others. These developments focus on speeding up the translation of the model logic into executable code whilst other enhancements provide support for multiple runs, statistical analysis and the production of common reports, graphs etc. that help by speeding up the experimental process. However the domain independence or breadth of these systems means that the user is still required to provide much of the detail logic of the model. This is likely to be a significant task Robinson (2004) suggests the conceptual modelling stage takes around one third of the total simulation project time. Thus, although these improvements could be expected to speed up model development to some extent, they are limited by primarily addressing the coding and experi-

146

Doug Love and Peter Ball

mentation parts of the simulation model development cycle leaving the conceptual modelling phase relatively untouched. Whilst the need to repeat the conceptual modelling stage is clearly a serious inhibitor on the use of such systems in the architecture design of the manufacturing system it is a less significant issue for refinement and detail design. The refinement stages of the manufacturing system design process will generate a need for modifications to an existing model even if the underlying conceptual simulation model remains largely unchanged. The ease and speed with which these can be done will have been aided by the improvements mentioned above but may still require longer than the engineer would wish. The length of the minor modify-experiment cycle may depend on the ease with which the engineer can interact with the model to implement the required changes and perform the necessary experiments and that, in turn, may depend on the nature of the simulation software. It could be argued that some simulation systems already allow models to be built using only data without any programming and some simulation software companies may argue that their interfaces are intuitive and can be learnt very quickly. But this would not be a view shared by a typical manufacturing engineer unfamiliar with simulation interfaces or the subtle tricks needed to get the systems to represent the logic required with recourse to programming. Ball and Love (1994) point out that interfaces may make simulation packages easier to use but this does not necessary mean easy to use, i.e. easy to use describes the simplicity by which the user can create the model from data from the problem domain. Data-driven simulators are usually defined as systems that allow a user to create a running model without the need to do any programming, for example see Pidd’s definition (Pidd, 1992). Configuration options are used to define or modify the operational logic of the model usually through menus choices and the setting of entity properties. Although it is true that this approach does use ‘data’ to define the model, it may still require the user make decisions that are normally associated with conceptual modelling, for example to define model inputs, outputs, and data requirements and to decide what components are to be included and the level of detail with which they will be represented. The more freedom the system offers to the user the wider its potential range of applications will be, but the more specialist knowledge will be needed to use it. O’Keefe and Haddock (1991) present a useful diagram that demonstrates the continuum from pure programming languages through the type of system described above to highly focused problem-specific simulators that merely require data population. The approach used in the cases described here is close to the problem specific end of the range; the ‘model’ is pre-built and the options offered are limited to those that are directly related to the manufacturing system design problem itself. They would be recognised by an engineer as part of the normal specification of the design and are expressed in domain specific language. The conceptual modelling decisions have already been made and are hard-wired into the system. The model is populated by the data that is loaded into it; in these cases the data describe the products, production processes and resources that make up the real system. Normally this data will be extracted from the company databases or ERP systems and formatted be-

Title Suppressed Due to Excessive Length

147

fore uploading although Randell and Bolmsj¨o (2001) built a demonstration factory simulation that showed it was: “feasible to run a simulation using the production planning data as the only information source”. The data used for this project are very similar to those required by SwiftSim (Love, 2009) - see below. Detailed configuration options may still be required but they are defined and presented in a form and language familiar to the engineer using a problem specific interface. This means there is no need for users to learn specialist simulation concepts and terminology. Of course other aspects of the simulation art are still required, especially those related to the experimentation. The avoidance of the conceptual modelling stage altogether and the fact that the coding stage is also eliminated means that the user can move from data gathering to running model very quickly since data upload and parameter and option setting are all that are required to create a running model. The Robinson (2004) suggestion that the project time is roughly split evenly between conceptual modelling, coding and experimentation (he excludes implementation from this) means that use of this type of data-driven simulator could save up to two thirds of the simulation project time. This paper reviews two case studies that illustrate this approach.

12.3 Case Applications 12.3.1 Case Study 1 Cell Design in an Aerospace Company The aerospace case is an example of the use of a data-driven simulator in the design of a cellular manufacturing system. This company manufactures complex parts for the aerospace industry with application in both the military and commercial markets and their customers include all the major manufacturers of aircraft. Moulding and related processes are used in their manufacture so that this application was slightly unusual in that the processes were very different from those seen in a conventional machine shop. The variety of parts in the cell’s product family was also substantial - around 3200 part numbers were considered ‘live’ and each part passed through around 10 operations. The number of work centres in the cell was more modest at around 70 although many contained multiple stations. At some stations individual parts were loaded and processed by the operator whilst at others parts were loaded in bulk and the processing took place unattended. The need to changeover might be triggered by a change in part number from one batch to the next or by a change in some other property or attribute of the part. Operators were multi-skilled and those skills differed from person to person and shift working was the norm. Special tooling was used extensively and in some cases travelled with the work through several operations and could be considered an important resource limitation. In some cases parts were assembled together at certain operations so that the process route data had to include bill of materials information. In some cases the constituent parts were made in the same cell whilst in other instances they were produced elsewhere. MRP

148

Doug Love and Peter Ball

generated works orders were to be used by the company to drive the production programme for the cell. The design team recognised the potential benefit of using simulation but were concerned that it would take too long to develop a usable model given the tight timeframe that they had been given for the project. The complexity of their processes and the size of the part family were also seen as likely to extend the development time needed for the model. On the other hand the ability to test the robustness of the design was recognised as especially important for high-variety cells where shifts in product mix can cause unforeseen problems in sustaining delivery performance and utilisation levels. Since the redesign involved reorganisation of existing facilities rather than the introduction of new processes it followed that much of the data held in the company’s ERP system could be used to populate the simulation model. A revised version of an existing data-driven batch manufacturing simulator had recently become available at Aston University so it was decided to use that package for the project. The original system (ATOMS, Bridge see 1991) had employed a manual user-interface in which the engineer typed in all the relevant data and, whilst some basic data could be uploaded from files, extensive manual editing was always required before a viable model could be generated. Although the core of the system was little changed the revised facilities meant much larger models could be run and a more comprehensive range of upload options were implemented through a spreadsheet interface. These developments meant that ERP data could be used without simplification to generate the model. SwiftSim (Love, 2009) relies entirely on the base manufacturing data and a range of configuration options (that are also defined by the uploaded data) to generate a running model of a manufacturing system. The data required is extensive but is no more comprehensive than would be needed to specify the manufacturing system design. The system does not offer any programming options at all - if the required functionality is not present then it cannot be added. To ensure that its range of functionality was as comprehensive as possible the original design was based on a study of cell design practise across a UK-based multi-national company. Engineers from the company’s design task forces located in plants across the country were interviewed to identify the features that the system needed to offer. The system was also refined by application in a number of in-house redesign projects. Domain data are used to create the model directly, i.e. the data are formatted, uploaded (or manually entered into the system), run options selected and the model then executes immediately. The user defines materials (i.e. part numbers), process routes, bills of materials, work stations, work centres, operators, work patterns, skill groups, control systems (MRP, Kanban), sequencing rules (FIFO, batching etc), stock policies, suppliers and lead times, demand sources (generated or input) etc.. The model is created directly from this data. Company terminology is used throughout so, for example, actual part numbers are used and operators are given their real names. The system can generate a range of standard reports that vary in the level of detail offered from simple tables of resource utilisation to event log files that record everything that happened in a run. The original ATOMS system provided a limited graphical, schematic, representation of the simulated system that could be used for

Title Suppressed Due to Excessive Length

149

debugging and diagnostic investigation. For this type of system the graphical display of the system status is rarely used when performing experimental runs but it remains very useful for diagnostics so that aspect will be a core focus of the new graphical extension currently being considered for SwiftSim. The concern to ensure the model was built as quickly as possible and the fact that the company had no experience of the modelling system influenced their decision to employ an external consultant (one of the authors) who had knowledge of both the manufacturing design process and the simulator. This meant that there was a learning curve faced by the consultant in becoming familiar with the companys products, processes etc. This approach ensured that the first model was produced quickly but the extra communications involved did slow the iteration cycle down during later stages in the project. The raw process and sales demand data were extracted from the company’s ERP system into spreadsheets where they could be readily reformatted for upload. The data for work stations, operators, materials, process routes (including bills of material) were all handled that way. Generating demand data proved to be a little more complex as an MRP calculation was performed in the spreadsheet to convert product demand to that for the cell family parts. This had the advantage of avoiding any distortions that might have been present in a works order history extracted from ERP. The disadvantage was that the spreadsheet calculation was slow, taking 6-8 hours on average. The absence of a programming capability did not prove to be a constraint as the simulator handled all the complications of the manufacturing processes without the need for any special ‘tricks’ or deviations. The time the system needs to create a model from a spreadsheet is very short (less than a minute) and run times are also reasonable taking around an hour to run a year’s simulated operation of the cell. However the time required for initial familiarisation and analysis, data extraction and reprocessing and data validation and correction meant that the first proper model took around 100 man hours to produce including the time needed to include program the MRP explosion into the spreadsheet. This time also included the consultants learning curve of around 20 hours that would have been avoided by a SwiftSim-trained engineer. Subsequent revisions to reflect design changes or different performance requirements could be accommodated much more quickly taking around 8 man hours to revise the data set, upload and perform a test run. These times are taken from contemporaneous log of the projects task times that was used to track progress and resources used. Once the base model had been created the engineers were able to obtain feedback on design changes quite rapidly although this cycle time would have been reduced and some of the initial creation problems may have been avoided if the engineers had used the simulator themselves from the beginning of the project. The engineers were able to use the standard reports from the system and generally they provided the information needed although the ability to show an animated graphic of the cell running was seen as very desirable, especially for communicating with both senior management and the shop floor.

150

Doug Love and Peter Ball

12.3.2 Case Study 2 High Volume Production Line Design The second case study is drawn from a high volume, engineered product environment. The company regularly introduces new products which trigger the development of new production lines. A production line is developed iteratively over a number of months and simulation is used as standard practice within those iterations. There are many individuals involved in the production line design process and although many regularly use simulation only a few are considered simulation experts. The focus of the design activity is the production line, with some links to the support, supply chain, etc. activities. The initial users of the simulation output are the wider design team to trigger redesign work or to confirm performance. The final simulation output is used as part of the senior management sign off process. The role of simulation in this case is to support the activities of the manufacturing engineers in removing risk from the design process and, importantly, triggering design changes that would typically result in a 10% performance improvement. Numerous simulation models are created during the design of a production line resulting from changes to numbers of machines, machine cycle times, process quality, expected output rates, etc. The models include details of buffers, selection rules, conditional routings, scrap rates and operator behaviour. Given their size (100 entities in a model is not unusual) and the scope of the potential changes, the models are rapidly rebuilt from scratch each time rather than modifying a base model. This rapid rebuilding of models is considered more robust than model modification and the scope of the changes required mean that such modifications could take longer than is available to the overall design team. The rapid building of simulation models is achieved through a tailored spreadsheet interface to a commercially available simulation package. The users work with the interface to specify the model through either manual entry or copy and pasting data from other design spreadsheets. The data entered represents the entities to be modelled as well as the control parameters. Populating the interface the first time for a new production line design typically takes several days, however, once achieved subsequent design changes can be accommodated easily with a day, often in hours. The early modelling work takes many days as the first models are run are deterministic and stochastic enhancements are progressively added and experiments performed. Once set up, the interface is able to build the model in the simulation package, run the model a number of times and retrieve the results. The interface contains only sufficient functionality to build models for that particular company. Therefore the user works within the user interface using terminology of a manufacturing engineer rather than generalised simulation terminology and is restricted to entering data typical of that companys requirements. The overall time from start to finish of modelling a given line is of the order of weeks. Relatively therefore the model build and run time is short for a given scenario. Overall modelling effort is actually dictated by the design iterations creating new scenarios. Manufacturing engineers use the simulation interface to build and run simulation models, sometimes with the guidance of the simulation experts. The company specific functionality of the simulation interface means that data specified in the inter-

Title Suppressed Due to Excessive Length

151

face that completely defines the model creation and execution is readily understood by all, whether or not they were part of initial modelling work. This contrasts with the typical view that simulation models built by others take time to fully understand. The size of the models means that manual creation of both the model logic and the graphical display would take a significant amount of time to create; potentially the speed is such that modifications would be triggered by the manufacturing engineers before the model was completed. Experimentation times are typically of the order of hours, sometimes scenarios are batched together and run overnight to use otherwise idle computers. The modelling approach therefore uses the power of a commercial simulator to model complex and varied systems and combines this with the simplicity of an interface dedicated to the particular companys work. This separation of the model creation from the power of the simulation software enables staff to quickly create models without having to develop a dedicated simulator or use significant staff time. In summary, the approach combines the power of a commercial simulation package with the speed and ease of use of a dedicated spreadsheet based interface in the language of the manufacturing engineer and allows rapid creation of models for experimentation by simulation experts and non-experts alike. The speed of the modelling is within the pace of the wider design team activity and genuinely informs the design process triggering design iterations and confirmation of performance before sign off.

12.4 Discussion of the Cases The paper has argued that the traditional relationship between manufacturing system design and simulation needs to evolve to truly draw on the benefits of simulation as an integral part of the dynamic design stage. The discussion went further to make the case that the dynamic design should be iterative starting at the concept stage and that to enable this to take place the interface to simulation systems should be in the language of the manufacturing engineers who are making the critical design decisions. The paper has presented two different cases where simulation models have built to directly contribute to the manufacturing system design process. The following discussion reviews how far these cases go towards enhancing the design outcome. Both cases address the integration of simulation into the manufacturing system design process. The cases demonstrate the influence of simulation on the design outcome as well as confirming performance. In both cases there was a simulation expert supporting the design activities and notably in case two those manufacturing engineers who are in the core of the design team are also users of the simulation system. The cases show the use of simulation to improve design performance, however, its influence on the concept design differs according to the point at it is deployed at the design stage. Whilst the first case demonstrated the influence on the design con-

152

Doug Love and Peter Ball

cepts in the second case simulation was utilised after the production line concepts emerged and therefore its role was to improve the performance of a given design option. Design iterations can vary in magnitude from parameter changes (such as cycle times and material control rules) to more fundamental structural changes (such as number of machines and routings). In case two most iterations were parameter changes however there were occasional more fundamental changes that resulted, as would be expected, in longer model creation times. Developments in simulators have potential to improve simulation model build times and in turn influence its role in manufacturing systems design. The development of a domain specific simulator requires both simulation and application domain knowledge. Two different approaches were illustrated the case two company used a standard commercial simulation system as the basis of their simulator whereas SwiftSim was developed through an academic research project with industrial collaborators. Interestingly the developments most valuable to these cases were core functionality improvements rather than those relating to animation. The role of the manufacturing engineer in the use of simulation varies in these cases. Case one was lead by the simulation expert where as case two was supported by the simulation expert. It has to be noted that simulation experts were used when subtle ‘tricks’ are required that are not a standard part of the interface functionality. A simulation specialist may be used to interface between the engineer and the model but this approach also has drawbacks. The specialist translates the engineer’s requirements into a model suitable for the purpose but the risk is that features are lost in translation and delays result. It may be that the popularity, mentioned earlier, of simple spreadsheet models with manufacturing engineers reflects a desire to directly control all aspects of the analysis. The level of data translation required from the manufacturing engineering design world into the simulation analysis world and back influences the level of robustness of the analysis process as well as the time taken to complete it. Both cases feature a minimum level of translation of data from the manufacturing engineers world to the simulation world, hence the manufacturing engineer could readily understand the model construction and outputs. Subsequently this minimises any nervousness with verification. Both cases present implications for model build time and indicate that they are built rapidly when compared to typically quoted figures from the literature. The rapid model building has had two impacts: firstly, the simulation output was influencing the design outcome rather than just confirming performance and, secondly, the level of detail possible is very high providing greater confidence in the design outcome.

12.5 Conclusions This paper has presented a discussion on the relationship between the manufacturing system design process and the simulation modelling process. It was argued that

Title Suppressed Due to Excessive Length

153

to improve the design outcome, the model building process needs to be reduced significantly to enable the results of simulation to truly influence the selection and refinement of design concepts. The detail of the two industrial cases demonstrates the challenges for simulation use as well as the benefits obtained. From this, key issues of integration of simulation, the influence on concept design, the functionality of commercial simulators, the role of the manufacturing engineer and data translation were identified and discussed. Overall the paper has demonstrated how domain specific, data driven simulators can enhance the manufacturing system design process.

References AlDurgham MM, Barghash MA (2008) A generalised framework for simulationbased decision support for manufacturing. Production Planning & Control 19(5):518 – 534 Ball PD, Love DM (1994) Expanding the capabilities of manufacturing simulators through the application of object-oriented principles. Journal of Manufacturing Systems (6):412–442 Bridge K (1991) The application of computerised modelling techniques in manufacturing system design. PhD thesis, Aston University Cochran JK, Mackulak GT, Savory PA (1995) Simulation Project Characteristics in Industrial Settings. Interfaces 25(4):104 – 113 Kamrani A, Hubbard K, Parsaei H, Leep H (1998) Simulation-based methodology for machine cell design. Computers & Industrial Engineering 34(1):173–188, cellular manufacturing systems:Design, Analysis and Implementation Lewis P (1995) A systemic approach to the design of cellular manufacturing systems. PhD thesis, Aston University Love D (1996) The design of manufacturing systems. In: International Encyclopaedia of Business and Management V4, Thompson Business Press, pp 3154–3174 Love D (2009) SwiftSim overview. URL http://oimabs.aston.ac.uk/swiftsim Love DM, Bridge K (1988) Specification of a computer simulator to support the manufacturing system design process. In: Proceedings International Conference Computer-Aided Production Engineering, SME, Michigan O’Keefe RM, Haddock J (1991) Data-driven Generic Simulators for Flexible Manufacturing Systems. International Journal of Production Research 29(9):1795– 1810 Paquet V, Lin L (2003) An integrated methodology for manufacturing systems design using manual and computer simulation. Human Factors and Ergonomics in Manufacturing 13(1):19–40 Parnaby J (1979) Concept of a Manufacturing system. International Journal of Production Research 17(2):123 –34 Pegden C (2005) Future directions in simulation modeling. In: Proceedings of the 37th Winter Simulation Conference, pp 1–35

154

Doug Love and Peter Ball

Pidd M (1992) Guidelines for the design of data-driven generic simulators for specific domains. Simulation 59(4):237–243 Randell L, Bolmsj¨o G (2001) Database driven factory simulation: A proof-ofconcept demonstrator. In: Peters B, Smith J, Medeiros D, Rohrer M (eds) Proceedings of the 33rd conference on Winter simulation, December 9-12, pp 977– 983 Robinson S (2004) Simulation: The practice of model development and use. John Wiley & Sons, Chichester Smith J (2003) Survey on the use of simulation for manufacturing system design and operation. Journal of Manufacturing Systems 22(2):157–171

Chapter 13

The Best of Both Worlds - Integrated Application of Analytic Methods and Simulation in Supply Chain Management Reinhold Schodl

This work attempts to discover how complex order fulfillment processes of a supply chain can be analyzed effectively and efficiently. In this context, complexity is determined by the number of process elements and the degree of interaction between them, as well as by the extent variability is influencing process performance. We show how the combination of analytic methods and simulation can be utilized to analyze complex supply chain processes and present a procedure that integrates queuing theory with discrete event simulation. In a case study, the approach is applied to a real-life supply chain to show the practical applicability.

13.1 Combination of Analytic Methods and Simulation Analytic models and simulation models are opposing ways to represent supply chain processes for purposes of analysis. “If the relationships that compose the model are simple enough, it may be possible to use mathematical methods (such as algebra, calculus, or probability theory) to obtain exact information on questions of interest; this is called an analytic solution” (Law and Kelton, 2000). Conversely, simulation models are quantitative models, which do not consist of an integrated system of precisely solvable equations. “Computer simulation refers to methods for studying a wide variety of models of real world systems by numerical evaluation using software designed to imitate the system’s operations or characteristics, often over time” (Kelton et al, 2002). The use of analytic models and simulation models in supply chain management harbors distinct merits and demerits (see Table 13.1). By combining analytic methods and computer simulation, one can potentially derive greater value than by applying one of these methods alone. This idea has been advocated since the early Reinhold Schodl Capgemini Consulting, Lassallestr. 9b, 1020 Wien, Austria, e-mail: : [email protected]

155

156

Reinhold Schodl

Table 13.1 Analytic models compared to simulation models Strengths of Analytic Models

Strengths of Simulation Models

No limitation to descriptive models, as analytic models can be prescriptive (i.e., optimization models) as well

Ability of representing systems with comprehensive stochastic cause-effect-relationships

High significance of results (conversely, conclusions derived from stochastic simulation carry risk, because analysis is based on a limited sample size of output values generated by repeated simulation runs)

Possibility to determine not only mean values, but also distributions of output variables by approximation

Lower time and effort to adapt an existing analytic model compared to building a new simulation model (as simulation models are generally built case-specific)

High acceptance among practitioners, as often better understandable and more transparent than analytic models (especially when animations are being added to simulation models)

days of computer simulation. Nolan and Sovereign integrate analytic methods with simulation in an early work in the area of logistics (Nolan and Sovereign, 1972). Later research utilizes the joint application of the methods in the field of supply chain management (for examples, see Ko et al, 2006; Gnoni et al, 2003; Merkuryev et al, 2003; Lee and Kim, 2002). Depending on the degree of integration, one can distinguish two different forms, i.e., hybrid modeling and hybrid models. “Hybrid modeling consists of building independent analytic and simulation models of the total system, developing their solution procedures, and using their solution procedures together for problem solving” (Sargent, 1994). An example is the evaluation of alternatives based on economic viability and operative feasibility by applying an analytic and a simulation model respectively. A further application is the verification of an analytic model via an independent simulation model (Jammernegg and Reiner, 2001). “A hybrid simulation/analytic model is a mathematical model which combines identifiable simulation and analytic models” (Shanthikumar and Sargent, 1983). Hybrid models are characterized by a higher degree of integration, as analytic methods and simulation are incorporated into a single model. Hybrid models can be classified according to the type of dynamic and hierarchical integration (see Table 13.2). Following the classification with regard to dynamic integration, this work presents a Type I model. Concerning the hierarchical integration, the presented model is a special case of Type IV, as the simulation model requires the analytic model’s output, but both models are hierarchically equivalent and represent the whole system.

13 The Best of Both Worlds

157

Table 13.2 Classification of hybrid models Dynamic Integration

Hierarchical Integration

Hybrid Model Type I: “A model whose behavior over time is obtained by alternating between using independent simulation and analytic models. The simulation (analytic) part of the model is carried out without intermediate use of the analytic (simulation) part” (Shanthikumar and Sargent, 1983)

Hybrid Model Type III: “A model in which a simulation model is used in a subordinate way for an analytic model of the total system” (Shanthikumar and Sargent, 1983)

Hybrid Model Type II: “A model in which a simulation model and an analytic model operate in parallel over time with interactions through their solution procedure” (Shanthikumar and Sargent, 1983)

Hybrid Model Type IV: “A model in which a simulation model is used as an overall model of the total system, and it requires values from the solution procedure of an analytic model representing a portion of the system for some or all of its input parameters” (Shanthikumar and Sargent, 1983)

13.2 Hybrid Models for Complex Supply Chains The analysis and improvement of complex supply chain processes is a unique challenge. Given the fact there is no universally accepted definition of the complexity of supply chain processes, we define the following building factors for complexity: number of process ele-ments (e.g., activities, buffers, information, resources) and degree of interaction between them, random variability (e.g., machine failures), and predictable variability (e.g., multiple product variants). The following two approaches show that hybrid models are particularly suitable for the analysis of complex supply chain processes. • The entire system is assessed by using analytic methods (e.g., queuing theory). Subsequently, the results of the assessment are used to construct a model of a sub-system, which then helps to conduct a more detailed analysis by means of simulation. This type of approach is used, for instance, to analyze a complex supply chain in the semi-conductor industry (Jain et al, 1999). • An analytic model is employed to assess a relatively large number of alternatives with relatively minimal effort. Promising alternatives are analyzed via simulation in more detail. For instance, such approach is employed to solve complex transportation problems (Granger et al, 2001). We now present a procedure to analyze complex supply chains with a balance between validity and effort. The procedure is different from the discussed approaches in the following ways. First, narrowing of the system’s scope by an analysis on an aggregated level is avoided, to incorporate the dynamic behavior of the overall system. Second, no preselecting of alternatives by an analysis on an aggregated level occurs, which prevents an unwanted rejection of promising process designs. The procedure consists of the following steps:

158

Reinhold Schodl

1. In the first step, the real system’s supply chain processes are modeled as an analytic model and analyzed according to queuing theory. The queuing model delivers values of performance indicators (e.g., waiting times) which are inputs for the complexity reduction in Step 2, as well as for the simulation model in Step 3. 2. This step aims to reduce complexity by identifying non-critical process steps that can be modeled in a simplified manner in Step 3. If variability is not being reduced, it has to be buffered by inventory, capacity, or times in order to maintain process performance. Inventory levels, capacity utilization, and waiting times represent the degree of buffering, and therefore act as indicators of how critical a process step is. These indicators can be obtained from the queuing model. Further indicators can be derived from the real system. An example is a process step’s relative position in the queuing network as, generally, variability at the beginning of the process has greater impact than at the end. 3. In this last step, the supply chain processes are modeled as discrete event simulation model. Process steps that are defined in Step 2 as non-critical are modeled in a simplified manner. Simplification can be achieved by modeling process steps without capacity restrictions. Waiting times caused by capacity limitations are then modeled in the simulation model as constants according to the values derived from the queuing model. Finally, the simulation model is applied to analyze alternative process designs.

13.3 Application in Supply Chain Management To demonstrate the practical applicability, we applied the described procedure to a supply chain in the electronic industry, with focus on a first-tier supplier producing printed circuit boards. In a period of six months, 407 different products consisting of 817 different components are produced by 325 machines. The value-adding processes are heavily influenced by a highly variable demand as a result of short-term market trends, as well as by frequent disruptions of production due to the application of complex technologies. The overall aim of the case study is to improve service levels while taking cost constraints into account. In particular, two design alternatives are considered and compared with the initial situation: first, capacity enlargement for the main bottleneck work centers and second, implementation of the concept Make-to-Forecast (Raturi et al, 1990) based on improved short-term forecasts by using data available from Vendor Managed Inventories. The core of the case study consists of three steps, as described below.

13.3.1 Step 1: Analysis Based on Analytic Methods The supply chain processes are modeled as a network of queues to be analyzed according to queuing theory. The software MPX (Network Dynamics, Inc.) is applied,

13 The Best of Both Worlds

159

which “... is based on an open network model with multiple classes of customers. It is solved using a node decomposition approach. [Each] ... node is analyzed as a GI/G/m queue, with an estimate for the mean waiting time based on the first two moments of the arrival and service distributions. Next, the MPX solution takes into account the interconnection of the nodes ... as well as the impact of failures on the service time and departure distributions” (MPX, 2003). The analytic model’s inputs include: • Demand data (primary demand in defined period, variability of customer order inter-arrival time), • Bill of material data, • Routing data (sequence of production steps, average setup time, variability of setup time, average process time, variability of process time, work center assignment), • Resource data (parallel machines in work centers, scheduled availability, mean time to failure, mean time to repair), and • Production lot sizes. The model is validated by comparing the mean production lead time with that of the real system, which differs by less than 5%. It is then applied to find values for each work center’s capacity utilization and average value-adding times for setting-up and processing, as well as average waiting times due to capacity restrictions. This output is required for the reduction of complexity in Step 2 and for the simulation model in Step 3.

13.3.2 Step 2: Reduction of Complexity If certain resources, such as work centers, are not modeled in the simulation model, the effort for building and running the model can be reduced. This is acceptable only if simplified modeling is limited to resources aligned to non-critical process steps. Process steps are classified as critical and non-critical based on multiple criteria, as follows: • The capability of a process step to deal with variability is an important factor in evaluating how critical a process step is. Generally, if variability cannot be reduced, it has to be buffered by capacity, time, and inventory. Fundamental indicators for the degree of buffering of variability are capacity utilization and lead time efficiency. Both measures are provided by the described queuing model. • Another factor is the relative contribution of a process step to the overall performance of the supply chain, which can be measured by a process step’s proportion of value-adding time and proportion of cost of goods sold. • Moreover, a process step’s relationship with other process steps is taken into account. The relative position of a process step within the network is a relevant indicator, as generally variability at the beginning of a process has greater impact than at its end. A further indicator is a process step’s assembling functionality,

160

Reinhold Schodl

as asynchronous arrival from previous process steps, is an important cause for delays.

13.3.3 Step 3: Simulation-Based Analysis The processes of the supply chain under study are modeled as a simulation model to be analyzed in detail with multiple performance measures. After building the model, verification and validation is carried out and an experimental design is developed to finally run simulation experiments. Critical process steps are modeled detailed, i.e., resources that carry out the process steps are represented in the model, including details about scheduled availability and random breakdowns. Resources aligned to non-critical process steps are not being modeled. Because of this simplification, waiting times caused by capacity restrictions cannot be determined by the simulation model. Thus, for non-critical process steps, the waiting times calculated by the analytic model are utilized and represented as constants in the simulation model. This approach guarantees a balance between representing reality as detailed as necessary while also keeping the effort to build and run the model as low as possible. The case study’s discrete event simulation model is implemented with the software ARENA (Rockwell Automation), which is based on the simulation language SIMAN. The simulation model accounts for various risks of the supply chain, especially variable demand, forecast errors, stochastic setup times and machine breakdowns. The model’s input is comprised of: • Demand data (order time, order quantity, desired delivery date), • Forecast data (forecasted order time, forecasted order quantity, forecasted desired delivery date), • Bill of material data, • Routing data (sequence of production steps, assignment of work centers, setup time, processing time), • Resource data (parallel machines in work center, scheduled availability, mean time to failure, mean time to repair, constant waiting time for simplified modeled work centers), • Production data and rules (production lot size, rule for dispatching production orders, rule for prioritization of production orders), and • Cost data (material cost, time-depended machine cost, quantity-depended machine cost). The length of the warm-up-period of the non-terminating simulation is decided by visual analysis of the dynamic development of inventory. The number of replications is determined by statistical analysis of the order fulfillment lead time. The output of the simulation model comprises performance measures whose definitions are in line with the well-established Supply Chain Operations Reference Model (Supply Chain Council, 2009). The defined scenarios are compared with

13 The Best of Both Worlds

161

multiple performance measures, i.e., delivery performance, order fulfillment lead time, capacity utilization, cost of goods sold, and inventory days of supply. Table 13.3 Effect of complexity reduction Degree of Complexity Reduction.

Error of Order Fulfillment Lead Time.

0% 9% 29%

0.60% 1.00% 2.90%

The focus of this paper does not lie in the presentation of the scenario’s specific results, but on a demonstration of the practical applicability of the presented approach to deal with complex supply chains. Therefore, the validation of the simulation model under different degrees of complexity reduction is of particular interest. The degree of complexity reduction is expressed as the proportion of work centers modeled in a simplified manner. Table 13.3 shows how complexity reduction affects the model’s error of order-fulfillment lead time. Complexity reduction of 9% results in a generally acceptable error of the order fulfillment lead time of 1%; for a complexity reduction of 29%, the error is still under 3%. For further validation, statistical analysis of the order fulfillment lead times of the customer orders was carried out. A Smith-Satterthwaite test is utilized, as the system and model data are both normal and variances are dissimilar (Chung, 2004). For a level of significance of 0.05 and a degree of complexity reduction of zero and 9%, there is no statistically significant difference between the actual system and the simulation. For a level of significance of 0.01, this is also true for a complexity reduction of 29%.

13.4 Conclusion Analytic models and simulation models are characterized by specific strengths and weaknesses. In this paper, we demonstrated a procedure that combines an analytic queuing model with a discrete event simulation model to utilize the specific benefits of both methodological approaches. A balance between validity of the results and effort for the analysis of the supply chain processes was accomplished.

References Chung C (2004) Simulation modeling handbook: A practical approach. CRC Press, Boca Raton

162

Reinhold Schodl

Gnoni M, Iavagnilio R, Mossa G, Mummolo G, Di Leva A (2003) Production planning of a multi-site manufacturing system by hybrid modelling: A case study from the automotive industry. International Journal of production economics 85(2):251–262 Granger J, Krishnamurthy A, Robinson S (2001) Stochastic modeling of airlift operations. In: Proceedings Winter Simulation Conference 2001, IEEE Computer Society Washington, DC, USA, vol 1, pp 432–440 Jain S, Lim C, Gan B, Low Y (1999) Criticality of Detailed Modeling in Semiconductor Supply Chain Simulation. In: Proceedings of the Winter Simulation Conference 1999, ACM New York, NY, USA, vol 1, pp 888–896 Jammernegg W, Reiner G (2001) Ableitung und Bewertung von Handlungsalternativen in einem Unternehmen der Elektroindustrie. In: Jammernegg W, Kischka PH (eds) Kundenorientierte Prozessverbesserungen, Konzepte und Fallstudien, Springer, Berlin, Berlin, pp 237–247 Kelton W, Sadowski R, Sturrock D (2002) Simulation with ARENA, 2nd edn. McGraw-Hill Science/Engineering/Math, Boston Ko H, Ko C, Kim T (2006) A hybrid optimization/simulation approach for a distribution network design of 3PLS. Computers & Industrial Engineering 50(4):440– 449 Law A, Kelton W (2000) Simulation modeling and analysis, 3rd edn. McGraw Hill, New York Lee Y, Kim S (2002) Production–distribution planning in supply chain considering capacity constraints. Computers & Industrial Engineering 43(1):169–190 Merkuryev Y, Petuhova J, Grabis J (2003) Analysis of dynamic properties of an inventory system with service-sensitive demand using simulation. In: Proceedings of the 15 th European Simulation Symposium-Simulation in Industry, Delft, The Netherlands, pp 509–514 MPX (2003) MPX WIN 4.3 - For use with Windows, User Manual. Network Dynamics Inc, Framingham Nolan R, Sovereign M (1972) A recursive optimization and simulation approach to analysis with an application to transportation systems. Management Science 18(12):676–690 Raturi A, Meredith J, McCutcheon D, Camm J (1990) Coping with the build-toforecast environment. Journal of Operations Management 9(2):230–249 Sargent R (1994) A historical view of hybrid simulation/analytic models. In: Proceedings of the Winter Simulation Conference, pp 383–386 Shanthikumar J, Sargent R (1983) A unifying view of hybrid simulation/analytic models and modeling. Operations Research 31(6):1030–1052 Supply Chain Council (2009) SCOR Model. URL http://www.supplychain.org/cs/root/s/scor model/scor model

Chapter 14

Rapid Modeling In A Lean Context Nico J. Vandaele and Inneke Van Nieuwenhuyse

Abstract Lean management is widespread but theoretical models that scientifically substantiate the lean practice are scarce. We show how queuing models of manufacturing systems and supply chains underpin the practice of Lean. Two quantitative performance models which relate the system parameters with the system performance in terms of lead time and throughput will be discussed, including an exogenous definition of Lean. We show that the ideal level (i.e., the Lean level) of the system buffers (safety capacity, work-in-process and safety time) are determined by the targeted system performance. Moreover, the lean concept is dynamic in nature: when either system characteristics or target performance change, the lean buffer levels change accordingly. The latter stresses the need for a comprehensive, analytical and consistent approach. This quantitative approach will be illustrated with lead time and throughput models.

14.1 Introduction In industrial practice, the concept of Lean operations management is the hype of the new millennium. It consists of a set of tools that assist in the identification and steady elimination of waste (muda), the improvement of quality, and production time and cost reduction. The concept of Lean operations is built upon decades of insights and experience from Just-In-Time (JIT) applications. Since the first articles and books Nico J. Vandaele Research Center for Operations Management, Department of Decision Sciences and Information Management, K.U. 3000 Leuven, Belgium, e-mail: [email protected] Inneke Van Nieuwenhuyse Research Center for Operations Management, Department of Decision Sciences and Information Management, K.U. 3000 Leuven, Belgium, e-mail: [email protected]

163

164

Nico J. Vandaele and Inneke Van Nieuwenhuyse

appeared, a tremendous number of publications became available for the practitioner (e.g. Womack et al (1990), Womack and Jones (1997), Liker (2004)). Many big companies adopted company-wide leadership programs to get their company on the lean track. In academia, the Lean concept has received rather limited attention, based on the reproach that Lean did not offer much more on top of the traditional JIT body of knowledge, and hence should be qualified primarily as a philosophy without rigorous scientific foundations. This is only partly true. It is indeed true that the scala of lean management techniques is often rather descriptive, failing to offer analytical modeling and/or optimization tools that could guide managerial decision making. Cosnider for instance the heavily promoted tool Value Stream Mapping as presented in the best seller ‘Learning To See’ (Rother and Shook, 1999). Value Stream Mapping forces to map the logical structure of the processes and adds valuable quantitative data both on system parameters as on performance measures. However, Value Stream Mapping is incapable to give insight in the relationships between the data and performance measures as well as the links between various elements of the mapped processes. In this paper, we present two quantitative models to underpin the definition of lean in terms of the lead time and the throughput of a flow system. As will be shown, the quantification of Lean boils down to an adequate match between the system’s parameters and target system performance. A mismatch between the two can cause a system to be either obese, or anorectic.

14.2 The Quantification Of Lean In order to develop a quantitative approach to the Lean concept, we will rely on some basic stochastic models for flow systems. Flow systems are systems where a set of resources is intended to perform operations on flows (see Vandaele and Lambrecht, 2003). Some illustrative examples are listed in table 14.1. Table 14.1 Some flow system examples Flow System

Typical Resources

Typical Flows

Production line Production Plant Hospital Airport Traffic Roads Laboratory Equipment Computer network Mobile phone network Insurance company

Machines, workers Machines, internal transport Hospitals beds, physicians, nurses Counters, desks traffic lights laboratory assistants Servers data lines Antenna’s, transmitters, buffers Inspectors, account managers

Products Products Patients Passengers Cars, trucks Samples Data, messages Calls Files

14 Rapid Modeling In A Lean Context

165

These examples show the rich variety of flow systems. All these systems share some common physical characteristics: on their routing through the system, flows visit resources in order to be processed, and hence consume (part of the) capacity of the resources. This competition for capacity causes congestion: flows may need to queue up in front of the resources. This congestion in turn inflates the lead time of a flow entity through the system. These basic mechanics of a flow system imply that every decision related to the flow has consequences for the resource consumption over time. For instance, once lead time off-setting for manufacturing components in an assembly setting is performed, resources (i.e., capacity) need to be committed in order to be able to perform the required processes for the components. Vice versa, all resource-related decisions have an impact on the flow: scheduled maintenance for instance will temporarily impede flow, while sequencing decisions cause certain flows to proceed while other flows need to wait. Consequently, flow systems contain three fundamental decision dimensions: flows, resources and time. If a flow system is to be managed in an effective and efficient way, the management decisions must consider the flow, resource and time aspects simultaneously, symbolized by the intersection visualized in Fig.14.2.

Fig. 14.1 The basic dimensions and buffers of a flow system

In what follows, we assume the flow system to be stochastic: i.e. both the flows and the resources are subject to variability. In real-life systems, causes of system variability are omnipresent, like quality problems, resource failures, stochastic routings, randomness, etc. (see for instance Hopp and Spearman, 2008). It is known that the presence of variability influences system performance in a negative way (Vandaele and Lambrecht, 2002). Important system performance measures are resource utilization, flow time, inventory, throughput and various forms of service level. Some of these (e.g. flow time and inventory) are flow oriented, while others (such as utilization and throughput) are resource oriented. In order to maintain an acceptable performance in a stochastic environment, a flow system has to operate with buffers (Vandaele and De Boeck, 2003). Conform the three basic system dimensions mentioned above, three types of buffers may be used: inventory buffers (e.g. safety stocks, work-in-process,), capacity buffers (spare capacity, temporary labor,) and time buffers (safety time, synchronization buffers,...). Any particular combination of the three buffers leads eventually to a

166

Nico J. Vandaele and Inneke Van Nieuwenhuyse

specific performance level. These buffers can be interchanged in order to reach a desired performance level. For instance, short lead times combined with the absence of work-in-process -mandatory in e.g. Just-In-Time systems- can only be guaranteed with lots of excess capacity. The amount of excess capacity (capacity buffer) can be reduced if variability and/or randomness are eliminated (see Hopp and Spearman, 2008). Tight capacity typically leads to high utilization at cost of long lead times and high work-in-process. Consequently, there may be several combinations of fundamental buffers leading to the same performance. These can be considered as some kind of technically equivalent buffer combinations. A monetary evaluation in terms of costs and revenues, can lead to the best economic choice between the technically equivalent options. The nature of the system largely determines which type of buffer is feasible (theoretically, or in practice). For instance, service systems do not have the possibility to employ an inventory buffer. Legal advice or consulting services are typical examples. Likewise, large time buffers in emergency systems (fire brigades, medical rescue teams) are unacceptable. As a consequence, these systems are known to operate with huge amounts of safety capacity. In this view, a system is defined as “Lean” when the buffers present in the system are restricted to the minimum level necessary in order to support the target performance (Hopp and Spearman, 2008). This is referred to as the Lean level. Consequently, all buffering in excess of that necessary minimum can be considered as obese, in the meaning that there is too much buffering for the desired performance. This state of obesity can manifest itself into too much inventory, too much safety time or overcapacity. An obese system could reach its target performance with either smaller buffers, or a better allocation of buffers. If these excess buffers are systematically reduced while keeping up with the desired system performance, the systems gets leaner. However, excessive reduction of buffers will cause system performance to erode: the target performance will eventually become unachievable. In these situations, we characterize the system as anorectic. Note that the above definition of Lean is not static, as it depends both on the system characteristics and the targeted performance. Consequently, the stronger the stress on the system’s performance measures, the more buffers (minimally) will be necessary: hence, the Lean level will change. The specific Lean level of a system may also differ across different systems. For example, the right amount of safety capacity for a car assembly plant compared to a truck manufacturer is different if they both want to be considered lean for similar performance objectives (e.g. a customer order delivery time of three months). Further, if the performance objectives vary through time (for instance with increased competition), the Lean level will change to reflect the new conditions. Note also that the system’s characteristics vary over time (e.g. the inherent variability of the system may decrease as a consequence of system improvements, product mix changes, etc.). This also impacts the Lean level. Following these arguments, the definition of Lean is dynamic in nature. We will now illustrate these concepts with two basic stochastic models, an M/M/1 queuing model and a model of a production line with limited work-in-process allowance.

14 Rapid Modeling In A Lean Context

167

14.2.1 Lead Time: Safety Time Versus Safety Capacity In this section we consider a system consisting of only one single server, processing a single product type. The system’s queueing behavior can be modeled as an M/M/1 system (see e.g. Anupindi et al, 2006), with an arrival rate μ and a processing rate μ . The desired customer service levvel is defined by S (0<S<1), and Ws refers to the S percentile of the flow time of products through the system. In a MTO system, Ws would be the lead time quote necessary in order to guarantee a delivery service level of S. The performance measures of interest are listed in table 14.2. Table 14.2 System performance Performance measure

Definition

Expression

Utilization Expected Time in the queue (waiting time) Expected Number of units in the queue Expected Time in the system (lead time) Expected Number of units in the system Lead time quote (S percentile) Safety time Safety capacity

ρ Wq Nq W N Ws Wsa f e Csa f e

λ /μ (1/μ )(ρ /(1 − ρ )) ρ 2 /(1 − ρ ) (1/μ )/(1 − ρ ) ρ /(1 − ρ ) (1/μ )/(1 − ρ ) ln[1/(1 − S)] Ws −W = W ∗ (1 − ln[1/(1 − S)]) 1−ρ

Given these performance measures we can quantify the concept of buffer substitution, the dynamic definition of lean and the issue of six sigma. This will be shown in figs. 14.2 and 14.3 respectively. The concept of system buffering is illustrated in fig. 14.2, where the W and Ws (where S equals 95), are shown as a function of the utilization ρ (the arrival rate varies from 0.05 to 0.95 while the service rate equals 1). First the strong non-linear behavior of the lead time as a function of ρ can be observed. As a consequence, the lead time quote, which includes safety time, gets increasingly larger with larger utilizations and with higher percentages of service. Therefore we can conclude that the amount of safety time grows with increasing utilization. It can be clearly seen that a smaller amount of safety time can be reached with a higher level of safety capacity and vice versa. In general, the desired lead time is determined by market conditions. If the company policy is such that a service level S has to be provided, this desired lead time quote needs to coincide with Ws . Given Ws and W for the company, the amount of safety time can be derived. From the relationships in table 14.2, the corresponding utilization equals 1 1 − ln 1−S ρ= (14.1) μ × Ws and the corresponding safety capacity equals

168

Nico J. Vandaele and Inneke Van Nieuwenhuyse

Fig. 14.2 Safety time versus safety capacity

1 1 − ln 1−S Csa f e = 1 − ρ = . μ × Ws

(14.2)

1 1 − ln 1−S Csa f e = μ × (Ws + Wsa f e )

(14.3)

In this way we can derive

where the trade-off can be clearly observed. At this point we can state the quantitative definition of Lean: it is the status where the desired performance is reached with the minimum amount of buffering. For a target service level S and the customer-determined lead time quote Ws , the resulting utilization ρ is given by equation (1). If a system wants to operate at a lower utilization level (or, equivalently, with more safety capacity), this results in a lower lead time quote. As such, we have overcapacity with respect to the performance needed: this is the obese state of the system. On the other hand, utilizations exceeding ρ lead either to lead time quotes that do not meet the market requirements, or to service levels that do not meet the company’s targets. This is the anorectic state of the system. Finally, an extension of the model (G/G/1, see Hopp and Spearman, 2008, for a discussion) can be used to quantify the scoring of six sigma projects. From the equation of WS, we can draw the relationship for various improved cases related to the base case (M/M/1), materialized in different levels of variability of arrival and/or service processes. This typically shifts the quoted lead time curve to the lower-right corner. For an equally defined customer preferred quoted lead time, the quoted lead

14 Rapid Modeling In A Lean Context

169

time Ws can be realized operating at higher utilizations; the same performance with more sales volume, which is inspired either by an expansion based strategy or a rationalization strategy, in case of the same sales volume with less resources. An alternative way to profit from the project improvements is to operate at the same utilization which simply leads to sharper lead time quotes. This embraces an aggressive, competitive strategy. Of course, all combinations of higher utilization and improved performance are alternative paths towards the shifted improvement curves. This is visualized in Fig. 14.3, where WS1, WS2, and WS3 resemble low variability, high variability and medium variability respectively. Typically, as Six Sigma projects attack system variability under a continuous improvement framework, the three curves stand for the subsequent and systematically implemented improvements.

Fig. 14.3 Quantification of Six Sigma, increasing levels of variability

14.2.2 Throughput : Work-In-Process Versus Throughput The production line consists of m identical single-machine stations in series. Each server has exponentially distributed processing times, with processing rate μ . Hence, the system is balanced and the bottleneck rate (BNR) equals μ . The line behaves as a CONWIP system: the work-in-process is held constant and is equal to N units. The performance measures of interest are listed in table 14.3 (Hopp and Spearman, 2008). These relationships may be used to illustrate the trade-off between safety capacity and work-in-process. Figure 14.4 illustrates the relationship between N and T H, for a line consisting of 5 stations with BNR= μ =1 unit per minute. We call this

170

Nico J. Vandaele and Inneke Van Nieuwenhuyse

Table 14.3 System performance Performance measure

Definition

Expression

Expected Time in the queue (lead time) Expected Throughput Safety capacity Utilization of each server

W TH Csa f e u

m/μ + (N − 1) × μ N/(m + N − 1) × μ BNR − T H N/(m + N − 1)

setting the “base case”. The figure shows the strong non-linear, concave behavior of T H in terms of N. Safety capacity Csa f e hence decreases as WIP increases. Given the characteristics of the system (i.e., the processing rate μ of each of the servers), higher T H can be obtained at the price of higher WIP, and lower safety capacity.

Fig. 14.4 TH in terms of N for a 5-station line with BNR = = 1 unit per minute (base case)

The expression for T H in Table 14.3 can be used to quantify the trade-off between safety capacity and WIP for systems targeting a market-determined throughput rate T H. As T H=N/(m+N-1)* μ , a decrease in N without jeopardizing T H can only be obtained by increasing the bottleneck rate μ (and, hence, by increasing the capacity of the line). Consequently, the same throughput level TH can only be obtained with lower WIP at the price of extra safety capacity. This is visualized in Fig.14.5. Assuming a market-determined T H equal to 40 units per hour, the basecase system would require N=8 units, with Csa f e =20 units/hr. The WIP level can be cut by half (N=4 units) without impacting the throughput rate (T H=40 units per hour) when the capacity of the system is increased to BNR=μ = 80 units per hour = 4/3 units per minute, implying a safety capacity Csa f e =40 units/hr.

14 Rapid Modeling In A Lean Context

171

Fig. 14.5 T H and SC in terms of N, for the base case and the increased capacity case

Given the systems characteristics and the market-determined T H, the lean level of WIP and Csa f e can be defined as that combination of WIP and Csa f e that yields T H. A system employing higher WIP levels can be characterized as obese: indeed, as the market determines T H, the inherent capability of the system to achieve throughput higher than T H will not result in additional sales. Conversely, a system employing lower WIP levels is anorectic, as it is incapable of achieving the desired T H. The impact of six sigma projects can be illustrated in an analogous way. It is known from the literature that the reduction of system variability shifts the T H curve to the upper left corner (e.g., Hopp and Spearman, 2008). This is shown in Fig.14.6 (T H curve for reduced variability case). Consequently, six sigma either allows to obtain a target T H with lower WIP (and, hence, lower working capital) or to increase T H for the same level of WIP (if this is desired in view of satisfying additional sales). Obviously, all combinations of lower WIP and improved T H represent alternative paths towards the shifted improvement curves.

14.3 Conclusion In this paper we offered a quantitative approach to underpin Lean operations management. We showed that, based on stochastic models, lean is a dynamic concept,

172

Nico J. Vandaele and Inneke Van Nieuwenhuyse

Fig. 14.6 Quantification of Six Sigma

which depends on of the desired system performance (which tends to be imposed by the market), and the system’s characteristics. The models can be used to illustrate the substitution of different buffer types (safety time, safety inventory and safety capacity) needed in order to reach the target performance. In addition, the system improvements from six sigma projects can be analyzed accordingly. Future research may focus on how these concepts can be extended towards larger and more realistic networks.

References Anupindi R, Deshmukh S, Chopra S, van Mieghem J, Zemel E (2006) Managing Business Process Flows, 2nd edn. Prentice-Hall Hopp W, Spearman M (2008) Factory physics. McGraw-Hill Liker J (2004) The Toyota way: 14 management principles from the world’s greatest manufacturer. McGraw-Hill Rother M, Shook J (1999) Learning to see. Lean Enterprise Institute Vandaele N, De Boeck L (2003) Advanced Resource Planning. Robotics and Computer Integrated Manufacturing 19(1-2):211–218 Vandaele N, Lambrecht M (2002) Planning and Scheduling in an Assemble-toOrder Environment: Spicer-Off-Highway Products Division. In: Song J, Yao D (eds) Supply Chain Structures: Coordination, Information and Optimization, Kluwer Academic Publishers, pp 207–255

14 Rapid Modeling In A Lean Context

173

Vandaele N, Lambrecht M (2003) Reflections on stochastic manufacturing models for planning decisions. In: Shantikumar J, Yao D, w Henk M Zijm (eds) Stochastic Modeling and Optimization of Manufacturing Systems and Supply Chains, Kluwer, pp 53–86 Womack J, Jones D (1997) Lean Thinking: Banish Waste and Create Wealth in Your Corporation. Touchstone, London Womack J, Jones D, Roos D (1990) The Machine that changed the World. Macmillan Publishing Company

Part III

Case Study and Action Research

Chapter 15

The Impact of Lean Management on Business Level Performance and Competitiveness Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Abstract In this paper we describe a model that investigates the impact of lean management on business competitiveness. We hypothesize that business competitiveness depends on organizational competences (including both the static level of operational capability and the dynamic capabilities of improving and adapting to changing internal and external conditions) and business performance. The lean literature provides an unbalanced picture of the elements of business competitiveness: while several researches discuss the impact of lean on static operational measures, there are much less studies about the relationship between lean and 1) organizational changes and responsiveness, and between lean and 2) business performance. In the empirical part of our paper we focus on the latter issues using both case studies and questionnaires. With our case based research (using two original cases and relying on several ECCH cases) we can clearly highlight how lean affects, through employees, organizational responsiveness and how it leads towards higher business competitiveness. Our analysis is unique in the sense that we could relate the casebased analysis to the perspective of employees, since in our original cases several employees (83 and 97) filled in a questionnaire that showed the impact of lean tools Krisztina Demeter Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary, e-mail: [email protected] D´avid Losonci Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary, e-mail: [email protected] Zsolt Matyusz Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary, e-mail: [email protected] Istv´an Jenei Department of Logistics and Supply Chain Management, Corvinus University of Budapest, Fovam ter 8, H-1093 Budapest, Hungary, e-mail: [email protected]

177

178

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

and methods on them, as well as their opinion about the improvements both at operational and business levels.

15.1 Introduction Nowadays, lean management is having its second heyday (Schonberger, 2007; Holweg, 2007). Several companies, many of them outside the automotive industry, implement lean management hoping to achieve competitive advantage. Their hope is fueled by the success of Toyota and other car manufacturers and their suppliers (Liker and Wu, 2000). Large international surveys support that pull, customer induced production systems, the basics of lean management, are inevitably the sources of competitive advantage today (Laugen et al, 2005). Unfortunately these researches did not investigate whether improving competitive resources and better operational performance really affect financial performance. Huson and Nanda (1995) were unable to present a clear link between JIT and profitability, while Voss (1995, 2005) suggests that competitiveness would improve - but that term is not defined in any way. One can also rarely find empirical studies that focus on the organizational change requirements of lean transformation beyond the usual lean tools and principles, though it has been widely accepted for decades now that human resources play an outstanding role in lean transformation, thus lean requires substantial changes in employees and managers’ perspectives and everyday work (Sugimori et al, 1977; Hines et al, 2004). In this paper we discuss the changes triggered by lean (introduction of new tools, methods and principles) and their results through case studies. We emphasize the role of human resources in this process and hence we present not only the top managements opinion about lean management, but also the workers impressions. We begin with a definition of business competitiveness, which sets up the theoretical frame for the research. Then the existing literature about the relationship between lean management and business competitiveness is summarized. After having described the research methodology, the empirical results of our study follow. The final part discusses the results and the limitations of the research.

15.2 The Research Framework 15.2.1 The Building Blocks of Business Competitiveness We can hardly find overall and well developed definitions of business competitiveness in the operations management literature. Therefore we cross the boarder of operations management and start the discussion of business competitiveness based on the definition of Chik´an (2006):

15 Impact of Lean Management

179

“Business competitiveness is a competence of the company that allows the company to provide products and services for customers within the standards of social responsibilities, that are (i) preferred to the products and services of other competitors and (ii) provide profit for the company. The prerequisite of this business competitiveness is that the company is capable of properly evaluating and adapting to the internal and external changes, while achieving sustainable competitive advantage.” (Chik´an 2006, p.46). Thus, in order to achieve business competitiveness, the company has to provide a service package, which is considered by the customer to be better than the competitors (competitive advantage) while being valuable for the company too (i. e. providing profit). Sustainable competitive advantage refers to the fact that the adequate service package should be provided not only in the present but by adapting to the external and internal environment also in the future. The top, broken line part of Fig. 15.1 represents this model, where each leg of business competitiveness provides a measurable “picture” of organizational capabilities. Business competitiveness is affected by capability to operate, capability to change and their market recognition (business performance). The business performance behind sustainable business competitiveness originates from capability to operate, where the adaptation to dynamic environment is guaranteed by the capability to change. This approach, however, does not provide link between the building and results of capabilities. Thus we propose a two level model, where both the results and the building blocks behind can be discovered. The level “under” the results relies on Fig. 15.1, which refer to various methods and levels of combining organizational resources according to defined objectives (Swink and Hegarty, 1998; Hayes and Pisano, 1994; Gelei, 2007). In lean environment a new tool/principle can be treated as a new combination of organizational resources, e. g. introduction of cell as a production unit. Undisputedly, organizational capabilities finally materialize in some kind of performance measure. The realized performance (cost/price, quality, flexibility, reliability and services, see Flynn and Flynn 2004; Li 2000 or the potential capability for this performance (Chik´an, 2006) forms the three measurable legs of business competitiveness (top, broken line part of Fig. 15.1). If we take the previous example again, a cell means shorter lead time and more flexible production processes on the operation performance side. Business competitiveness can be investigated in its true complexity if we rely both on the realized performance and on the resource combination perspectives as complementers. The organization, represented by the three measurable legs, can only adapt to the dynamic environment, retain and improve operational competitiveness if it is improving continuously and combines organizational resources in new ways (e. g. based on lean management principles). The real sources of competitiveness stem from improved individual capabilities, as well as from redesigned and continuously changing working place practices and operating routines.

180

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Fig. 15.1 Components of business competitiveness. Based on Chik´an (2006) and Gelei (2007)

15.2.2 Lean Management and Business Competitiveness Lean management was a hot topic already in the early 80s in operations management (Schonberger, 2007) and flourished in the US (Holweg, 2007) under the name of just-in-time (JIT). In the ’90s lean management became the dominant strategy for organizing production systems (Karlsson and Ahlstr¨om, 1996). Moreover, Hines et al (2004) considered it as the most influential paradigm of operations management. Despite this, the effects of lean management on business competitiveness were anecdotic or case-based, lacking any deeper insight into real managerial issues. From another point of view (Voss and Blackmon 1994, cited by Davies and Kochhar 2002, operating practice in the field of operations management is crucial for operating performance, and operating performance is crucial for operational competitiveness. Following this logic Voss et al (1997) state, without empirical support, that outstanding operating performance leads to outstanding business performance and competitiveness. Adapting this logic to lean management, leaning production improves production performance, and thus production competitiveness, all of which contribute to business performance and competitiveness. Thus lean management, as one of the best practices of world class manufacturing, finally leads to improved business competitiveness (Voss, 1995; 2005). In overall, the relationship between lean and competitiveness is strong intuitively, but real practical evidence in this issue is missing.

15 Impact of Lean Management

181

15.2.2.1 The Impact of Lean Management on Business Competitiveness The Capability to Operate Schmenner (1988) concluded, that “Out of many potential means of improving productivity, only the JIT-related ones were statistically shown to be consistently effective”. Shah and Ward (2007) found a positive relationship between lean management and outstanding operating performance, and added that the relationship is well accepted among researchers and practitioners (see their referred sources e. g., Krafcik (1988); MacDuffie (1995); MacDuffie et al (1996); Shah and Ward (2003); Womack and Jones (1996)). According to the literature, lean practices impact inventory turnover, quality, lead time, labour productivity, space utilization, flexibility (volume and mix) and costs heavily (Crawford et al, 1988; Huson and Nanda, 1995; Flynn et al, 1995; MacDuffie et al, 1996; Karlsson and Ahlstr¨om, 1996; Sakakibara et al, 1997; Boyer, 1998; McKone et al, 2001; Cua et al, 2001). Thus lean practices inevitably impact operating performance dimensions positively, and moreover, concurrent applications of various practices seem to have synergetic effect, they strengthen each other (Crawford et al, 1988; Cua et al, 2001; Flynn et al, 1995; Sakakibara et al, 1997; Boyer, 1998; McKone et al, 2001; Shah and Ward, 2007). To summarize this part, there is far enough empirical support that lean management contributes to business competitiveness through the capability to operate. 15.2.2.2 The Impact of Lean Management on Business Competitiveness The Capability to Change According to Fig. 15.1, the capability to change consists of four areas: (1) market relations, (2) personal skills, (3) decision making and communication and the level of (4) innovativeness. The relationship of lean management with these areas in most of the cases is clear, but less evident. Nonetheless, all of them are important in lean transformations. Just think about the importance of (1) market relationship quality in (i) balancing production load forward and backward in the supply chain, (ii) identifying the customer value (the first principle of lean thinking, Womack and Jones (1996), or (iii) organizing JIT supplies (a basic lean element). The (2) human factor is crucial in lean transformations: “Needless to say, sophisticated technologies and innovative manufacturing practices alone can do very little to enhance operational performance unless the requisite human resource management (HRM) practices are in place” (Ahmad and Schroeder, 2003, p.19). This statement is supported by the fact that human resources (under the name of cross-functional work force) are among the most frequent practices within lean management (Shah and Ward, 2003, 2007). (3) Decision making and communication systems play a central role in todays organization where information flow and knowledge have enormous effect on value creating processes. Latest HR practices such as empowerment and decentralization, which occur in lean as well, shape the structure of these systems. Areas (2) and (3) overlap heavily, they could not be handled isolated from each other. Hence we integrate them in the empirical part of this study. The effect of lean

182

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

management on (4) the innovativeness of the company (e.g. R&D expenditures) still shows contradicting results. Several researchers state that excessive elimination of waste “cripples” innovative ideas and the extent of developments decreases (Lewis, 2000). A counter-example is Toyota, which brought Prius on the market far earlier than its competitors, years before the era of hybrid-driven cars (Liker, 2008). The human factor serves as a basis for two elements of the capability to change, and both seem to determine the success of the lean transformation. Since areas (2) and (3) are relevant from the very beginning, they should be developed parallel with lean tools and principles during the lean transition. In spite of this these elements are rarely in the focus of empirical works, especially from shop floor workers point of view. The Human Factor in Lean Management Emphasizing people is a must in lean management: it comes from the logic of its operations. Since process dependence increases with the elimination of the buffers, production problems arise. Thus the demand for motivated and adjustable work force is obvious (Sugimori et al, 1977; MacDuffie, 1995). Shah and Ward (2007) reached the same conclusion: employees working in cross-functional, self managed teams are faster and more efficient in solving identified problems. MacDuffie (1995) and Shah and Ward (2007) proved that companies relying on HR practices as an integral part of their lean production system can achieve higher results (this is also supported by Wood (1999)). Interestingly, HR literature does not discuss the issue of lean (Wood, 1999) with the exception of Birdi et al (2008). If we consider the effect of lean on employees, there is are two opposing views (Delbridge et al, 2000). Supporters, mostly researchers of OM, see the positive effects of lean management on employees (Legge, 2005). Others (Berggren, 1993; Landsbergis et al, 1999; Lowe, 1993; Skorstad, 1994; Wood, 1999) emphasize the “dark side” of lean (e. g. work intensity, reduced autonomy, overtime, increased horizontal load etc.) Our paper brings some new thought into this discussion by analyzing lean achievements (using business competitiveness as a framework) and HR related changes through the eyes of employees. Based on a comprehensive literature review (Sugimori et al, 1977; Crawford et al, 1988; Flynn et al, 1995; MacDuffie, 1995; Sakakibara et al, 1997; Boyer, 1998; McLachlin, 1997; Cua et al, 2001; Hines et al, 2004; Shah and Ward, 2007) we summarize the most important HRM practices in relation to lean management: • • • •

Education and training, cross-functional work force; Decentralization and empowerment; Team work; Information flow and feedback.

The elements above overlap with the most important HRM practices of the dominant HRM model, which considers people as valuable assets (see Legge 2005 and Pfeffer 1998).

15 Impact of Lean Management

183

15.2.2.3 The Impact of Lean Management on Business Competitiveness Business Performance Surprisingly, in spite of the popularity of lean management in the last decades, operations management still not supported empirically the relationship between lean management and business performance. Impacts on operating performance are obvious, but few studies tie operating performance to financial performance. The work of Huson and Nanda (1995) is exceptional in this respect, even if their conclusion is that the real impact of just-in-time on profitability is ambiguous. Lewis (2000), in his casebased research, argues that “Becoming lean does not automatically result in improved financial performance”, since “[t]he benefits of lean production can very easily flow to powerful players” (Lewis, 2000, p. 975). Nonetheless, the investigation of the effect of lean management on financial performance is very important. Besides proving the direction of the relationship, another goal is to uncover those elements that influence the quality of the relationship. Based on the literature review (which reflects the view of the top management) Table 21.1 summarizes the questions investigated in this paper. First we analyze the three elements of competitiveness with the help of case studies that reflect top management opinion ((1)-(3)). After the operational results of our companies (I) we discuss the ability of responsiveness that supports internal and external adaptation (II), where we give special attention to HR related changes because of their important role in lean implementation and sustainability. Finally we investigate the relationship between lean and financial performance (III). Throughout the questionnaire analysis we assume that good results may motivate employees. The real motivating factor, though, is the feeling of being part of the changes (V), not just the experience of the good results ((IV) and (VI)).

15.3 Methodology 15.3.1 Case Studies and Survey Strengthening the role of qualitative research (case study research included) is a long-time need in the field of OM (Meredith et al, 1989). Until recently many excellent OM researchers have written some kind of ”teaching note” for case study research. (e.g. McCutcheon and Meredith, 1993; Meredith, 1998; Stuart et al, 2002; Voss et al, 2002), nowadays it is one of the standard empirical research methods (Gupta et al, 2006). Though in the past few years there were published only a handful of lean case studies. McLachlin (1997) used a methodology based on Yin (1989) to evaluate management initiatives necessary to implement JIT. Lewis (2000) assesses the relationship between lean production and competitive advantage by using case studies. In our paper we used two case studies (namely the cases of R´aba Automotive Components Manufacturing Ltd. and OKIN) to analyze the links between

184

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Table 15.1 Questions investigated in this paper Case studies

Lean literature on business competitiveness

Surveys

Do our case companies I. Capability to operate IV. Employees regard lean lean efforts resulted (3.2.1.1) Well documented, the (3.2.2.1.) as successful as in the expected positive effects of managers do? improvements? lean management on operative measures are confirmed What kind of changes II. Capability to change V. What do employees were caused by the (3.2.1.2.) All elements are (3.2.2.2.) perceive from leanapplied lean tools or essential in lean, in oriented HR practiprinciples in the elelean transformation ces implemented in ments of capabilities personal skills and the case companies? to change? What kind decision making and of infrastrural changes communication seems changes supported lean? to be crutial What is the effect of lean on financial performance? Which factors influence this relationship?

III. Business performance VI. (3.2.1.3) Anecdotal evidences, (3.2.1.3.) few empirical evidences

Do employees see a relationship between lean and profitability?

lean and competitiveness. The case studies were based on interviews (with middle and top management), company visits, top management questionnaires and company documents. Primary company selection criteria were an open attitude and a solid determination towards lean. In Table 21.2 we summarized the lean management tools applied by our two case companies, which indicate the selected companies fulfillment of the selection criteria. The surveying method is a common tool for measuring management attitudes, while it is very uncommon to conduct employee surveys on a large sample in the field of lean, hence our approach is somewhat unique. The employee questionnaire consisted of 51 questions, and it was inspired by a previous survey (Tracy, 2004). The questionnaire intended to grab the anticipations of the companies, the goals and implementation of lean transformation, the transformational effects, results and changes on working conditions, tools, applied technology and intra-firm communication. 15.3.1.1 R´aba Automotive Components Manufacturing Ltd. R´aba Automotive Components Manufacturing Ltd. is one of the three divisions of R´aba Automotive Holding Plc. The case was written about the plant near the city of M´or (called R´aba in this paper) with turnover around 30 Million Euro and 589 employees. It is the main plant for producing seat accessory parts (seat frames, car

15 Impact of Lean Management

185

Table 15.2 Lean practices in case companies Workplace practices, operating routines

R´aba Okin

Buyer, supplier relations Supplier base within the plant Kanban based raw material supply Direct contact with material suppliers, incoming goods quality check Intensified communication with customers Just-in-time order fulfillment Product development and technology New equipment Low cost automation, transformation of on hand equipment Organization and control of manufacturing processes Rearrangement of production lines and storage (value stream mapping) Unified material-, info.-flow and organiz. (warehouse, production, dispatch) Work environment (5S) Pull material flow, Just-in-time, kanban One piece flow Production cells Clear definition of tasks and responsibilities Redesign of process control system and practices Redesign of inventory management Implementation of intra-process control (received materials, WIP) Kaizen workshop / incremental improvement Visual management (e. g. whiteboard) Reduced changeover time Human resource related changes Lean coordinator (system engineer) Team work Training for the management (lean knowledge) Training for middle managers and workers New (partially performance based compensation system) motivation system Internal (open) forum (for better communication) Refurbishment of work environment “Empowerment”, quality responsibility of workers

foams and seat trims) for Suzuki, a large OEM in Hungary. Moving towards lean management was essential in order to survive after some years of unprofitable operations and downsizing. Altogether 83 employees filled in the questionnaire at R´aba (62% of all employees in the working places affected by lean transformation).

186

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

15.3.1.2 OKIN Hungary Kft. OKIN Hungary Ltd. has new owners, a German investment group since 2007. It has around 300 employees with a site in north-east Hungary, next to the city of Hajd´udorog. It assembles furniture motion mechanics in large variety. Product design, orders, deadlines and customer relationship management is executed in the German headquarter. Lean transformation was forced by the owners, because the Central European wage advantages eroded strongly after new capacities have been created lately in Far Eastern countries. The employees of four assembly lines filled in the questionnaire at the company. Practically each employee who was present at the time wrote his/her opinion (93 people).

15.3.2 Discussion The Impact of Lean Management on Business Competitiveness 15.3.2.1 Case Studies The Capability to Operate Following the lean transformation R´aba experienced improvements in dimensions of costs, productivity, delivery lead time, delivery dependability, inventory turnover, inventory level, space utilization, labor productivity and volume flexibility. In some dimensions the improvement was huge. Connecting services were not affected by lean. Okin achieved great improvements in productivity (shorter setup times), manufacturing lead time, quality and inventory turnover, thanks to the lean transformation and the better organizing of processes. By studying the results of lean companies the improvement of elements of operational performance (costs, productivity, quality, manufacturing and delivery lead times, volume and time flexibility) seems evident this is in line with the expectations. These changes suggest success, and they reflect the proper application of tools used for reorganizing the manufacturing process. The Capability to Change We can conclude from the R´aba and Okin cases that the improving measures were caused by both the reorganization of manufacturing and the changes made in the supporting infrastructure. The companies made the most changes in the manufacturing processes, in their control and in the (2) area of human resources (see Table 21.2). It is worth mentioning that from the other dimensions of responsiveness the area of (3) decision making and communication systems (which is closely connected to people) changed strongly, while the area of (1) market relations and (4) innovativeness did not change or only to small extent. The lack of innovativeness could be explained by the fact that it is not required by the companiescustomers. The sole

15 Impact of Lean Management

187

obligation of the companies is to meet the customer specifications. Relationship with the customer at R´aba was strong before the lean transformation too (justinsequence supply), and in the case of Okin it was a logical step forward to form tighter cooperation with the customers. Lean was crucial in creating more intensive communication and relationship on the input side too. Workers at both companies learned lean basics to the necessary extent for their job. Workers were not given further education, on the job training and/or their previous knowledge were enough for the new tasks. There was also a lean manager appointed who coordinated the whole process. At Okin the lean transformation remained basically centralized and was built around the lean manager. Manufacturing decisions, which influenced areas of (2) human resources and (3) decision making and communication, were delegated to the lowest level. With the participation of the workers a new flow (from bottom to up) appeared in the communication system (kaizen workshops, work process design) beginning a transition towards modern management systems. Our research highlights the fact that lean does not intend to improve competitiveness through innovativeness (technological advance, R&D) this may explain the mild interest for this topic from researchers. Lean affects responsiveness through (2) personal skills and the closely connected (3) decision making and communication elements. Our results suggest that human resource management (motivation, education, communication, decision making) appears as one of the most critical factors in lean. This topic should get a much bigger interest in itself, not just as an integrated part of the manufacturing system. Business Performance During the years following lean implementation the output and sales revenue increased at both companies (1), but it would be a mistake to consider lean as the only factor behind this phenomenon. Lean should be regarded as a consequence of growing as well as a cause of it. At R´aba lean implementation as a way of efficiencyseeking was forced by the capacity reserved by the customer (anticipated growth) and the unprofitable business, while the improving performance created further utilizable capacity. At Okin the main causes were also capacity issues and more profitable prices. In both cases the companies were able to improve efficiency indicators ((3) and (6)) in a way that no great investments were needed. This is further strengthened by the increased inventory turnover (though Okin has a very special inventory policy). The companies business performance (ROS (4) and operating profit (3)) do not improve automatically with the implementation of lean. Business performance is affected by several other factors, e. g. industry, market position (R´aba: second-tier supplier; Okin: mother company), competition intensity (R´aba: increasing presence of competitors and OEMs in the Central European region; Okin: Far Eastern manufacturing sites), power (R´aba: OEM as customer; Okin: mother company transfer prices), product characteristics (complex, substitution, product range), product development capability etc.

188

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Table 15.3 R´aba’s business performance (2004-2007)1 Source: Company data 2004

2005

2006

2007

(1) Sales (thousand Ft) 9 630 042 10 032 572 12 544 893 16 768 960 (2) Total expenses (thousand Ft) -10 390 086 -10 258 467 -12 135 328 -16 073 747 (3) Operating profit (thousand Ft) {(1)-(2)} -760 044 -225 895 409 565 695 213 (4) Return on sales (3)/(1) 3.26% 4.15 (5) Infr., plant and equip. (thousand Ft) 2 924 334 2 401 493 2 401 493 2 603 025 (6) Sales on tangible assets (Ft) {(1)/(5)} 3.293 4.178 4.178 6.442 (7) Inventories (thousand Ft) 1 143 899 1 028 104 1 028 104 1 098 665 (8) Inventory turnover {(2)/(7)} 9.083 9.978 9.978 14.630 1

The column in bold refers to lean implementation year.

Table 15.4 Okins business performance (2004-2007). Source: Company data 2004 (1) Sales (thousand Ft) (2) Total expenses (thousand Ft) (3) Operating profit (thousand Ft) {(1)-(2)} (4) Return on sales {(3)/(1)} (5) Infr., plant and equip. (thousand Ft) (6) Sales on tangible assets (Ft) {(1)/(5)} (7) Inventories (thousand Ft) (8) Inventory turnover {(2)/(7)}

857 931 -870 166 -12 235 370 743 2.314 5 087 171.057

2005

2006

2007

929 775 1 443 278 1 557 893 -995 340 -1 505 677 -1 530 321 -65 565 -62 399 27 572 1.77% 370 176 379 185 424 123 2.512 3.806 3.673 11 453 12 275 10 576 86.906 122.662 14.630

In Fig. 15.2 we summarized the findings of the case studies. Similarly to the results in Table 21.1, we concluded that lean had an obvious positive effect on the capability to operate. The companies show great improvement in their operational performance following the lean transformation. Measures of capability to change are also better, which reflect better market relationships, more skilled workforce, more advanced decision making systems and more intensive communication. The capabilities to operate and change together suggest the improvement of business competitiveness. But did the market accept this improving performance? Business indicators became better at both companies, to a greater extent at R´aba and to a slighter extent at Okin. According to the case studies the relationship between lean and business performance is not so strong as between lean and the capabilities to operate and change, because here we have to deal with several other influencing contextual factors. 15.3.2.2 Survey Our case studies in accordance with literature show the “beneficial” influence of lean on organizational dimensions. This statement is based on managerial interviews, managerial surveys and company data. Concerning the employees, there are

15 Impact of Lean Management

189

Fig. 15.2 Lean management and business competitiveness

very few (empirical) research about them and those as we said earlier focus mostly on the working conditions of lean. We do not know anything about whether they perceive the “proved” success during their daily work. As the knowing of success could be inspiring, therefore their opinion and experience may facilitate the management of lean. The opinion of the employees, who operate the whole system, hence could give significant insights for the better understanding of the system. During the analysis of the employee surveys we kept the three “legs” of competitiveness, but with a slight modification: inside the capability to change we investigated only the critical HR practices identified by previous lean literature (see Table 21.1). The Capability to Operate Questions in Table 15.5 are about the dimensions of capability to operate. The employees could mark those statements with which they agreed. From the previously given performance measures improvement in productivity was the most chosen one at both companies (83% and 50%), though the frequencies differ significantly. The distribution of employee answers indicates that the lead time/cycle time, the scrap ratio and quality became significantly better at both companies. As for the remaining measures (inventories, costs, process stability), R´aba performed better. There is also another difference between the companies in the respondent rate, which was higher at R´aba. This can be partly explained by the fact, that the changes made at R´aba were more deep and better communicated. It is

190

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Table 15.5 Capability to operate - shop floor workers’ perceptions Performance (might more answers!)

Respondant rate (%) R´aba Okin Pearson Chi-Square N = 83 N = 93 (statistical significance)

After lean transformation... ...decreased in our company. Lead time/cycle time Inventories Scrap Costs All answers

49 26 48 44 167

34 13 44 13 104

3.772 (0.037) 4.498 (0.027) 0.237 (0.371) 19.639 (0.000)

After lean transformation... ...improved in our company. Productivity Process Quality All answers

83 45 38 166

50 15 48 113

18.502 (0.000) 17.326 (0.000) 1.633 (0.131)

nonetheless strange that despite the radical changes only 40-50% of the employees perceived some improvement. The operational results of lean appear at shop floor level, though to much less extent than among the top managers or what could be expected after a radical improvement. The difference between employee perceptions and reality can originate from many sources, e. g. internal communication, focal points of employee reward system or employee ignorance. It is worth thinking over how can the achieved results help in the acceptance and sustainability of lean. The Capability to Change In this part we are concentrating on the presence and effects of critical lean HR practices identified in part 2.2.2.1 previously. We investigate only the (2) personal skills and (3) decision making and communication parts of the capability to change. Results in Table 15.6 suggest that there are significant differences between some of the HR practices and the whole transformation was deeper at R´aba. Training. According to the literature, training is one of the key points, though the employees of the companies evaluated it only as “I slightly agree”. Employees perceived some improvement compared with the previous routines, but only to a small extent. The most plausible explanation for this is that the employees only got the necessary knowledge about lean basics, but beyond this there was no further education (see below for more details). Cross-functional workforce. Workers perceptions reflect higher horizontal workload and more supplement activities to do. The latter originates in the quality approach of lean management: the worker is responsible for his/her work environment, since it affects product/service quality. Supplement activities, namely 5S and smaller maintenance tasks (introduced by both firms) do

15 Impact of Lean Management

191

Table 15.6 Capability to change shop floor workers about critical HR practices in lean2 HR practices (R´aba, Okin) Training Learning is essential at my company. (78, 91) Employees were or will be given some form of training on how to use the technology/tools that is required to implement lean (81, 86) Cross-functional workforce Since lean I have to know more kind of operations. (82, 89) Since lean I have to do more supplement activities (82, 91) Empowerment and decentralization For decisions concerning my work my opinion is also taken into account. (83, 90) Since lean I have to do more supplement activities (82, 91) I have the opportunity to improve processes. (80, 88) My boss allows me to be creative. (80, 89) Lean innovation mistakes are tolerated. (80, 90)

Average R´aba

Average Okin

F (Statistical significance)

3.88 (1.562) 3.21 (1.410) 8.729 (0.004) 2.88 (1.308) 3.13 (1.532) 1.293 (0.257)

2.30 (1.274) 2.63 (1.265) 2.788 (0.097) 2.55 (1.441) 2.78 (1.237) 1.291 (0.257)

2.30 (1.274) 2.63 (1.265) 2.788 (0.097) 3.22 (1.490) 2.93 (1.356) 1.717 (0.192) 3.00 (1.322) 3.02 (1.339) 0.012 (0.912) 3.31 (1.365) 3.04 (1.269) 1.743 (0.189) 3.39 (1.355) 3.22 (1.356) 0.630 (0.428)

Team work Within my organization, management and employees 1.65 (0.706) 2.00 (0.789) 0.069 (0.792) work together to solve problems. (83, 89) My coworkers supported/support me in lean 2.73 (1.043) 3.11 (1.352) 4.277 (0.040) implementation (80, 89) Teams were or will be developed to implement 1.83 (0.792) 2.91 (1.30) 41.405 (0.000) lean. (80, 87) Communication I understand why lean is/was implemented. (76, 90) I got the necessary knowledge about the essence and background of lean transformation. (81, 90) Before lean implementation my manager clarified my tasks. (81, 88) During lean implementation my manager clarified my tasks. (80, 90) I was told the reasons to implement lean. (81, 88) My managers told me when and how lean would be implemented. (81, 90) I was informed about the results of lean. (77, 79)

2.12 (0.923) 2.64 (1.164) 10.131 (0.002) 2.69 (1.281) 3.28 (1.529) 7.301 (0.008) 2.54 (1.275) 3.26 (1.410) 11.986 (0.001) 2.53 (1.253) 2.91 (1.295) 3.882 (0.050) 2.10 (1.020) 2.97 (1.504) 18.908 (0.000) 2.09 (0.883) 2.91 (1.403) 20.616 (0.000) 2.10 (0.867) 3.35 (1.396) 44.880 (0.000)

2

All questions were asked on a 1 to 6 scale, where 1 means total agreement, 2 stands for agreement, 3 is for slight agreement, 4 is for slight disagreement, 5 stands for disagreement and 6 means total disagreement.

not require professional training, only more basic lean knowledge. So seeing the slightly “positive” rating of learning in Table 15.6. we state that the companies ef-

192

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

forts in job enrichment (more manufacturing operations) build much more on the effective exploitation of existing worker knowledge and proper organized processes. In other words: (professional) training supports mostly technological changes. In lean (transformation) one should consider and not waste (not exploit) professional worker knowledge, and use it as a valuable resource (as our case companies did). In addition basic lean training is an unavoidable premise, just as the deepening of the training at all hierarchical levels in order to sustain lean later on. One practical consequence of this is that managers in lean environment should pay more attention to the learning phase of newly employed people. Empowerment and decentralization. The firms rely on employees active participation in improvement activities. The shop floor workers can be either “advisors”, or “makers”, i. e. they have the opportunity to build their own ideas into the reorganized processes. Emphasizing trial and error approach of improvement activity (as perceived in our cases) provides an innovative environment for lean efforts. Although the basis for worker participation is good, as they have the opportunity to take part in projects in a tolerant atmosphere, opinions in Table 15.7. suggest that there is place for further improvement. Only about 40-50% of shop floor employees feel involved. According to operators perception middle managers have the most important role in disseminating and applying lean. This organizational level connects top managers lean commitment with daily professional practice. Table 15.7 Involvement in lean implementation shop floor workers’ perception Involvement (more answers are possible!) The following person or people were or will be involved in implementing lean Owner Top management Middle management Employees Suppliers

Respondant rate (%) R´aba Okin Pearson Chi-Square N = 81 N = 83 (statistical significance)

2.5 2.5 75 52 1.2

14.5 14.5 41 40 2.4

7.546 (0.005) 0.078 (0.453) 0.078 (0.453) 2.416 (0.081) 0.315 (0.509)

Team work. In lean, the unit of work organization is a team, especially in the case of R´aba. In our companies the foremen/managers have the leading position in team work (problem solving). This refers to the fact that they manage problem solving activities, coordinate and frame “leaning”, and dominate the radical changes of the implementation phase. These findings support the previous paragraph: active participation of top and middle management is a substantial success factor in leaning the shop floor. Communication. Comparing the two firms this element shows the most remarkable difference, with R´aba having an overall advantage in all dimensions. The gap (at least partially) can be explained by the depth of the changes and at the same time one should be aware that the “amplitude” can and should affect communication strategy. The offensive communication deserves extra attention on

15 Impact of Lean Management

193

each stages of the implementation (before, during, and after): informing workers about backgrounds and reasons, tasks and responsibility and evidently about results. The top-down flow of data and information can be tightly connected to active management role, which is an additional signal of their central role. The averages on result-feedback (bottom line in Table 15.6) display and confirm our explanation on differences in capability to operate measurements Table (15.5): employees in R´aba are better informed. The commanding value of the automotive supplier in this question might be misleading, since operators perceptions, as we argued earlier, lag significantly behind the real figures. To summarize, HR practices related to lean management are of high priority in successful lean companies. According to shop floor perceptions their actual deployment depends on the depth of the transition and lean commitment of managers. Our case companies suggest that workers, thanks to the fact that lean exploits operators knowledge more effectively, are responsible for more “core” and supplement activities. Shop floor employees can actively participate in process improvements. Anyway, the problem-solving teamwork is mainly dominated by managers. Communication has considerable role from the first steps, employees are informed along the lean journey. Business Performance Although the distinction is more outstanding, the pattern of answers resembles the operative Figs. (Table 15.8). Almost unanimously, R´aba workers tie lean and profitability, and this relationship seems to be much stronger than the one with manufacturing performance measurement. In the case of Okin workers did not perceive improved profitability. One powerful rationale is the depth of changes, another that the recent history of R´aba is marked by losses and lay-offs, meanwhile Okin operations covered its costs and grew constantly. The case studies indicate that workers are aware of the effect of lean on company performance (both operational and business), but they underestimate this effect. It is worth thinking over whether better feedback could improve employee commitment and satisfaction. Table 15.8 Business performance shop floor workers’ perceptions Involvement (more answers are possible!) Lean affected the companys profitability positively.

Respondant rate (%) R´aba Okin Pearson Chi-Square N = 83 N = 93 (statistical significance) 62

9

51.539 (0.000)

194

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

15.4 Conclusion The findings presented in this paper illustrate that becoming lean can contribute to business competitiveness. (1) Our case studies confirm the previously ”proven” positive relationship between lean management and operative performance measures. The data pointed out that lean implementation can enormously improve operative performance in the early stages of the transition. (And behind the measurable results can also have some hardly measurable effects: stability, order and cleanness.) This positive relationship is obvious company-wide: not only (top) managers know about it, but the majority of shop floor workers as well. Although in the latter case the perception of the extent of improvements lags far behind the real outcomes. The more active and effective communication of the operative success (if employees can influence them directly) could support the acceptance of lean and enhance workers lean commitment. (2) Human resource occurs in lean literature as a central element in the transformation process in spite of this it is rarely in the focus of empirical works. In accordance with the academic society we found that beside the production-related tools our case companies apply HR- related practices most frequently. Beyond (i) the growing importance of the middle management level, this new HR approach appears (ii) in training, (iii) in more effective exploitation of existing expertise (covering problem solving, improvement activities, and job enrichment), (iv) in team work and (v) in intensified communication on the shop floor level. It may be a challenging issue for an OM professional: lean is not only about operations, because people require at least as much attention as production and service processes. (3) Although lean is a “fashionable” management and academic “fad”, surprisingly there is not almost any research about its possible financial impact. According to financial data case companies started lean in a period of “motivating crisis”: hallmarked by operating losses and anticipated growing demand. Our case data refer to positive relationship between lean and business measures. Cases also highlight that several factors (e. g. competition in the industry, product characteristic, and customer power) can affect business performance, and any of these might be stronger than the potential financial outcome of improved operational performance. Workers perception is mainly shaped by communication and the earlier performance of the company. Limitations In case of researches like this one, the question of validity and reliability is crucial. We used two means to enhance them: (i) the work is based on managers and shop floor workers’points of view; (2) we combined qualitative (interviews, company documents, visits) and quantitative (surveys, company documents) sources during data gathering and explanation phases. In spite of researchers endeavor, the paper, with special regard to its findings, should be handled carefully, since companies differ in size, operate in different industries and business environment, and go along their own lean “path”. The number of case companies together with the methodol-

15 Impact of Lean Management

195

ogy used to analyze them clearly means a limitation. So the research can not conclude with general statements. However, we believe that the chosen research framework served our research objective: the adaptation of business competitiveness on lean companies.

References Ahmad S, Schroeder RG (2003) The impact of human resource management practices on operational performance: recognizing country and industry differences. Journal of Operations Management 21(1):19 – 43 Berggren C (1993) Lean production The End of History. Work, Employment & Society 7(2):163–188 Birdi K, Clegg C, Patterson M, Robinson A, Stride CB, Wall TD, Wood SJ (2008) The impact of human resource and operational management practices on company productivity: A longitudinal study. Personnel Psychology 61(3):467 – 501 Boyer KK (1998) Longitudinal linkages between intended and realized operations strategies. International Journal of Operations and Production Management 18(4):356–373 Chik´an A (2006) A v´allalati versenyk´epess´eg m´er´ese. Egy versenyk´epess´egi index e´ s alkalmaz´asa. P´enz¨ugyi Szemle 51(1):42–56 Crawford KM, Blackstone JH, Cox JMJ (1988) A study of JIT implementation and operating problems. International Journal of Production Research 26(9):1561– 1568 Cua KO, McKone KE, Schroeder RG (2001) Relationships between implementation of TQM, JIT, and TPM and manufacturing performance. Journal of Operations Management 19:675–694 Davies AJ, Kochhar AK (2002) Manufacturing best practice and performance studies - a critique. International Journal of Operations & Production Management 22(3):289–305 Delbridge R, Lowe J, Oliver N (2000) Shopfloor responsibilities under lean teamworking. Human Relations 53(11):1459–1479 Flynn BB, Flynn EJ (2004) An exploratory study of the nature of cumulative capabilities. Journal of Operations Management 22:439–457 Flynn BB, Sakakibara S, Schroeder RG (1995) Relationship between JIT and TQMpractices and performance. Academy of Management Journal 38(5):1325–1360 Gelei A (2007) Besz´all´ıt´o-t´ıpusok e´ s azok alapveto kompetenci´ai a hazai aut´oipari ell´at´asi l´ancban. PhD thesis, Budapesti Corvinus Egyetem Gupta S, Verma R, Victorino L (2006) Empirical research published in production and operations management (1992–2005): Trends and future research directions. Production and Operations Management 15(3):432–448 Hayes RH, Pisano GP (1994) Beyond World-Class: The New Manufacturing Strategy. (cover story). Harvard Business Review 72(1):77 – 87

196

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Hines P, Holweg M, Rich N (2004) Learning to evolve - A review of contemporary lean thinking. International Journal of Operations & Production Management 24(10):994–1011 Holweg M (2007) The genealogy of lean production. Journal of Operations Management 25:420–437 Huson M, Nanda D (1995) The impact of just-in-time manufacturing on firm performance in the US. Journal of Operations Management 12(3-4):297 – 310 Karlsson C, Ahlstr¨om P (1996) Assessing changes towards lean production. International Journal of Operations and Production Management 16(2):24–41 Krafcik JF (1988) Triumph of the lean production system. Sloan Management Review 30(1):41–52 Landsbergis PA, Cahill J, Schnall P (1999) The Impact of Lean Production and Related New Systems of Work Organization on Worker Health. Journal of Occupational Health Psychology 4(2):108 – 130 Laugen BT, Acur N, Boer H, Frick J (2005) Best manufacturing practices: What do the best-performing companies do? International Journal of Operations & Production Management 25(2):131 – 150 Legge K (2005) Human resource management: Rhetorics and Realities. Anniversary edition edn, Palgrave Macmillan, chap Human resource management, pp 220– 241 Lewis MA (2000) Lean production and sustainable competitive advantage. International Journal of Operations & Production Management 20(8):959 –979 Li L (2000) Manufacturing capability development in a changing business environment. Industrial Management and Data Systems 100(6):261–70 Liker J (2008) The Toyota Way: 14 Management Principles from the Worlds Greatest Manufacturer. McGraw-Hill Professional, USA Liker JK, Wu YC (2000) Japanese automakers, US suppliers and supply chain superiority. Sloan Management Review 42(1):79–94 Lowe J (1993) Manufacturing reform and the changing role of the production supervisor: The case of the automobile industry. Journal of Management Studies 30(5):739 – 758 MacDuffie J (1995) Human resource bundles and manufacturing performance: Organizational logic and flexible production systems in the world auto industry. Industrial and labor relations review 48(2):197–221 MacDuffie JP, Sethuraman K, Fisher ML (1996) Product variety and manufacturing performance: evidence from the international automotive assembly plant study. Management Science 42(3):350–369 McCutcheon DM, Meredith JR (1993) Conducting case study research in operations management. Journal of Operations Management 11(3):239–256 McKone KE, Schroeder RG, Cua KO (2001) The impact of total productive maintenance practices on manufacturing performance. Journal of Operations Management 19(1):39–58 McLachlin R (1997) Management initiatives and just-in-time manufacturing. Journal of Operations Management 15(4):271–292

15 Impact of Lean Management

197

Meredith J (1998) Building operations management theory through case and field research. Journal of Operations Management 18(4):441–454 Meredith JR, Raturi A, Amoako-Gyampah K, Kaplan B (1989) Alternative research paradigms in operations. Journal of Operations Management 8(4):297–326 Pfeffer J (1998) The Human Equation: Building profits by putting people first. Harvard Business School Press, USA Sakakibara S, Flynn BB, Schroeder RG, Morris WT (1997) The impact of just-intime manufacturing and its infrastructure on manufacturing performance. Management Science 43(9):1246–1257 Schmenner R (1988) Behind labor productivity gains in the factory. Journal of Manufacturing & Operations Management 1(4):323–38 Schonberger RJ (2007) Japanese production management. Journal of Operations Management 25:403–419 Shah R, Ward PT (2003) Lean manufacturing: context, practice bundles, and performance. Journal of Operations Management 21(2):129–149 Shah R, Ward PT (2007) Defining and developing measures of lean production. Journal of Operations Management 25(4):785–805 Skorstad E (1994) Lean production, conditions of work and worker commitment. Economic and Industrial Democracy 15(3):429 Stuart I, McCutcheon D, Handfield R, McLachlin R, Samson D (2002) Effective case research in operations management: A process perspective. Journal of Operations Management 20:419–433 Sugimori Y, Kusunoki K, Cho F, Uchikawa S (1977) Toyota production system and kanban system. Materialization of just-in-time and respect-for-human system. International Journal of Production Research 15(6):553–564 Swink M, Hegarty WH (1998) Core manufacturing capabilities and their links to product differentiation. International Journal of Operations and Production Management 18(4):374–396 Tracy M (2004) Lean Transformation Questionnaire. URL http://www.oakland.edu/lean/download/LeanMfgSurvey forEmployees.pdf#search=%22lean questionnaire%22 Voss C, Blackmon K (1994) Total Quality Management and ISO 9000: A European Study. Centre for Operations Management, London Business School, London. Voss C, Ahlstrm P, Blackmon K (1997) Benchmarking and operational performance: Some empirical research. Quality Management & Technology 4(4):273– 285 Voss C, Tsikriktsis N, Frohlich M (2002) Case research in operations management. International Journal of Operations and Production Management 22(2):195–219 Voss CA (1995) Alternative paradigms for manufacturing strategy. International Journal of Operations & Production Management 15(4):5 – 16 Voss CA (2005) Alternative paradigms for manufacturing strategy. International Journal of Operations and Production Management 25(12):1211–1222 Womack JP, Jones DT (1996) Lean thinking: Banish waste and create wealth in your corporation. Simon&Schuster UK Ltd

198

Krisztina Demeter, D´avid Losonci, Zsolt Matyusz and Istv´an Jenei

Wood S (1999) Human resource management and performance. International Journal of Management Reviews 1(4):367 –414 Yin R (1989) Case study research: design and methods. Sage Publications, Newbury Park, CA, USA

Chapter 16

Reducing Service Process Lead-Time Through Inter-Organisational Process Coordination Henri Karppinen and Janne Huiskonen

Abstract The management of public sector service operations has gained much attention in the scientific literature during the last fifteen years. As in the industrial world, also in the service world, different types of processes exist, requiring different kind of tools and improvement actions. A group of challenging service processes are the so called ‘fluid service processes’ that are considered as uncontrollable, peopledominated, diagnosis-focused, and the traditional process improvement tools do not seem to work in them. The focus of the study is inter-organisational cooperation in fluid service process delivery. The specific focus of the paper is in understanding and reducing the service process lead-time. The study is based on three interconnected action research projects conducted in a Finnish municipality. The results of the study show that the traditional process development approach is not enough when trying to solve process-related problems in the inter-organisational context, such as the lead-time of a service process.

16.1 Introduction Service process development has become an important topic both in private and public sector services. In the current financial turbulence, making the service operations even more efficient while at the same time emphasizing customer-orientation is a challenging approach for every organisation. It is, however, an inspiring starting point for research in the service sector, which has certain traditions but is still Henri Karppinen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN53851 Lappeenranta, Finland, tel: +358-5-621 2649, fax: +358-5-621 2699, e-mail: [email protected] Janne Huiskonen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN53851 Lappeenranta, Finland

199

200

Henri Karppinen and Janne Huiskonen

in a developing stage. The interest in this study is a specific area of services: the interorganisational service context, where many organisations from the private and public sector, with diversified objectives and motivation to participate in cooperation, work together in order to produce a service for the customer. In the literature the topics of inter-organisational cooperation and relationships as well as a more operational view the service process management are widely discussed and analysed, but unfortunately separately. Service process management is also a developing area of research and theory. Our intention is to emphasize that the research of the service sector should be about seeing service as it is, and only after that trying to form the needed approaches and tools, not trying to fit the services into existing models based on the industrial world. The research work described in this study work was performed in a Finnish Municipality in three different service sector processes.

16.2 Research Objectives and Literature Review We define the research gap and the target of the study on the basis of the setting explained in the introduction. The research gap is very practical world-oriented, and in order to reach our objective we have selected the action research as the methodology, and our aim is to maintain practical relevancy while trying to influence theory creation. The starting point of our study was a prolonged lead-time problem in three service processes. Long lead-time causes problems in terms of costs, but also when measuring the service quality and the service delivered for the customer. The customer has a significant role in all three processes, and all of them are very labour intensive. In the case processes, the biggest challenge is having many different organisations involved in the service delivery. The participation is also very intensive, not the usual buyer-supplier relationship, but more like a joint activity or joint venture. We started by analysing the existing literature in two separate and usually not interconnected themes: inter-organisational cooperation/relationships without common organisational structure and managing complex, customer and labour intensive multi-stage service process. Oliver (1990) presents a joint program cooperation form, which is a specific programme of two agencies working together when planning and implementing common activities but without a common organisational structure. The prerequisite according to Oliver is that the objectives of two individual agencies can only be achieved by cooperation. Joint programs are formalized arrangements which tend to institutionalize and stabilize the inter-organisational exchange of resources. Oliver mentions that this kind of cooperation form is usual when dealing with social services and cooperation related to them. A linkage to managing joint operations is not mentioned, indicating a common problem in this literature. Much of the literature focuses on static exchange relationships with an economical rather than an operational perspective (e.g. Ouchi, 1979; Dekker, 2004; C¨aker, 2008). A single attempt to raise the important question related to the operational view of a cooperative service process has been made by Provan and Sebas-

16 Reducing Through Inter-Organisational Process Coordination

201

tian (1998). They mention the idea of an informal or formal integration structure that aims to coordinate the services their client needs. This integration form varies between single information transaction to a full-scale sharing of resources and programs. Lundgren (1992) discusses the coordination of activities in the industrial network context. He states that coordination in networks usually means organising functions and flows, activities and relationships within a network to increase the effectiveness of the activities. Coordination of activities will cause changes in the resource structure, but in the network context it also offers possibilities to form new kinds of combinations of different resources and activities. Although Lundgren discusses industrial networks, similar characteristics apply also for the service context. The coordination focuses on cooperation, the process of interaction between the members in the network. Laing and Lian (2005) have formed four different categories of factors involving the coordination of interactions in the service context: trust, closeness, process factors, and organisational policy factors. Our target is to find key elements of the processes in our specific case systems. We base our processual view on Wemmerl¨ov (1989), who uses two different process categories: rigid service processes and fluid service process. Our interest is in fluid service processes, which are described as follows: they usually require relatively high technical skills; a great amount of information is needed in order to specify the nature of service needed exactly; and the service worker goes through an unprogrammed search processes and makes several judgement decisions, meaning that the process is not well defined; the volume of people handled per a unit of time is low; and the workflow uncertainty is high, process normally involves only one customer at a time, the response time to a customer-initiated service request is often fairly long. Wemmerl¨ov adds that fluid processes are often people-dominated and they often exist in highly professional organisations. The challenges in managing these kinds of processes are high, and because of the wide scope and unpredictable service requests, forecasting of service flows should be tightly connected to the resource use per time unit (load), and most efforts on development should be focused on the expertise of individual persons and the information on which they base their process control decisions. Though according to Wemmerl¨ov (1989), standardization of the fluid process is difficult or worthless, Bowen and Youngdahl (1998) present an opposite idea in which they combine lean thinking and the product-line approach. Their idea is that mass customization is possible both in the industrial world and the service world, and it is all about “having flexible processes and structure when producing variable or even individually customized products or services” (op.cit, 1998, p.222). One of the elements they include in their idea is the networked organisation that does not exclude responsiveness, flexibility or focus on individual customers. The third part of the literature analysis concerns defining the lead-time in the context of multi-staged fluid service process. According to Wemmerl¨ov (1989, p.32) “Accurate time standards are difficult to derive, and, due the variance in tasks and processing times, often not worth the effort developing.” From the customer perspective, the lead-time of a service is important because it affects on ‘service expe-

202

Henri Karppinen and Janne Huiskonen

rience’ and further on ‘service quality’. From the service provider’s perspective, the lead-time is often connected to costs; the longer the customer is in the process, the higher the costs are. We see the service process lead-time as a relevant and important measurement, and it should not be undervalued. In this study we consider the true lead-time as the time the customer is within the process, and both passive and active time should be included in the lead-time. The research gap is based on an observation that we do not have enough focused theory about connecting areas of inter-organisational cooperation and managing a multi-staged service process and the need existing in real service systems. Our research target and the effort of filling the gap “to analyze lead-time related problems in the fluid service process and to find the factors that have an influence on lead-time improvement efforts”

16.3 Lead-Time Related Problem Analysis The problem analysis was conducted in three different public sector service process using action research methodology. The selection of methodology was guided by two main needs: a need to increase understanding of the researchers from the research point of view, in order to benefit the theory building (primary), and a need to benefit the client system with feasible and process oriented solutions (secondary). The client system included two healthcare processes and one residential application process in a Finnish municipality. The original objectives in the process development projects included improving the productivity and service quality, improving the customer flows and lead-time as well as achieving better customer satisfaction, both with internal and external customers. The first process (process A) involved children under the age of 7. The patients in this process typically suffered from developmental disorders and problems that were caused directly or indirectly by their parents. The second process involved young persons in the age from 13 to 20. These patients had developmental disorders and parent related-problems, but also mental, alcohol and drug abuse problems (process B). The third process (process C) was a service process for people applying for an apartment or a house from the municipality, and unlike in the first two processes, the process itself was much less visible to the customer. The action research projects included all six main steps of the action research cycle: data gathering, data feedback, data analysis, action planning, implementation, and evaluation (Coughlan and Coghlan, 2002). In process A the action research steps were conducted in 9 workshops ( day), in process B in six, and in process C in three workshops. The research team included three researchers: a facilitator, an observer in the workshops, and a researcher not participating in the workshops but taking part in the data analysis and evaluation. This kind of setting was needed because of the validity, and subjectivity challenges related to action research as a methodology (Coughlan and Coghlan, 2002; Zuber-Skerrit and Fletcher, 2007). The integration of methodological and problem solving steps is presented in the fig.16.1.

16 Reducing Through Inter-Organisational Process Coordination

203

Fig. 16.1 Integration of methodological steps and process related problem solving

We base our view on observations made in the workshops, focusing on the process mapping, problem analysis and solution definition. Already in process A we learned that process level coordination, as presented in the process management literature, is inadequate because it tends to create a situation where problem analysis causes too aggregated solutions. Flexibility is needed not only in the service processes but also in the problem analysis. The process mapping and analysis chart (modelling part) in the case process B is presented in the fig.16.2. The analysis indicates that in these kinds of processes, the most difficult problem when trying to improve the lead-time of a service process is a too static view of the process. The thinking of managers is too much focused on setting high service standards, measurements, quality systems etc., and at the same time the employees on the operating level are too much focused on individual tasks. The service process level does not exist at all, or at least not the kind of process perspective founded in the service process literature. Flexibility and responsiveness do not work because the service process definitions and operational policies are based on static and unrealistic definitions. The service process lead-time is a sum of different service paths, often tailored for individual customers, with the result that the original lead-time targets are never met. The original targets and lead-time measurement however presuppose that the process has a well defined workflow and the customer will get ‘standardized service’ with low variance. On the basis of problem analysis, we can also state that the process level view that should include managing the process flows and controlling the interfaces between

204

Henri Karppinen and Janne Huiskonen

Fig. 16.2 Process map and analysis chart in the case process B

the process events, did not change if the load in the process changed. This meant in processes A and B that alternatives were originally not defined at all for different process flows. If the process flow stopped, then the customers/patients were moved to an unplanned and often unintended place to wait for the planned process flow to continue again. In processes A and B, the lead-time measured (if measured) was not a result of preset and locked service paths, but rather the unplanned (true) service paths. At the same time, when the process alternatives were not defined, the process included events where, according to the process descriptions, different actors produced the same activity, but no flow control existed. As process level control did not exist, the solutions made at the event level, were not optimised and the lead-time became longer and longer. Despite the fact that there should have been a cooperative process, the focus was purely on single events. In process C, the application process was a centered solution where three different major apartment/resident owners bought centralised service from the municipality service provider. The problem in the centralised process caused in the worst case a stop in the service for all three companies, even if the ‘problematic customer’ was a customer of only one company. In general the problems were process flow control focused (flow control was based on diagnosis made in individual events e.g doctors, experts; process load based control decisions did not exist). Simplified, a fluid service process is all about recognizing the state of the process proactively and making control decisions that are process-loading and flow-based. This does not exclude

16 Reducing Through Inter-Organisational Process Coordination

205

in the professional service context the quality of a diagnosis made in an individual event. The solutions aiming for improved lead-time were in the process A and B forming a new inter-organisational and multi-professional “care group” especially for the early stages of the service process. Also some structural changes and policy solutions were made in the first phases of a process e.g reducing process entries. In the process C, the radical solution was to replace the original process with three different application processes for each major company that were operated by the companies not the municipality. The most important problems and solutions made are presented in the fig.16.3.

Fig. 16.3 Lead-time improvement related solutions in processes A, B and C

16.4 Discussion As an answer to the research gap defined, we found out that in all case processes the dependency/relationship setting between actors was created first, and only after that the single service processes were planned. This creates a locked situation, where the cooperation is based on preset dependencies, not on the actors’ interests/objectives, or the service delivered for the customer. In our opinion the inter-organisational setting creates a need to develop the service process, not only on the process level but

206

Henri Karppinen and Janne Huiskonen

also on the inter-organisational level and the single event level. Unlike the literature related to the subject, we consider that a fluid service process is controllable in the inter-organisational setting. The idea is that if flexibility and responsiveness are to be maintained on the process level, the actions can and should be controlled and the operations standardized on the inter-organisational level. Only when the process-related policies, organisational roles, rules, the required service paths and alternatives for the different service loads are set on the interorganisational level, can the service process planning begin. The process level planning should be based on polices created at a higher level of cooperation, and the service process should be planned to be active, or even proactive, not static like the service literature describes. Agreeing about common policies does not mean that static process boxes and arrows for a single option only have to be formed. The idea of three-level planning is that the service process is intelligent and responsive, a viable system. This is achieved by doing the process planning in cooperation with the organisations, and setting needs for the resources and the process alternatives needed in the service delivery together. At the single event level, which is the smallest entity in our model, the actions are based on process/event-related professionalism. Therefore the flexibility at this level is based on making the diagnoses needed, and having the right selections of the service process alternatives for different customer cases and process loads. These procedures, the decision-making rules, must be pre-planned, and the person who diagnoses only selects the right path for the customer case based on the state of the service process (intelligent service management). As a summary, the inter-organisational level is for the service process policy, rules, and role setting; the process-level planning must be based on finding the right process paths and the alternatives for the different process loads, and the event-level must be based on the required professional skills and diagnosing the customer cases, as well as strict rules when controlling the process flows. The study has some scientific limitations, and one of them is the idea of ‘levels of coordination’ being formed through the observations made during the action research projects, not beforehand. The decisions made in case process C were, however, a result of new thinking, emphasizing “the levels of coordination”. The second limitation is that we do not have enough quantitative data to validate our observations and client system view related to service process lead-times (based mostly on qualitative data). The implications for future research and also for practitioners are similar: we need to test these ideas further, to find out whether the model of ‘intelligent service’ is beneficial and applicable in other services process, as well. One important issue related to future research is the challenge of modelling intelligent service, and maintaining at the same time the relevant information and intelligibility. We intend to focus also in the future on the ‘service-intelligence’-models in the diagnosis-focused service processes.

16 Reducing Through Inter-Organisational Process Coordination

207

References Bowen D, Youngdahl W (1998) “Lean” service: In defense of a production-line approach. International Journal of Service Industry Management 9(3):207–225 C¨aker M (2008) Intertwined coordination mechanisms in interorganizational relationships with dominated suppliers. Management Accounting Research 19:231– 251 Coughlan P, Coghlan D (2002) Action reseach for operations management. International Journal of Operations and Production Management 22(2):220–240 Dekker H (2004) Control of inter-organizational relationships: Evidence on appropriation concerns and coordination requirements. Accounting, Organizations and Society 29(1):27–49 Laing A, Lian P (2005) Inter-organisational relationships in professional services: Towards a typology of inter-organisational relationships. Journal of Services Marketing 19(2):114–128 Lundgren A (1992) Coordination and mobilisation processes in industrial networks. In: Yxelsson B, Easton G (eds) Industrial Networks A New View of Reality, Routledge, London Oliver C (1990) Determinants of interorganizational relationships: Integration and future directions. Academy of management review 15(2):241–265 Ouchi W (1979) A conceptual framework for the design of organizational control mechanisms. Management science 25(9):833–848 Provan K, Sebastian J (1998) Networks within networks: Service link overlap, organizational cliques, and network effectiveness. Academy of Management Journal 41(4):453–463 Wemmerl¨ov U (1989) A Taxonomy for Service Process and its Implications for System Design. International Journal of Service Industry Management 1(3):20– 40 Zuber-Skerrit O, Fletcher M (2007) The quality of an action research thesis in the social sciences. Quality Assurance in Education 15(4):413–436

Chapter 17

Is There a Relationship Between VC Firm Business Process Flow Management and Investment Decisions? Jeffrey S. Petty and Gerald Reiner

Abstract This study on the management of business process flows of venture capital (VC) firms explores the relationship between the utilization rate of the human resources within the VC firm and deal (project) rejection rate under consideration of contextual factors. We employ an exploratory research design (a historical case analysis) as well as quantitative model oriented research based on empirical data in order to understand what is really going on in terms of VC firm processes with regard to their system dynamics. We utilize a longitudinal data set comprising 11 years of archival data covering 3,340 investment decisions collected from a European-based VC firm. The results indicate that, over time, there are considerable dynamics in the VC decision making process. Specifically, the investment decisions of venture capitalists are influenced by firm-specific factors related to the human capital resources of the firm, namely management capacity. Implications of these results for research and practice, in venture capital as well as other service industries, are discussed.

17.1 Introduction Venture capital (VC) firms, which are typically staffed by a small team, receive deals according to a stochastic arrival rate over the life of a fund. During this time the team is responsible for the deal flow process that involves the evaluation of deals, structuring investments, managing portfolio companies and liquidating the fund’s portfolio (Tyebjee and Bruno, 1984; Fried and Hisrich, 1988), as well as managing the firm itself. These firm processes are based upon assumptions within a deterministic enJeffrey S. Petty Lancer Callon Ltd., Suite 298, 56 Gloucester Road, UK-SW7 4UB London e-mail: [email protected] Gerald Reiner Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected]

209

210

Jeffrey S. Petty and Gerald Reiner

vironment (no demand variability, etc. is taken into consideration) and are typically driven by financial performance measures rather than operational ones. Hence, the question arises as to how the “quality” of the process output is affected during the life of a venture fund (e. g. Is there a higher risk of rejection, even for a potentially suitable project, at different times based upon capacity problems within the VC firm?). The literature focused on VC decision making (Macmillan et al, 1985; Dixon, 1991; Zacharakis and Shepherd, 2005) does not take this firm-specific aspect into consideration and focuses typically on the “quality” of the potential deal based upon: (i) the company’s management team, (ii) the market, (iii) the product or service, and (iv) the venture’s financial potential. Thus, the existing literature fails to address the potential impact of the firm-specific processes and resources (Barney, 1986, 1991; Hitt and Tyler, 1991; Mahoney and Pandian, 1992) on the strategic decisions (Eisenhardt and Zbaracki, 1992), and ultimately the strategy and performance, of a firm. As such, until now there have been no direct time related requirements considered for designing, planning or managing the VC firm’s processes. The management and delivery of a VC firm’s product and related processes, as well as other professional service firms, can typically be characterized as “maketo-order” (Naylor et al, 1999). Therefore, in general, the question arises about the quantity of available firm resources (e. g. people, systems, capital) that are required to successfully pursue the firm’s strategy within the specified time and with the specified level of service (e. g., Jammernegg and Reiner, 2007). This capacity management approach is suitable for classical relationships between a service provider and its customers/clients wherein the scope of services and time requirements can be specified in a contract. However, what happens in terms of resource allocation and process effectiveness if there is no clear classification of order requirements, especially with respect to time related aspects, possible? Therefore, in our study we deal primarily with the following research questions, (1) Over time, what is the impact of firm structure (lean staffing) on deal evaluation and decision making? (2) What are the firm-specific processes that influence deal evaluation and decision making under consideration of dynamic aspects? (3) Is there a relationship between the utilization rate changes over time and the related investment decisions of VC-firms?

17.2 Research Method To address the research questions outlined above, we chose an exploratory research design (a historical case analysis) which is recommended for investigating phenomena that are subtle and/or hardly understood . This type of research design permits a thorough understanding of the phenomena in question, which is of great importance for developing new knowledge on complex and dynamic phenomena such as the impact of organizational structure and staffing on VC decision-making over time in a VC firm (Fried and Hisrich, 1988). By collecting and analyzing data in a single-case,

17 VC Firm Business Process Flow Management and Investment Decisions

211

longitudinal setting, the research design adopted in our study focuses on validity and accuracy rather than generalizability, and provides the basis for the development of new theory which then can be further advanced following a multiple-case replication logic and through large-scale survey research (Eisenhardt, 1989; Strauss and Corbin, 1998; Yin, 2003). As this study seeks to explore the factors affecting the VC’s processes over time, the use of archival data analysis is preferred over an interview or survey approach because it allows for the collection and analysis of the different measures over several time periods. This approach also provides access to information that helps to gain a more realistic view of the actual environment as well as the actions that were made by the subjects at the time. Thus, this helps to enhance the validity of the data as it eliminates recall bias on the part of the subject as well as other limitations often associated with self-reported techniques (Hall and Hofer, 1993; Shepherd and Zacharakis, 1999). We also conduct quantitative modeloriented research, especially under consideration of empirical data, based upon the results of the qualitative research activities. Bertrand and Fransoo (2002) pointed out that the methodology of quantitative model-driven empirical research offers a great opportunity to further advance theory (Davis et al, 2007). In general, quantitative model-based empirical research provides the ability to generate models of causal relationships between control variables and performance variables. These models are then analyzed or tested using different scenarios involving varying levels of constraints on the subject variables. The primary concern of this research approach is to ensure that there is a fit between the actual observations and actions and the resulting model, which is based upon reality. Utilizing a combination of different research approaches thus enables us to address the aforementioned research questions.

17.2.1 Data Collection and Sample The data used in the model was collected from the archival records of a Europeanbased VC firm and included the investment/rejection decisions on more than 3,600 deals that had been received by the firm over an 11-year time period. The data set was created by reading all 7,284 passages of text in the firm’s deal flow data base as well as related emails and memos in the archived deal files. The database entries for deals that had made it beyond the initial screening phase into the evaluation and due diligence phases typically contained a synopsis of the VC’s findings and views; a random sample of 350 deals was selected in order to compare the notes in the files to the comments entered in the action log and there was no evidence of gross omissions or any material rewording of comments, so the database was deemed a reliable data source. The time a deal spent in the selection process ranged from one day to more than a year and, after eliminating those deals that lacked sufficient information (e. g. date of submission or VC decision was missing) to be included in the study the resulting sample included 3,340 deals. The firm was staffed by a small team of VCs and the average acceptance rate of deals submitted to the firm over the entire period was 1%, which is consistent with the description of VC firms

212

Jeffrey S. Petty and Gerald Reiner

and industry averages reported in many other studies. We will use the term “firm” to describe the VC firm, whereas the terms “company”, “deal”, and “proposal” all apply to the entrepreneurial ventures evaluated by the VC.

17.2.2 Data Analysis The initial qualitative analysis involved an interpretative approach (Glaser and Strauss, 1967; Roberts, 1997; Strauss and Corbin, 1998) to the documents containing the VC’s views and decisions related to the deals they had reviewed over the 11-year time period. Those comments representing the firm’s view or reason for the final decision, both explicit and implied, for the ultimate investment/rejection decision were collected and coded along with descriptive information pertaining to each deal, which included the date the deal was received by the firm, the date of the final investment decision. Additional context specific factors including the number and tenure of the VCs in the firm, the role of the VC firm in each of their portfolio investments (e. g. lead investor, co-lead, investor) and any extraordinary events (e. g. raising money for a new fund) over the 11-year time period of the study were also included in the analysis. In order to discuss our research questions in greater detail we also developed a quantitative model based upon the results of first research activities. In particular, we modeled the VC firm’s process operations in detail (see also Figure 21.2) based upon the descriptions of the VC decision making process in the literature (Tyebjee and Bruno, 1984; Fried and Hisrich, 1994; Shepherd et al, 2005). Briefly stated, the principle steps depicted in Tyebjee and Bruno’s (1984) model of the VC process are: (a) origination, which includes all those activities related to sourcing proposals, (b) screening, the VC’s initial appraisal of the documentation that typically results in the majority of proposals being rejected on the basis of obvious flaws and/or firm specific criteria, (c) evaluation, which involves a more in-depth assessment of the company’s management team, financials, product and market, (d) structuring, wherein the VC and the company negotiate the specific deal terms and (e) post-investment activities, which encompasses the VC’s on-going monitoring and control activities as well as operational support of their portfolio companies and, ultimately, liquidation of portfolio deals. Shepherd et al (2005) created a VC process model based on simple Jackson open queuing networks where the expressions of venture flows and status are averages in the long run. This research work provides important theoretical input for developing our quantitative empirical model. However, their research differs from this study in two important ways. First, they do not have actual data from a VC firm so their study remains more theoretical in nature (e. g., application of Jackson open queuing network). Second, their focus is more on an individual’s allocation of time and the influence of opportunism rather than an organizational level analysis. Based on this previous research work we develop a VC process model adapted from the main operational activities that takes into consideration empirical data and focus on the relationship between operations management within the VC firm and the acceptance as well as rejections of projects.

17 VC Firm Business Process Flow Management and Investment Decisions

213

This model will provide time specific utilization information for further statistical analysis.

Fig. 17.1 Process Model of VC firms

We will describe the main mathematical equations, input data as well as variables (flows and stocks) and performance measures of our model. The above described initial exploratory research proved the input data for our analysis, i. e., proposals (Pt ), resources (Rt ), rejections I (RIt ), rejections II (RIIt ), rejections III (RIIIt ), activity times for screening, evaluation structuring and portfolio (AIt , AIIt , AIIIt , AIVt ), termination (Tt ), number of newly hired employees (Ht ) and number of employees departing from the firm (Qt ) . The process operations of the venture capital firms are specified as follows. Number of proposals waiting for Screening (SCt ) is increased by Pt and reduced by the maximum processing rate (OIt ). OIt is determined by the activity time I (AIt ) as well as the available number of resources within the resource pool (Rt ) SCt = SCt−Δ t + (Pt − OIt )Δ t

Rt OIt = min , SCt−Δ t AIt

(17.1)

It = OIt − RIt where RIt ≥ OIt

(17.2) (17.3)

Number of proposals waiting for Evaluation (EVt ) is increased by inflow I (It ) and reduced by the maximum processing rate (OIIt ). OIIt is determined by the activity time II (AIIt ) as well as the available number of resources within the resource pool (Rt ) EVt = EVt−Δ t + (It − OIIt )Δ t

(17.4)

214

Jeffrey S. Petty and Gerald Reiner

Rt OIIt = min , EVt−Δ t AIIt

(17.5)

IIt = OIIt − RIIt where RIIt ≥ OIIt

(17.6)

Number of proposals waiting for Structuring (STt ) is increased by inflow II (IIt ) and reduced by the maximum processing rate (OIIIt ). OIIIt is determined by the activity time III (AIIIt ) as well as the available number of resources within the resource pool (Rt ) STt = STt−Δ t + (IIt − OIIIt )Δ t OIIIt = min

Rt , STt−Δ t AIIIt

(17.7)

IIIt = OIIIt − RIIIt where RIIIt ≥ OIIIt

(17.8) (17.9)

Number of projects within the Portfolio (POt ) is increased by inflow III (IIIt ) and reduced by the outflow/termination (Tt ) of projects. POt = POt−Δ t + (IIIt − Tt )Δ t

(17.10)

Finally, we will define the most important performance measure in the context of our research framework, i. e., utilization (Ut ). To be able to calculate Ut it is necessary to take the theoretical available number of resources within the resource pool (Nt ) as well as the real available number of resources within the resource pool (Rt ) into consideration based on allocation of employees to VC activities (i. e., AIt , AIIt , AIIIt , AIVt ) as well as the hiring (Ht ) and departure (Qt ) of employees. Nt = Nt−Δ t + (Ht − Qt )Δ t Rt = Rt−Δ t − (OIt AI + OIIt AII + OIIIt AIII + IIIt − OIt−Δ t AI −OIIt−Δ t AII − OIIIt−Δ t AIII − Tt − Ht + Qt )Δ t Ut = 1 −

Rt Nt

(17.11)

(17.12) (17.13)

Furthermore, we defined time related performance measures, to be able to validate our model. In particular, we calculate the flow time for each VC process operation based on waiting time as well as activity time, i. e., average screening flow time (SCT ), average evaluation flow time (EV T ), average structuring flow time (ST T ) as well as well as the average flow time within the portfolio (POT ). To be able to calculate the flow times (T L) we use a general distributed (times between arrivals and service time) queuing model (Hopp and Spearman, 1996). Based

17 VC Firm Business Process Flow Management and Investment Decisions

215

on the application of queuing models we were able to calculate the average time (periods in month) spent for evaluation before investment (BIT ) BIT = SCT + EV T + ST T

(17.14)

The average time a project deal spends in process before being rejected (PRT ) is defined as PRT =

SCT ∑ RIt + EV T ∑ RIIt + ST T ∑ RIIIt ∑ (RIt + RIIt + RIIIt )

(17.15)

Validation of the process analysis model is based on empirical data generated during the initial analysis (see Table 17.1) as well as simulated data based on 136 periods/months (T ). The model validation shows that our model is able to analyze the performance of VC process operations. Table 17.1 Model validation Performance measure Empirical data Model results BIT PRT

7.9 months 2.1 months

7.5 months 2.0 months

17.3 Results Throughout the period under study there were a surprising number of instances (n = 67) when the team of VCs within the firm simply did not have the management capacity to adequately evaluate a potential deal, even when the deal was acknowledged to be potentially viable (e. g. “Interesting but time constraints due to other due diligence.”). Similarly, during these periods of increased activity, deals that were viewed as potentially time consuming were rejected, despite any potential interest (e. g. “it has high potential but very much handson work and stirring will be required”). These instances demonstrate that the team’s capacity utilization, in combination with the characteristics of the potential “client”, play a significant role in the decision-making process. Thus, a firm’s strategy with respect to staffing and structure may need to be adjusted in order to adapt to changing demand needs. Also, once all of the 3,340 deal decisions had been coded, the occurrence of deal specific decision reasons in relation to the VC management team’s utilization were compared with the chi-square test. All analyses were performed with SPSS (version 15.0). Multiple scenarios using different VC team utilization rates were tested and the chi-square statistic for all tests was highly significant (P value of .000) thus indicating that the stated reason(s) for the decisions are associated with the VC team’s available safety capacity. Additionally, further quantitative analysis

216

Jeffrey S. Petty and Gerald Reiner utilization

arrival rate

rejection rate 70

100%

90% 60 80%

[%]

60%

40

50% 30

40%

30%

number of proposals

50

70%

20

20% 10 10%

0%

0 0

20

40

60

80

100

120

Time (months)

Fig. 17.2 Process Model results

revealed, contrary to what may be expected, that the arrival rate of new deals did not necessarily predict deal rejection rates or team utilization rates (see Figure 17.2). However, the relationship between the team’s utilization rate and deal rejection is much more pronounced, especially in the latter months of the fund when VCs are operating at a high utilization rate as a result of on-going portfolio management activities.

17.4 Discussion and Conclusion In this study we explore the relationship between utilization rate and decision making, specifically the rejection rate, under consideration of contextual factors. Based upon both qualitative and quantitative analysis we find that there are considerable dynamics in the VC decision making process, especially over time. In this context, we have shown that operations management and in particular capacity management provides further valuable insight. Second, our findings extend prior conceptualizations of the VC decision-making process by showing that the importance of decision making criteria can significantly change over the lifecycle of a fund. Developing and managing a portfolio places constraints on the resources (i. e., human resources) of the VC firm so that during periods of extensive due diligence, deal closings and managing existing portfolio companies there is less time to spend screening and evaluating new deals. Additionally, this study raises the question of whether or not the existing research focused on individual VC decision making is in fact represen-

17 VC Firm Business Process Flow Management and Investment Decisions

217

tative of the actual decision making within the firm. VC decision making is but one aspect of an organizational process and therefore researchers should not assume that a study of individual VC decision making is the same as VC firm level decision making. Much of the research that has been conducted on VC decision making appears to be suited to the initial screening phase of the process but without considering the current needs or requirements of the firm it is not practical to assume that we are developing models that accurately depict the firm-level processes and actions.

17.4.1 Implications Looked at from the perspective of the VC firm, our findings suggest that VCs should evaluate their existing management capacity. They should develop strategies to accommodate for times when they experience increased deal flow, above average due diligence activity or number of deal closings and hands-on management of the portfolio firms. While there has been limited research on how VC firms are organized and managed (Gorman and Sahlman, 1989), there is no evidence that VCs acknowledge or attempt to address any potential under-capacity in terms of management time, especially in the latter years of a fund. This missing focus on capacity management in combination with the performance evaluation of VC firms is solved in our research study by integrating state-of the art service operations management knowledge. The apparent constraint on VC management time over the life of a fund adds to the existing literature focused on the allocation of a VC’s time. Although the issue of VC time allocation has typically been concerned with the management of the fund’s portfolio (Gorman and Sahlman, 1989; J¨aa¨ skel¨ainen et al, 2006), our findings provide additional evidence that both pre and post-investment activities (Gifford, 1997; Shepherd et al, 2005) may influence the decisions of the VCs within a firm. Furthermore, considering that VC management capacity has a considerable impact on the deal selection process, entrepreneurs need to be aware that there will be instances when the entrepreneur is basically at the right place at the wrong time: their business proposal is compelling, yet the VC firm does not have the capacity to evaluate it and therefore, out of necessity, makes the decision to reject it. As is the case in most service settings, entrepreneurs may be limited in what they can do to influence the expert’s opinions of their proposals yet they may find it well worth the effort to attempt to learn about the current capacity of the firm in order to avoid those times when the firm is overloaded with work in terms of management time, thus improving the chance that the decision maker will be able to devote their full attention to the proposal. Finally, given that other service firms (e. g. legal, advertising, financial, and consulting) are also characterized by “make-to-order” processes, lean staffing, and demand variability, there may be possibilities for generalization of the research results across the service sector.

218

Jeffrey S. Petty and Gerald Reiner

17.4.2 Limitations and Future Research A key limitation of this study is that the researchers were not present in the VC firm during the actual deal evaluation process so there is no way to guarantee that all of the relevant information was recorded in the firm’s database. Additionally, the fact that the data was obtained from a single VC firm limits the ability to generalize the findings across the industry as a whole. Although not addressed in this study, there are many other factors, such as the size and location of the VC firm (Gupta and Sapienza, 1992) and biases on the part of individual VCs (Franke et al, 2008; Matusik et al, 2008; Shepherd et al, 2003), which may also have an effect on an individual VC’s investment decisions. Despite these limitations this study shows that the criteria used by VCs are not consistent over time and provides evidence that there is much to be discovered about the VC decision making process that cannot be accomplished using existing approaches. Based upon the results of this study, selected firm-specific factors possessing a variable nature (Cyert and March, 1963) appear to have a greater impact on managerial action and decision making than previously reported. As such, more longitudinal research is required within the context of service firms that will enable researchers to capture the complexity of the task environment as well as the resulting decisions. Although each firm pursues a different strategy suited to their specific goals all venture capitalists operate under similar constraints when it comes to staffing, portfolio demands, and investment restrictions so similar results are expected in studies conducted in other firms. Further limitations related to our empirical quantitative model are that we did not take into consideration the use of resources for sourcing deals and termination (or exit) of the portfolio companies. However, in reality, the majority of deals received by a VC firm are unsolicited so the sourcing activity is more passive in nature (Fried and Hisrich, 1988), which would imply that the time requirements on the part of the VC are quite minimal. Further research work should deal also with the identification and evaluation of potential VC firm process improvements (capacity management, etc.) and finally the implementation of the process improvements to be able to finish the entire research cycle (Mitroff et al, 1974) in combination with longitudinal research.

References Barney J (1991) Firm Resources and Sustained Competitive Advantage. Journal of Management 17(1):99 – 120 Barney JB (1986) Organizational Culture: Can It Be a Source of Sustained Competitive Advantage? Academy of Management Review 11(3):656 – 665 Bertrand J, Fransoo J (2002) Operations management research methodologies using quantitative modeling. International Journal of Operations and Production Management 22(2):241–264 Cyert R, March J (1963) A behavioral theory of the firm

17 VC Firm Business Process Flow Management and Investment Decisions

219

Davis J, Eisenhardt K, Bingham C (2007) Developing theory through simulation methods. The Academy of Management Review (AMR) 32(2):480–499 Dixon R (1991) Venture Capitalists and the Appraisal of Investments. Omega 19(5):333 – 344 Eisenhardt KM (1989) Making fast strategic decisions in high-velocity environments. Academy of Management Journal 32(3):543 – 576 Eisenhardt KM, Zbaracki MJ (1992) Strategic Decision Making. Strategic Management Journal 13(Special Issue):17–37 Franke N, Gruber M, Harhoff D, Henkel J (2008) Venture Capitalists’ Evaluations of Start-Up Teams: Trade-Offs, Knock-Out Criteria, and the Impact of VC Experience. Entrepreneurship: Theory & Practice 32(3):459 – 483 Fried VH, Hisrich RD (1988) Venture Capital Research: Past, Present and Future. Entrepreneurship: Theory & Practice 13(1):15 – 28 Fried VH, Hisrich RD (1994) Toward a Model of Venture Capital Investment Decision Making. FM: The Journal of the Financial Management Association 23(3):28 – 37 Gifford S (1997) Limited attention and the role of the venture capitalist. Journal of Business Venturing 12(6):459 – 482 Glaser B, Strauss A (1967) The discovery of Grounded Theory: Strategies for qualitative research. Chicago, Aldine Pub. Co. Gorman M, Sahlman WA (1989) What do venture capitalists do? Journal of Business Venturing 4(4):231 – 248 Gupta AK, Sapienza HJ (1992) Determinants of venture capital firms’ preferences regarding the industry diversity and geographic scope of their investments. Journal of Business Venturing 7(5):347 – 362 Hall J, Hofer CW (1993) Venture capitalists’ decision criteria in new venture evaluation. Journal of Business Venturing 8(1):25 – 42 Hitt M, Tyler B (1991) Strategic decision models: Integrating different perspectives. Strategic Management Journal 12(5):327–351 Hopp WJ, Spearman ML (1996) Factory Physics: Foundations of Manufacturing Management. Irvin Inc., Chicago J¨aa¨ skel¨ainen M, Maula M, Sepp T (2006) Allocation of Attention to Portfolio Companies and the Performance of Venture Capital Firms. Entrepreneurship: Theory & Practice 30(2):185 – 206 Jammernegg W, Reiner G (2007) Performance improvement of supply chain processes by coordinated inventory and capacity management. International Journal of Production Economics 108(1-2):183 – 190 Macmillan IC, Siegel R, Narasimha PNS (1985) Criteria used by venture capitalists to evaluate new venture proposals. Journal of Business Venturing 1(1):119 – 128 Mahoney JT, Pandian JR (1992) The Resource-Based View Within the Conversation of Strategic Management. Strategic Management Journal 13(5):17–37 Matusik S, George J, Heeley M (2008) alues and judgment under uncertainty: Evidence from venture capitalist assessments of founders. Strategic Entrepreneurship Journal 2(2):95–115

220

Jeffrey S. Petty and Gerald Reiner

Mitroff II, Betz F, Pondy LR, Sagasti F (1974) On managing science in the systems age: Two schemas for the study of science as a whole systems phenomenon. Interfaces 4(3):46 – 58 Naylor J, Naim M, Berry D (1999) Leagility: Integrating the lean and agile manufacturing paradigms in the total supply chain. International Journal of Production Economics 62(1-2):107–118 Roberts C (1997) Text analysis for the social sciences: Methods for drawing statistical inferences from texts and transcripts. Lawrence Erlbaum Associates Shepherd DA, Zacharakis A (1999) Conjoint analysis: A new methodological approach for researching the decision policies of venture capitalists. Venture Capital 1(3):197 – 217 Shepherd DA, Zacharakis A, Baron RA (2003) VCs’ decision processes: Evidence suggesting more experience may not always be better. Journal of Business Venturing 18(3):381 – 401 Shepherd DA, Armstrong MJ, L´evesque M (2005) Allocation of attention within venture capital firms. European Journal of Operational Research 163(2):545 – 564 Strauss A, Corbin J (1998) Basics of qualitative research: Techniques and procedures for developing grounded theory. Sage Publications Inc Tyebjee TT, Bruno AV (1984) A Model of Venture Capitalist Investment Activity. Management Science 30(9):1051–1066 Yin R (2003) Case Study Research Design and Methods, 3rd edn. Sage, Thousand Oaks, CA Zacharakis A, Shepherd DA (2005) A non-additive decision-aid for venture capitalists’ investment decisions. European Journal of Operational Research 162(3):673 – 689

Chapter 18

What Causes Prolonged Lead-Times in Courts of Law? Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

Abstract The paper highlights the challenges in process performance issues in large public sector professional organizations. Factors causing process inefficiencies and prolonged lead-times in two Finnish Courts of Law are introduced and analyzed.

18.1 Introduction The Finnish Constitution states that everyone has the right to have his/her legal case heard properly and without undue delays before a legally component court of law. This statement is also written into the European Declaration of Human Rights. Finnish courts have struggled with prolonged lead-times and have received appeals regularly from the European Court of Human Rights concerning unreasonable duration in the handling of judicial cases. Complaints about delays in courts are not solely Finnish phenomena nor are they something new. The court systems in many countries have been criticized for years for being inflexible, for taking too long, and for demanding more and more resources (Martins et al, 2007; McWilliams, 1992; Smolej, 2006). This research started with a call for help from the Finnish Ministry of Justice wanting to study the court system processes in order to find ways to reduce the Petra Pekkanen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN53851 Lappeenranta, Finland, e-mail: [email protected] Henri Karppinen Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN53851 Lappeenranta, Finland Timo Pirttil¨a Department of Industrial Management, Lappeenranta University of Technology, P.O. Box 20, FIN53851 Lappeenranta, Finland

221

222

Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

time that cases stay in the process without endangering the quality of decisions or increasing the resources. The backbone of court system operations is, like in all professional organizations, autonomous work of highly motivated and educated individuals (Brock et al, 1999; Lowendahl, 2005; Mintzberg, 1983). In the court system, the judges also need to be completely independent and “beyond control” to ensure objective ruling. Still, at the same time the court system is a process with a set of sequential tasks and activities linked together, concerning different participants. It is a process that demands continuous and coordinated flow of a very large number of individual and infinitely different types of cases. Court systems are organizations balancing between the needs of independent professional work and effective massproduction processes. The problems with process performance indicate that process effectiveness issues have not been given the attention they need in different areas of justice organizations operations. These features are often considered, at least to some degree, opposing. There is a fear that increasing the process viewpoint and process performance will lead to unfavorable circumstances for professional work and thus weaken the quality of the decisions made. The aim of the research project was to find ways to help court system organizations find new ways of actions that take the divergent requirements into account better. The first part of the task was to find out what the exact problem was and what caused it. This paper concentrates on defining the lead-time problem and identifying and analyzing the reasons and sources for inefficiencies and prolonged lead-times. The main research question is: • What are the main factors in court systems current way of actions which have caused and influenced the problems in process performance and prolonged leadtimes? The analysis is based on experiences gained from large process improvement projects in two Finnish Courts of Law. The case organizations are introduced first. After that, the improvement projects and the data collection methods are presented. In section 4, an analysis of the factors behind the prolonged lead-times is introduced. Finally, concluding remarks are made.

18.2 Case Organizations The Finnish court system is tripartite for civil and criminal cases. The first level is the District Courts. The decisions of District Courts can normally be appealed in a Court of Appeal. The decisions of the Courts of Appeal, then, can be restrictively appealed in the Supreme Court. In addition, there are special courts, for example Insurance Court and Administrative Courts.

18 What Causes Prolonged Lead-Times in Courts of Law?

223

18.2.1 Helsinki Court of Appeal The first case court in this study is the largest Court of Appeal in Finland. It handles about 4000 cases annually, which make about 30 The cases are prepared and presented for decision by legally trained referanderies, who are called Senior Assistant Justices. After preparation, one of the judge members, who are called Senior Justices, goes through and verifies the prepared case. A responsible judge and referandary are appointed for every case. The cases are then decided in a Court session by a composition of three Senior Justices. In the case court, there are 170 employees and it operates in seven departments. Each department operates independently and is headed by one of the Senior Justices. The case handling operations are presented in the Fig.18.1

Fig. 18.1 Case handling operations in Helsinki Court of Appeal

The needed preparation time varies according to the complexity of the case. The cases are divided to five size-groups: S, M, L, XL, and XXL. There are two types of cases: criminal cases and civil cases. The civil cases are usually more complex and require more preparation time. The cases are also prioritized and categorized to three classes according to the assessed urgency of the case. The first priority level concerns “emergency” type of cases, which need to be handled immediately, for example child guardian issues or restraining orders. Other cases are divided to priority level 2 or priority level 3 with several criteria according to the nature of the felony or dispute. There are two main ways to handle individual cases: a written procedure or a main hearing. In a main hearing, the parties involved are present and witnesses are heard. The average lead-time in 2006 was 12 months, but the dispersion of leadtimes was huge, from weeks to several years. From the year 2003 on, the case court has solved annually more cases than have arrived. The proportion of very old cases was quite bad when the improvement project started: 34 % of the pending cases were

224

Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

older than 12 months. The age of the pending cases at the start of the improvement project is shown in Fig.18.2.

Fig. 18.2 Pending cases in Helsinki Court of Appeal on 4 May 2006

18.2.2 Insurance Court The Insurance Court is a special court for social security issues. It handles 10 000 cases annually. There are 120 employees and the court operates in three departments. There is only one case handling procedure, a written procedure. At the moment there is no formal division of cases, either by size or urgency. The average lead-time was 14 months in 2007, varying from a couple of months to several years. In recent years, the Insurance Court has solved much more cases annually than have arrived, and the number of pending cases is diminishing fast. The problem is that while the number of pending cases has almost halved in one year, the large dispersion of lead-times has not changed, or the number of very old cases dropped. The number of cases pending and their age in the years 2007-2008 are presented in Fig.18.3.

18 What Causes Prolonged Lead-Times in Courts of Law?

225

Fig. 18.3 Age of pending cases in the Insurance Court (30 September 2007 and 30 September 2008)

18.3 Process Improvement Projects and Data Collection The research and data collection was carried out using the action research approach, which is a generic term covering many forms of action-oriented research (Coughland and Coghlan, 2002; Gummesson, 2000). The process improvement project in Helsinki Court of Appeal started in May 2006 and in the Insurance Court in June 2008. The improvement team in both courts consisted of members from all organitional levels, altogether 15 persons per team. The main stages of the improvement projects were data gathering, data analysis, action planning, implementation, and evaluation. The work was done in several workshop meetings of the improvement teams. The project in Helsinki Court of Appeal is now in the evaluation stage and the project in Insurance Court in the action planning stage. The research group has actively participated in the process improvement projects as external experts and change facilitators from the beginning. The main source for data has been active observation and monitoring of the improvement work in group workshop meetings. Complementary data collection methods have included for example collecting operational statistics and generating numerical analyses from the database of clients and interviews of 60 members of personnel (30 in both case courts) concerning the problems behind process performance and process improvement potentials.

226

Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

18.4 Analyzing the Factors Behind the Process Inefficiencies and Prolonged Lead-Times The most apparent problem in the lead-times of both case courts was the fact that complex and large cases get stuck in the process for some reason. This has created both large dispersion of lead-times and several complaints concerning unreasonable duration. Because both case courts solve more cases annually than arrive, the piling up of large cases is not strictly related to a lack of resources. The next task was to analyze the process and the way of actions in order to find out what causes and furthers this phenomenon. The analysis revealed four main categories of factors, introduced in Fig.18.4.

Fig. 18.4 Sources for prolonged lead-times

18.4.1 Inappropriate Goal Setting and Performance Measurement Systems The performance indicators used in the courts are selected by the Ministry of Justice, which also sets targets and monitors their accomplishment. Public sector organizations are said to be still facing more problems associated with performance measurement than private sector organizations. One very typical trap is the use of too simplified output measures or the concentration on managing one single success factor at a time. (Rantanen et al, 2007) The most important goal and performance indicator used in the courts is the number of annual output. It is emphasized a lot and monitored carefully. This does not encourage preparing the complex, often badly overdue cases, and makes it feasible to increase the total output by ignoring the more complex cases. The overemphasis on the annual output indicator has also led to competition between departments and

18 What Causes Prolonged Lead-Times in Courts of Law?

227

restrained cooperation between them. A lot of energy is used in optimizing the number of solved cases, and the last part of the year is always employed by solving the small cases and getting the output goal filled. In the Insurance Court, the output of referanderies is monitored even more carefully. They need to prepare eleven cases every week, which has also led to a quite inflexible system. The only monitored goal and indicator for lead-time is the average lead-time of solved cases, which is monitored annually. This also makes it more feasible to solve the smaller cases. All the indicators used describe past performance and output, and there is no indicator showing what is left behind, what the current situation is, or what would be a goal for a maximum lead-time. It is quite obvious that not a lot of attention is paid to the choice of the most appropriate performance indicator, or to the dangerous and negative effects of wrong goal and performance indicator choices. The use of simplified output goals and indicators is very likely due to lack of time and knowhow in the Ministry of Justice and difficulties in defining more comprehensive outcome goals and measures when the final product is quite abstract and variable.

18.4.2 Lack of Process Efficiency-Oriented Management Practices A typical feature in professional organizations is the fact that managers are chosen by substance skills rather than managerial capabilities, which means that the best professional becomes the manager - not the best manager. Independent professionals are not, in addition, the easiest to manage, and the management of professional organizations depends much on negotiation, consensus and the good work ethic of professionals (Lowendahl, 2005; Mintzberg, 1983; Rantanen et al, 2007). The experience, knowledge and interest in process performance issues vary a lot between individual managers in the case courts. Some managers follow lead-times and process performance very carefully, some managers not at all, and do not even feel that it is important. The promotion system in the case courts is based only on achievement in judicial issues and the training of referanderies concentrates only on these issues. The courts are places for the referendaries to get experience and training and qualify as Senior Justices. This is why the referendaries change departments or even the court in every couple of years. This tradition has its benefits, but it also leads to a lack of clear responsibilities and a sense of duty over the complex cases in the current department. The high turnover of referanderies makes the whole system very vulnerable. The process is very referendaryled, and they have huge responsibility for the start-up and smooth running of the handling process. The judge’s responsibilities are unclear, and their responsible role is quite trivial in practice. The fact that judges need to be completely independent poses a lot of challenges for the management. Practically nothing can be done in situations where the work ethics of an individual judge fails. The judges can not be fired or their salary reduced. Even though the need to be totally independent was originally meant to cover only the content issues of a ruling, the convention has spread also to working methods.

228

Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

While the management must respect this status, they must also be able to intervene if backlog of cases increases without a good reason. Some of the managers do not feel superior to other judges, and do not see it to be their place to intervene in “colleagues” work. On the other hand, there are individual managers who follow the situation and intervene almost too much, which is also experienced as quite oppressive. However, the general opinion is that more managerial feedback (positive and negative) is needed and that the managers should follow the lead-times and backlogs of an individual employee more carefully and take actions if necessary, but it should be done in a constructive and respectful manner. The management system relies on a very clearly articulated hierarchy, where status and ranking are emphasized, but clear responsibilities and chains of command are missing. The follow-up and management duties are almost solely the responsibility of the Head of the Department. Lower-level superiors are appointed (Superior Referandary), but they do not have any formal manager status. Their role is mainly to distribute the cases evenly to the referendaries and to monitor the case load of the whole department. By expanding the management duties, there would be more time and resources to concentrate also on managing the process performance-related issues.

18.4.3 Lack of Production Planning at All Levels At the start of the improvement project it was a general concern and notion that it is not possible to increase productivity without increasing resources. A large number of cases are solved but without any orderliness, leading to the aging of complex cases. Restraining the number of very old cases is not a question of doing more cases but a question of doing things according to some kind of a plan and order. There is practically no production planning or production scheduling in the case courts. The measures and follow-up indicators used have led to the fact that individual workers plan their work mostly by picking out the cases that best help to meet the expectations. A very common opinion in the start of the projects was also that you can not plan professional work or you should not chain up brainwork with schedules. It is true that the lack of deadlines for cases makes planning seem quite useless in the eyes of an individual professional. The fact that no planning practices are used, leads to the fact that it is very hard to find uninterrupted time for preparing the complex cases. When the time for longer preparation is not planned and scheduled, new, more high-priority, cases always emerge and the preparation for complex cases stops. The setup times grow enormously when getting acquainted with the case material is done several times all over again. The weekly number quota for referanderies makes it even harder to find the preparation time needed for complex cases. One fact that complicates the planning is that the individual buffers of cases is so large and unmanageable that the occupants no longer know and recall the situation of an individual case nor the age

18 What Causes Prolonged Lead-Times in Courts of Law?

229

of it. The piles of cases just keep growing on your desk and you have no plan for preparing them. There is a lot of compulsory waiting time in the case handling process that is completely wasted without long-term planning. The court session will not be started to be arranged and called together until the preparation of the case is completed. The arrangements for a court session are a difficult coordination task of getting all parties present. This leads to situations where the court session may take place months or even years after the preparation has been completed. In practice the referandary and the judge need to get acquainted with the case material all over again nearer the court session.

18.4.4 Traditional Values and Attitudes of the Legal Profession Every organization and profession creates its own set of values, beliefs and attitudes. When talking about the legal profession, which has its origins in the antiquity, these values and attitudes, have a very long and traditional history. The legal professionals take much pride in their occupation and they enjoy a lot of respect from whole society as well. This status relies heavily on the complex knowledge base of judges and lawyers and the long traditions of methods and routines (Becher, 1999; Schein, 2004). In this sense there is very little room for appreciation of productivity and effectiveness. This thinking is reflected in all aspects of court operations: what is valued is measured, managed, educated and done. The quality of rulings made is followed, for example, very carefully and lot of time is spent on the spelling and phrasing of final acts and checking every little detail in the final decisions. So far lead-time has not been very high in the valuation lists and has been seen as an enemy for the traditionally valued aspects of quality. While legal professionals might not be the most dynamic or change-oriented professionals, they are very ethical and hard working and do enormous work in order to fulfill what is expected of them. The prolonged lead-times are a cause of stress but the professionals have felt quite powerless in front of this evergrowing problem. When they see that something really works and helps, despite of preconceptions, the acceptance of change initiatives is good, especially among younger professionals. Changes in values and attitudes will not happen overnight, and the improvement steps need to be justified and taken gradually.

18.5 Conclusions When starting to describe courts of law as organizations, the first word that comes to mind is professional organizations. But unlike for example small private law firms, the justice courts can also be described as production plants, operating in a multistaged process. Added with the features arising from public administration, it can be

230

Petra Pekkanen, Henri Karppinen and Timo Pirttil¨a

concluded that the courts of justices are organizations with diversified requirements for operations. If the organization and its practices are designed with too much emphasis on one single feature, the other ones will suffer. This paper has concentrated on describing the lead-time problem and its causes in courts of law. It can be concluded that process performance problems are not a matter of producing too little numbers or a problem of average lead-time. The problem is the fact that some cases get prolonged so much that the courts get complaints and backlogs emerge. Since the number of produced cases is very good in both case courts presented in this study, the source for the problem is not a lack of resources, nor is it an outcome of any other single factor. It is a complex mixture of causes and effects connected to unsuitable design parameters and way of actions, added with management practices and a value system with long history and traditions. Several public sector professional organizations are facing a similar situation as the pressures for better efficiency and productivity are constantly increased. They need examine the operational practices thoroughly and pinpoint the sources for inefficiency, and find ways to transform. In justice courts the first step is to recognize and accept the fact that they have also process effectiveness-related demands that need to be delivered as well as other quality criteria. It is largely an issue of broadening the image of the organization and its tasks, and translating that into operation. Big changes have been made to the measurement and follow-up systems, management practices and to the planning and scheduling practices in the case courts. For example in reforming the planning practices it became evident that scheduling supports professional work, not impedes it, and the time that can be spent on an individual case is longer when uninterrupted preparation time has been scheduled. All the improvement initiatives planned and taken in the case courts have proved that good results can be achieved and that the quality of justice decisions and process effectiveness are not exclusionary, but they even support each other. While long traditions of working methods and values take long time to change, it is possible to find ways to balance the requirements of professional work and process efficiency better.

References Becher T (1999) Professional practices: Commitment & capability in a changing environment. Transaction Publishers, London Brock D, Powell M, Hinings C (1999) Restructuring the professional organization: Accounting, health care and law. Routledge Coughland P, Coghlan D (2002) Action reseach for operations management. International Journal of Operations and Production Management 22(2):220–240 Gummesson E (2000) Qualitative methods in management research. Sage Publications Lowendahl B (2005) Strategic management of professional service firms, 3rd edn. Copenhagen Business School Press

18 What Causes Prolonged Lead-Times in Courts of Law?

231

Martins AL, Helgheim BI, de Carvalho JC (2007) A logistics approach to shorten lead-times in Courts of Law A case study. Proceedings from the 19th Annual NOFOMA Conference, Reykjavik Iceland McWilliams JM (1992) Setting the record straight: Facts about litigation costs and delay. Business Economics 27(4):19 Mintzberg H (1983) Structure in fives: Designing effective organizations. PrenticeHall International Rantanen H, Kulmala H, Lnnqvist A, Kujansivu P (2007) Performance measurement systems in the Finnish public sector. International Journal of Public Sector Management 20(5) Schein E (2004) Organizational culture and leadership, 3rd edn. Jossey-Bass Smolej M (2006) Time Management in Nordic Courts: Review of proposals and policies aimed at reducing delays in courts. Report series in european commission for the efficiency of justice (cepej) at its 8th plenary meeting in strasbourg, European Commission for the Efficiency of Justice (CEPEJ) Strasbourg

Chapter 19

Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains Ralf Elbert and Robert Sch¨onberger

Abstract Concerning the topic of “How Regional Value Chains Speed Up Global Supply Chains”, the research question of this paper will be the acceleration potential of logistics clusters in the entire supply chain. Acceleration potential can occur e.g. as lead time reductions or as agile and quick responses. Logistics clusters can make it easier for companies to gain an increase in innovation ability and productivity, and so enable a higher reactivity. The increase is being caused by a time advantage, which on the other hand results from a better product-market position, a connection of resources, core competencies and knowledge, and declining transactions cots within the cluster. Therefore Porters diamond model, which he used to transfer his own model of national advantages to the field of regional clusters, will be extended by an additional factor, that is to say time.

19.1 Introduction “The world seems to move faster” - this statement is not new, but it never felt stronger than today. For companies it implies that traditional business models encounter limitations and in addition a strong conflict occurs. Always being one innovation ahead to the competitor, securing immediate reactivity and a 100 percent stock availability are opposed to an “it-may-cost-nothing”-mentality and the findings of lean management. It is clear that it is harder for a single company, as a “lone standing fighter”, to meet these challenges of time and money. But cooperation in networks can offer Ralf Elbert University of Technology Berlin, Chair of Logistics Services and Transportation e-mail: [email protected] Robert Sch¨onberger University of Technology Darmstadt, Chair of Clusters & Value Chain e-mail: [email protected]

233

234

Ralf Elbert and Robert Sch¨onberger

an additional potential of responsiveness and flexibility in a long run. As a special form of inter-organisational cooperation, clusters 1 . were discussed intensively in recent years. Almost 20 years ago it was Michael E. Porter who transferred the term “clusters” into the managerial economics and connected it with competitive advantages and so set the focal point for the current enthusiasm on clusters. Manifold studies reveal and document that clusters are a source for success of corporations (see e.g., S¨olvell et al 2003; Rosenfeld 1997; Porter 1998; Roelandt and den Hertog 1999 and Porter 2003). So far, cluster development emerges as a legitimate instrument of economic policy - the economic potential of clusters receives attention from policy-makers at all political levels throughout Europe: The European Commission, national, regional as well as city governments are promoting regional clusters. On the European policy level see e.g. “The European Cluster Memorandum” from December 2007 that is supported by national and regional agencies for innovation and economic development and addressed to policy makers at the national and European levels. In Germany, the Federal Ministry of Education and Research spends 600 million Euros for the development of 15 clusters all over the country in the next years. Furthermore, there are cluster development programs on the federal state level e.g. in Bavaria (“Bayerische Clusterpolitik”), Hesse (“1. Clusterwettbewerb des Landes Hessen”) and North Rhine-Westphalia (“RegioCluster.NRW”) where cluster development is supported with financial means as well as training and consulting of cluster managers (see Haasis and Elbert, 2008, p. 21-22). Clusters as a specific inter-organisational form were also recognized by the logisticians immediately. They quickly realized - because of their general knowledge about inter-organisational cooperation e.g. in networks and supply chains - the opportunities of this kind of cooperation (see Pfohl and Trumpfheller, 2004, p.3). So it is no surprise that until today more than 40 logistics clusters have been established in Germany (see Elbert et al, 2009, p.65). To find out how logistics clusters can create time advantages for the involved companies, a case study in an explorative research design and on the basis of a text analysis points out, which statements are made by the logistics clusters regarding time and speed. This paper shows an examination of the websites of the 40 identified logistics clusters in Germany in order to detect passages, which refer to time and speed. First references will be discussed, whether potential of acceleration can be identified in the Porter diamond and which logistics concepts are made possible through regional cooperation and how logistics clusters enable - if speed is the determining factor - a reduction of lead times and a fast turning with comparably small costs in the new mega-hubs logistics clusters.

1

Often also described with the terms industrial districts (see Markusen (1996) and Marshall (1920), industry cluster (see Cortright, 2005, p.8), innovative milieus (see Franz, 1999, p. 112-114), hot spots (see Pouder and St. John, 1996, S. 1194), sticky places, regional innovation networks and hub-and-spoke-districts (see Markusen, 1996, p. 296-297))

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains

235

19.2 Competitive Advantages through Clusters In its broadest sense, a network can be thought as social, economic and/or political relations between individuals and organizations (see Schubert, 1994, p. 9). Value creation networks are characterized by cooperation between economic and legal independent companies in order to realize competitive advantages (see Jarillo 1988, p. 32 and Liebhart 2002, p. 113). A regional value creation network consists a large number of companies from the same region, which are tied to each other by cooperative and competitive horizontal and vertical links (see Ritsch, 2005, p. 28). This is what Porter defined as clusters: a geographic concentration of companies, specialized suppliers and service providers, as well as companies in related sectors and institutions, which are all connected in a special field of interaction. The characteristics of this connection could be a common branch and industry or related process technologies. According to Porter “the enduring competitive advantages in a global economy lie increasingly in local things - knowledge, relationships, and motivation that distant rivals cannot match” (see Porter, 1998, p. 78). It is again Porter, who analyzes clusters the first time from its competitive strategy framework and describes them as a superior economic structure. Clusters affect competition in three broad ways: they improve the productivity of cluster companies, increase their innovation capability and stimulate the creation of new demand. Porter explains these effects with environmental influences that converge in six attributes that have the greatest influence on a company’s ability to innovate and upgrade. These attributes, which he terms diamond, “shape the information firms have available to perceive opportunities, the pool of inputs, skills and knowledge they can draw on, the goals that condition investment, and the pressures in firms to act” (see Porter, 1990, p. 111). His diamond model represented in the illustration below covers the following four determinants: The factor conditions describe the position of a region concerning the availability of production factors. This covers all relevant factors for an industry such as work force, raw materials and services. The demand conditions describe the kind of regional demand for products or services of an industry. The related and supporting industries mark the presence of international competitive related industries or suppliers. The firm strategy, structure and rivalry determine finally the conditions of a cluster, how companies are organized and led, as they cooperate and how the regional competition looks like. As further two determinants the chance and the government were later added, which can shape the regional competition sustainable and give at the same time important impulses for the cluster development. The development of clusters is either founded by particularly favourable conditions in one of the four original determinants of the diamond model or it can be released by acts of business, which cannot be traced to special local conditions. The further development of clusters explains Porter with a reciprocal process, which runs between the determinants of the diamond model. He understands the diamond as a strengthening model, in which a positive development in one determinant leads to an improvement of the other competitive conditions. He assumes that a certain number of participants in the cluster - a critical mass - must be achieved, thus the

236

Ralf Elbert and Robert Sch¨onberger

Fig. 19.1 The Porter diamond model (source:Porter 1990, p.72)

interactions in the cluster initiate an inherent dynamism and automatically lead to a settlement of specialized subcontracting firms, special training and research institutions, as well as to a further development of the infrastructure (see Porter, 1990, p. 216). The mostly macro-oriented analysis of Porter and others is very important in order to set up the topic ‘cluster’ in the mind of scientists, politicians and company leaders. But these mostly economically and economically-geographically driven studies do not explain how the success impact of clusters can be generated. Even if regional agglomerations and also clusters in various industries have been a research subject for decades, it was until now no focus on how regional value chains in clusters speed up global supply chains. Bringing clusters and logistics together, it is definite to say that logistics clusters are clusters, where logistics services are the central attribute of the connection between the participants (see Elbert et al, 2008, p. 315). Following the argumentation of Porter, logistics clusters can be described as concentrations of logistics companies, specialized suppliers for logistics companies, logistics service providers, firms in logistics-intensive industries as well as public and private institutions related to logistics, which can compete and cooperate with one another. The term logistics company covers both, the logistics industry (engineering and plants for the intralogistics and material handling, software producer for logistic applications and system integrators, etc.) and the logistics service providers (players like contract logistics, shipping companies, warehouse operators, IT service providers, consulting firms, etc.). Referring to the related devices, research institutions with an emphasis on logistics as well as logistics associations and logistics-relevant institutes for standardization could be named as potential cluster participants. Depending upon the development of a logistics cluster, vertical interdependences by supplier-customer relations on the one hand and horizontal interdependences by companies on the same stage in the value chain on the other hand can occur (see Elbert et al, 2008, p. 316).

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains

237

19.3 Research Framework In section 19.2 it was shown that clusters can generate competitive advantages regarding the Porter diamond. In order to deduce the emergence of these competitive advantages, it is required to analyze the sources of value creation in clusters, where basically three categories can be identified: “product-market position”, “resources, core competencies and knowledge”, as well as “transaction costs ” (see Elbert et al, 2009, p. 63).

19.3.1 Product-Market Position According to the market-based view value chains in clusters allow on the one hand bundling of existing regional product-market positions and on the other hand the development of new product-market positions (see M¨oller, 2006, p. 40). By pooling of already developed markets, companies can profit among themselves of the market positions of the cooperation partners and develop individually new markets on their own. For companies without network participation such a market entrance is much more difficult and becomes more time and cost intensive. Particular the bundling of financial and organisational resources within networks enables beyond that an accelerated and for the single firm more economical development of new markets (see Wrona and Schell, 2003, p. 320). In addition regional cooperation with already established competitors can reduce the existing rivalry, whereby economic rents can increase temporarily (see Zahn and Foschiani, 2002, p. 271).

19.3.2 Resources, Core Competencies and Knowledge According to the resource-based view (see Penrose, 1959; Prahalad and Hamel, 1990; Barney, 1991; Amit and Schoemaker, 1993). and its enhancements the creation of value in networks is induced through bundling and generation of resources, of core competencies and of knowledge. The aggregation and composition of complementary resources enables the development of individually strengths as well as the reconciliation of existing weaknesses. At the same time network-specific intangible and only with difficulties imitable knowledge can be generated as new core competency, which leads to a competence-based ability to innovate by the involved companies (see Duschek, 2002, p. 172). The capacity for innovation of the companies represents a central competitive advantage, which positively affects the value creation within the network. Regarding the knowledge which can be developed, the bundling in a regional network enables also the research and development within specialized areas and reduces thereby the risk for each individual company. Beyond that network-specific value creation results in particular from the transfer and the enhancement of knowledge as well as learning from and with the partners in the

238

Ralf Elbert and Robert Sch¨onberger

cluster (see Zahn and Foschiani, 2002, p. 271). The bundling of the existing and the generation of new knowledge is based on the confidence between the partners as well as o the existing competence in relationships as another central core competency (see Wrona and Schell, 2003, p. 320).

19.3.3 Transactions Costs According to the transaction cost theory different kinds of transaction costs can be affected by cooperation in regional networks (see Woratschek and Roth, 2003, p. 156). Cost advantages arise in networks in particular as a result of scale effects, which cannot be realized individually by a single company (see Zahn and Foschiani, 2002, p. 270-271). From the transaction cost theory view the costs per transaction can be reduced by investments in relational capital, since repetitive transactions between a small group of regional network participants are reducing the initiation and arrangement costs; in addition scale and scope effects with a rising contract volume in the region can be obtained and the average completion costs can be reduced. An extensive exchange of information leads to a reduction of the information asymmetries and therewith connected a reduction of the control costs in the region (see Dyer, 1997, p. 543-544). However these advantages do not result automatically regarding the transaction costs. Instead inter-organisational cooperation can require on the one hand a stronger coordination and organization of the activities. On the other hand a lack of confidence and reputation of the companies can lead to opportunistic behaviour in the regional network and requiring thus higher safety precautions (see Williamson, 1991, p. 291). In both cases higher transaction costs would result from cooperation in a regional network. As it is obvious from the discussion above value creation within clusters is based on two mechanisms. First, clusters can improve firm’s productivity and second, increase their innovation capability. Both allow companies to produce a superior output with similar or lower costs thereby improving their competitive position. Whereas agglomerations lead to shared commonalities across companies they still act independently in the market place. It is only through cooperation that companies engage in joint regional value creation systems, in which an independent transformation process takes place. The companies’ input, consisting of joint configuration of value activities as well as their combined resources and capabilities, are transformed by the reinforcing effects of the Porter diamond and lead to upgraded products and innovations. The following figure illustrates how clusters - starting from a simple agglomeration of companies - can be a source of innovation and productivity through cooperation, in combination with an activating cluster management. The Porter diamond reinforces the underlying sources of competitive advantages leading to superior productivity and enhanced innovation capabilities of its related companies. Time as determining factor is being added as additional advantage.

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains

239

Fig. 19.2 Time advantages on the way to innovation and productivity

Thus, it becomes evident how a cluster generates time advantages and so acceleration potential for the involved companies. Only remains the transmission on logistics and the empirical confirmation that it is possible to gain time advantages through clusters in the global supply chain.

19.4 Empirical Analysis In a really short time more than 40 logistics clusters were established in Germany. A deeper contemplation shows that all sizes, ages and forms of organization can be found within the logistics clusters. On the one hand there are big and established ones, which are supported by ministries or regional governments; on the other hand there are small and young logistics clusters, which - it seems like that - are still searching for their role in the global supply chain. But most important is that it does not look like that logistics clusters are only a phenomenon, which is going to disappear during the next years. Since most logistics clusters have a long-ranging business model, it is to assume that their positions will even strength. The logistics clusters which can be identified in Germany and were analyzed are put together in the Appendix.

240

Ralf Elbert and Robert Sch¨onberger

If logistics clusters will be relevant players in the supply chain it is important to know, what kind of advantages companies are effectively able to gain. As shown in section 19.3 the Porter diamond enables to realize time advantages through cluster activities. Although most logistics clusters are young and new actors in the supply chain, all forty clusters have web pages and an internet presence. Therefore it is reasonable to use this form of publication for a desktop research and to find out more about what clusters think about their own acceleration potential in achieving innovations and higher productivity. Several logistics clusters are obviously working on questions of speed, agile and quick responses, as well as reactivity. Most clusters reference on: • Infrastructure and the possibilities on working on an improvement together within the cluster to achieve faster movements. • Cooperation with customs to reduce latency and to increase the passage. • Local traffic jams warning systems to reduce waiting times and to reduce energy usage, environment pollution and traffic transport and the negative effects of traffic. • Increase of transparency to reduce contact times. • Building up a freight exchange to increase the productivity of all carriers. • Reducing throughput times through area-wide, intelligent, traffic control systems to reduce the risk of traffic jams in the region. • Connecting the regional players through regional networking and formulation and implementation of joint logistics projects to speed up the ability to react within the cluster and the region. • Making it easier to adduce logistics services by special infrastructural offers and to be able to innovate. • Capturing, structuring and operationalization of the term flexibility to create the basis for advantages for small and medium sized businesses and increasing their productivity. • Knowledge exchange in several ways to accelerate the knowledge spill-over. • Establishing ideas on how to segment customers by lead-times to increase the reactivity. • Promotion and acceleration of research and development to transfer creative impulses into marketable logistics products and services. Figure 19.2 shows, how, on the basis of an expedient “Product-Market Position”, a beneficial combination of “Resources, Core Competencies and Knowledge” as well a reduction of the “Transactions Costs”, regional logistics clusters can generate time advantages. This model therefore sets direct connections from the basis up to time advantages and the increase of innovation and higher productivity. A look into the logistics clusters practice brings up concrete examples for the importance of these connections: • Product-Market Position → Innovation & Productivity In many logistics clusters - for example in the logistics net Berlin Brandenburg or in the competence centre logistics Bremen - the cluster participants bundle their

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains

241

interests also for the development and for the use of existing industrial real estate areas and promote thus the settlement of new companies, as well as the growth of the existing companies and the entire region. In the same time they improve the development of the labor force by common education and vocational trainings, or by the common investment into the infrastructure and enable so their development and a better extent of the utilization. The activities mentioned affect on the one hand the factor conditions, i.e. the existing human, principal and natural resources. On the other hand the settlement of new companies affects the demand conditions and the competition between the logistics cluster participants. Beyond that some logistics clusters concentrate on the networking between the logistics industry and related, supporting industries. A very good example for this is the logistics cluster metropolis Ruhr, which wants to lead the Ruhr region by the interdisciplinary connection of logistics and IT “from the traffic-technical center” to the “information-logistics center of Europe”: A regional bundling of the available as well as a development of new product-market positions. • Resources, Core Competencies and Knowledge→ Innovation & Productivity The knowledge transfer between the cluster participants - as basis, in order to strengthen and develop common strengths - represents a frequently specified goal by the analyzed logistics clusters. In this coherence interdisciplinary cooperation is aimed between companies, universities and research institutions, as for example “Bavaria innovatively: Cluster logistics” particularly emphasizes as an activity, for example. So especially for small and medium-size companies it should be easier to get in touch with research institutions. The opportunities for knowledge transfer are offered by workshops, lectures, working groups and further meetings. Common core competencies are developed, if the cluster selects a certain topic within logistics to create a basis for competence-based innovation ability. As example the intralogistics network in Baden-Wrttemberg could be named. • Transactions Costs → Productivity The majority of the current activities of the logistics clusters, like e.g. newsletter mailings, the implementation of an internet portal as well as the organization of different events - as for example the logistics cluster North-Rhine/Westphalia or the network logistics Leipzig are doing - have the goal to create trust between the logistics clusters participants. These activities are to be rated as investment into the relations capital and form thus the basis for cooperation in the logistics clusters. Beyond that the information transfer also enables the networking of economics, science, research and politics. Proximity is created by the exchange between the companies, which leads to a reduction of the initiation and arrangement costs, for example. The arising needs of coordination and organization make an own cluster office necessary. Nearly two thirds of the observed logistics clusters already have their own cluster office; an example for this is the logistics network Thuringia. The discussion shows that several ways are gone within the logistics clusters to create a basis for cooperation and for development of requirements in the region to accelerate the logistics services. Nowadays most issues related to an acceleration potential, and named on the websites, are an improvement in infrastructure to

242

Ralf Elbert and Robert Sch¨onberger

speed up traffic and action in the logistics clusters area. But especially joint research and development activities show the intension to create new logistics services. And bringing the local logistics actors together in working groups or in discussion forums is a chance to exchange knowledge and - even more - establishes the basis for cooperation. Because people are close together and using the chance of short distance, a time advantage leads into faster innovations and higher productivity.

19.5 Further Research and Outlook With a description of the sources of value creation and the combination with the self-strengthening effects of the Porter diamond it was possible to show how clusters can create a time advantage on the way to an increase of innovation and higher productivity. Short ranges, fast contacts and response times, as well as the knowledge spill-over in a cluster lead into the time advantages and into the fact that redundant resources can be reduced. The web-pages of the 40 identified logistics clusters in Germany give an impression what kind of acceleration potential lies in the regional networks. Since in most regions the logistics clusters are relatively new, the results of the cluster work are not measurable yet. So for future research it is necessary to keep an eye on the cluster development and to follow up what they are doing in the region concerning time and speed. If logistics clusters are succeeding to dismount their position in the supply chain as new mega hubs, the question, as well, is taking into account whether such regional networks can activate additional adding value and whether in the future the production will follow logistics.

19.6 Appendix Logistics Clusters in Germany • • • • • • • • • • • • • •

Arbeitsgruppe Logistik Arbeitskreis Logistik Oldenburg Cluster Nutzfahrzeuge Schwaben (CNS) Cluster-Offensive Bayern: Cluster Logistik CNA Center for Transportation & Logistics Neuer Adler e.V. Cross Border Logistics CBLog FAV (Forschungs- und Anwendungsverbund Verkehrssystemtechnik) Berlin ForLog Gesamtzentrum fr Verkehr Braunschweig e.V. (GZVB) GVZ (Gterverkehrszentrum) Augsburg Intralogistik-Netzwerk in Baden-Wrtemberg e.V. KLOK Kooperationszentrum Logistik e.V. Kompetenzzentrum Logistik Bremen KTMC - Kompetenzzentrum Telematik . Mobile Computing . Customer Care

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains • • • • • • • • • • • • • • • • • • • • • • • • • •

243

Last Mile Logistik Netzwerk GmbH LOGIS.NET Logistics-Ruhr.com Logistik in Schwaben Logistik Initiative Rhein-Erft Logistik Netzwerk Thringen e.V. Logistik RheinMain Logistikagentur Oberfranken Logistikcluster “LogistikLand NRW” Logistik-Image-Kampagne Baden-W¨urttemberg Logistik-Initiative Baden-W¨urttemberg Logistik-Initiative Duisburg/Niederrhein Logistikinitiative Emsland Logistik-Initiative Greven Logistik-Initiative Hamburg Logistikinitiative Log-I SH Logistikinitiative Mecklenburg-Vorpommern Logistikinitiative Niedersachsen LogistikNetz Bayerischer Untermain LogistikNetz Berlin-Brandenburg LogistikRuhr Logport77 logRegio L¨ubeck MoWiN.net - Mobilittswirtschaft in Nordhessen Netzwerk Logistik Leipzig/Halle Research Cluster for Dynamics in Logistics

References Amit R, Schoemaker PJH (1993) Strategic Assets and Organizational Rent. Strategic Management Journal 14(1):33–46 Barney J (1991) Firm Resources and Sustained Competitive Advantage. Journal of Management 17(1):99 Cortright J (2005) Making Sense of Clusters Regional Competition and Economic Development. Last check 30.07.2008, URL http://www.brook.edu/metro/pubs/20060313 Clusters.pdf. Duschek S (2002) Innovation in Netzwerken: Renten-Relationen-Regeln. Wiesbaden Dyer JH (1997) Effective Interfirm Collaboration: How Firms Minimize Transaction Costs and Maximize Transaction Value. Strategic Management Journal 18(7):535–556 Elbert R, Sch¨onberger R, M¨uller F (2008) Regionale Gestaltungsfelder f¨ur robuste und sichere globale Logistiksysteme. Strategien zur Vermeidung, Reduzierung und Beherrschung von Risiken durch Logistik-Cluster. In: Pfohl HC, Wimmer T (eds) Wissenschaft und Praxis im Dialog. Robuste und sichere Logistiksysteme, Hamburg, pp 294–322.

244

Ralf Elbert and Robert Sch¨onberger

Elbert R, Sch¨onberger R, Tschischke T (2009) Wettbewerbsvorteile durch LogistikCluster. In: Wolf-Kluthausen H (ed) Jahrbuch Logistik 2009, Korschenbroich, pp 61–67. Franz P (1999) Innovative Milieus: Extrempunkte der Interpenetration von Wirtschafts- und Wissenschaftssystem. Jahrbuch fr Regionalwissenschaft 19:107–130 Haasis HD, Elbert R (2008) Bringing regional networks back-into global supply chains: Strategies for logistics service providers as integrators of logistics clusters. In: Kersten W, Blecker T, Fl¨amming H (eds) Global Logistics Management,, Berlin, pp 21–31 Jarillo JC (1988) On Strategic Networks. Strategic Management Journal 9(1):31–41 Liebhart UE (2002) Strategische Kooperationsnetzwerke. Entwicklung, Gestaltung und Steuerung. Wiesbaden Markusen A (1996) Sticky Places in Slippery Space: A Typology of Industrial Districts. Economic Geography 72(3):293–313 Marshall A (1920) Principles of Economics. Lodon M¨oller K (2006) Wertsch¨opfung in Netzwerken. Mnchen Penrose ET (1959) The Theory of the Growth of the Firm. Oxford Pfohl HC, Trumpfheller M (2004) Das ATHENE-Projekt. Auf dem Weg zur Netzkompetenz in Supply Chains. In: Pfohl HC (ed) Netzkompetenz in Supply Chains. Grundlagen und Umsetzung, Wiesbaden, pp 1–10 Porter ME (1990) The Competitive Advantage of Nations. London Porter ME (1998) Clusters and Competition: New Agendas for Companies, Governments, and Institutions. In: Porter ME (ed) On Competition, Boston Porter ME (2003) The Economic Performance of Regions. Regional Studies 37(6/7):549–578 Pouder R, St John CH (1996) Hot spots and blind spots: Geographical clusters of firms and innovation. Academy of Management Review 21(4):1192 – 1225 Prahalad CK, Hamel G (1990) The Core Competence of the Corporation. Harvard Business Review 68(3):79 – 91 Ritsch K (2005) Wissensorientierte Gestaltung von Wertsch¨opfungsnetzwerken. Aachen Roelandt T, den Hertog P (1999) Cluster Analysis and Cluster-Based Policy Making in OECD Countries: An Introduction to the Theme, Ch 1. In: OECD (ed) Boosting Innovation: The Cluster Approach, Paris, pp 9–23 Rosenfeld SA (1997) Bringing business clusters into the mainstream of economic development. European Planning Studies 5(1):1–15 Schubert K (1994) Netzwerke und Netzwerkans¨atze: Leistungen und Grenzen eines sozialwissenschaftlichen Konzeptes. In: Kleinaltenkamp M, Schubert K (eds) Netzwerkans¨atze im Business-to-Business-Marketing. Beschaffung, Absatz und Implementierung Neuer Technologien, Wiesbaden, pp 8–49 S¨olvell O, Lindqvist G, Ketels C (2003) The Cluster Initiative Greenbook. Stockholm Williamson OE (1991) Comparative Economic Organization: The Analysis of Discrete Structural Alternatives. Administrative Science Quarterly 36(2):269 – 296

19 Logistics Clusters - How Regional Value Chains Speed Up Global Supply Chains

245

Woratschek H, Roth S (2003) Kooperation: Erkl¨arungsperspektive der Neuen Institutionenkonomik. In: Zentes J, Swoboda B, Morschett D (eds) Kooperationen, Allianzen und Netzwerke. Grundlagen Ans¨atze Perspektiven, Wiesbaden, pp 141–166 Wrona T, Schell H (2003) Globalisierungsbetroffenheit von Unternehmen und die Potenziale der Kooperation. In: Zentes J, Swoboda B, Morschett D (eds) Kooperationen, Allianzen und Netzwerke. Grundlagen Ans¨atze Perspektiven, Wiesbaden, pp 305–332. Zahn E, Foschiani S (2002) Wertgenerierung in Netzwerken. In: Alback H, Kaluza B, Kersten W (eds) Wertsch¨opfungsmanagement als Kernkompetenz, Wiesbaden, pp 265–275

Part IV

Survey and Longitudinal Research

Chapter 20

Measuring the Effects of Improvements in Operations Management Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

Abstract This paper examines the relationship between a managerial focus on reducing inventory and improvements in value added. We analyze financial information on large non-service US based firms over the 25 year period from 1980 to 2004. Our results show a very strong correlation between the increase in value added and the decrease in days of inventory speed across all manufacturing industries. The results strongly support the operations management literature which claims a managerial focus on efficiency, in particular increasing the speed of operations, will result in significant value creation for firms. The results also imply that the concept of competition based on operational speed has not been transferred across all firms and the potential for improvement still exits in most industries.

20.1 Introduction Operational speed, defined as the lead time from order handling, through production and delivery, to the customer, has long been recognized as one of the common characteristics of successful companies in competitive business environments. All operational management methods to improve operations are supposed to make processes Vedran Capkun HEC School of Management, 1, rue de la Liberation, 78351 Jouy-en-Josas cedex, France, tel. +331-39-67-96-11 fax. 70-86 e-mail: [email protected] Ari-Pekka Hameri, corresponding author Ecole des HEC, University of Lausanne, Internef, Lausanne 1015 Switzerland, tel +41 21 692 3460 fax 3495 e-mail: [email protected] Lawrence A. Weiss McDonough School of Business, Georgetown University, Old North G01A, Washington, DC 20057-1147, USA, tel. +1-202-687-3802 fax. 4031 e-mail: [email protected]

249

250

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

faster, more controllable and accurate. This includes business process engineering, total quality management, vendor managed inventories, supply chain integration, just-in-time, lean thinking, and activity based management. Among the best known firms which focus on operational speed are the computer assembler Dell and the apparel company Zara. These companies avoid the perilous impact of supply chain dynamics by operating at speeds where the capital bound by their operations is a fraction of the overall volume of their business. This provides them with the agility to react to sudden demand variations and outperform their competitors. These companies are also less dependent on forecasts and preplanned operations giving them an additional advantage and cost efficiency over their slower competitors. Schmenner, 1988; Stalk and Hout, 1990; Womack and Jones, 1996, all find a strong positive relationship between financial results and those firms who set operational speed as their key strategic approach. According to the operations management literature, each operation that is part of a business process should add to the value of the end-product. The more efficiently a company creates value which customers are willing to pay for, the greater the firm’s ability to deliver value to its stakeholders. One of the first steps to increasing the value creation capability of a process is to remove those operations that do not add value (in the sense of providing something a customer is willing to pay for). Essentially a process should create value without bottlenecks and the process variability, inherent or external, should be minimized (see Schmenner and Swink, 1998). Whether the term used is swiftness, operational speed, or reduced days of supply, the aim is to improve operational speed by reducing the lead times of the value creation processes of the company. The majority of success stories in operations management stem from automotive, machinery and job-shop (i.e. assembly) industries. It is important to review how just-in-time and other operational methodologies have affected these and other industries since their introduction in early 1980’s. The paper begins by presenting the relevant literature on lead time reduction and competition based on speed. This is followed by a description of the research hypothesis, sample description, and applied methodologies. Then we document the relationship between value creation and operational speed by using financial information from large U.S. companies. Next, we illustrate how development in operational speed has taken place in general and across different industries. Finally, we present a summary of our key findings with their managerial implications.

20.2 Literature Review Manufacturing processes have come a long way since Henry Ford’s moving assembly line and mass production. Ford’s focus was on output and cycle time, however, he also provided examples of how repeating tasks improve lead time along the traditional learning curve - like the case of disassembling excess war ships after the First World War (Ford, 1922). Scale and cost centric manufacturing were a managerial

20 Measuring the Effects of Improvements in Operations Management

251

focus until the 1970s. Then, the quality movement turned the focus to continuous improvement and errorless operations. By the end of the 1970s, quality management had become entrenched in manufacturing operations. The arrival of Just-in-Time (JIT), and its emphasis on waste, inventory reduction, and operational flexibility, created a managerial focus on operational speed in the early 1980’s. Schonberger (1982), in one of the first books on JIT, describes the benefits stemming from reduced set-up times and smaller lot-sizes. He also warns of the perils of monster machines and excessive investments in technological marvels which require near 100% utilization levels to justify the financial investment. Hall (1983) and Monden (1983) document the increase in efficiency obtained by Japanese production facilities. All these books use examples from car assemblers and their suppliers with a few references to other machine assembly industries. Thus, the initial focus of the early JIT literature was on a relatively repetitive type of production with assembly of complex products. This trend continued with other supportive studies reporting on job shops with huge product variety, and numerous different operations and product routings. Goldratt and Cox (1984) and Suri (1994, 1998) refined the managerial focus to a relentless reduction of bottlenecks and lead time. These approaches, theory of constraints and quick response manufacturing, rooted their message on flow and lead time reduction by representing a plethora of cases from job shops and machine assembly companies. The same message was being delivered by other scholars and practitioners under different labels like time based competition (Stalk, 1988; Stalk and Hout, 1990) and lean manufacturing (Womack et al, 1990; Womack and Jones, 1996). Little (1961) shows that speed is directly proportional to average inventory. This means speed will increase with inventory reductions1. Schmenner (2001) provides an overall perspective on the history of manufacturing and operations management with the simple phrase “swift, even flow”. He argues that the operational management emphasis should be an expeditious, well controlled, flow of material through various value adding operations without bottlenecks and excess variability. According to Schmenner (2001), throughout history, companies that focused on flow with an emphasis on speed and variability reduction have outperformed companies emphasizing other goals. This is consistent with the mathematical principles of operations management, based on queuing theory, which demonstrates the relationships between lot sizes, bottlenecks, lead times, and process variability (see Hopp and Spearman, 1996). Despite the sound mathematical grounds and numerous documented cases, speed based competition is still principally rooted in high volume and repetitive industries. The flagship of this trend remains the automotive industry, followed by job shops assembling machinery. Studies like Holmstr¨om (1995) and Hameri and Lehtonen (2001) indicate that most industries are dormant when it comes to operational speed. Some industries, like pulp and paper, have improved their productivity by investing in automation and through vertical integration including strong merger and acqui1

Unless of course there is a supply chain glitch, which would then cause a decline in speed. See Hendricks and Singhal (2003)

252

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

sition strategies (Hameri and Weiss, 2006). The Internet has also pushed operational speed into a new focus over the past few years. Firms now use the internet to improve information transparency across organizational boundaries. Piszczalski (2000) and Bruun and Mefford (2004) provide examples of how new communication technologies speed-up operations and reduce mistakes. They show how completely new business models based on automatic order handling and verification have emerged. There are some anecdotal studies on JIT and speed based competition in services (Barlow, 2002) and other manufacturing industries. Unfortunately, these are few and sporadic. The vast majority of the examples of speed based competition are based on industries related to automobiles, assembly of machinery, and computer equipment. Over half a century ago Forrester’s (1961) non-linear simulations on information and delivery delays in supply chains helped academics and managers understand how information distortion and order batching leads to ever longer lead times and inventory build up. JIT emerged in the early 1980s and brought supplier relations to the forefront. Activity Based Costing systems arrived a few years later, in the mid 1980’s. These systems allowed managers to quantify the cost of resource demands and operational processes. This provided managers with an improved understanding of the underlying economics of their operations and allowed them to make more informed decisions on their product and process choices (see Kaplan and Cooper, 1998). In the early 1990’s, the focus on the entire supply chain (from elementary suppliers to the end customer) gained extensive momentum. Today, supply chain management is a corner stone of modern operations management and supply chain structures and control principles are the subject of extensive academic research (e.g. Houlihan, 1987; Fisher and Raman, 1996; Fisher, 1997; Frohlich and Westbrook, 2002). As noted above, speed and lead time related research has focused principally on the automotive, machinery, and computer assembly operations. By contrast, supply chain research extends to all industries. It reviews supply chain structures with distributions centers and various forwarding parties included in the chain. The underlying theme in supply chain management is on information transparency, reliable lead times, and the clever positioning of various value adding operations in the long logistical chains. Companies with efficient in-house operations are also more likely to display competence in managing their supply chains. To summarize, operational speed, defined as the lead time from order handling, through production and delivery, to the customer, has been recognized as one of the common characteristics of successful companies in competitive business environments. Most of the research in the field of operational speed documents the numerous cases where lead time reduction has resulted in major advantages for the company. The vast majority of this research concerns industries where JIT and speed based competition was first introduced - namely automotive, job shops, and electronic equipment. Surprisingly, there are no studies which review different industries and their development over longer periods of time. Our paper aims to fill this gap by focusing on the longitudinal development in the reduction of operational speed (defined as days of supply) and the related increase in value creation.

20 Measuring the Effects of Improvements in Operations Management

253

20.3 Research Questions, Sample Selection, and Methodology This paper examines the relationship between value creation and three key elements ascribed to a managerial focus on operational efficiency: days of supply, new investments in plant and equipment, and expenditures on research and development. The key findings of the operations management literature indicates changes in value added should be at least partially explained by these three key variables, which defines the model to be tested in this study (Fig. 20.1). Following the literature review we set our research questions as: 1. What is the impact of a managerial focus on operations on value creation? 2. How does this link vary across industries? and 3. Is there a stronger correlation among industries that were early adopters of speed based operations strategies?

Fig. 20.1 The underlying model of the study: How capital expenditures, days of supply and R&D costs (our three independent variables), correlate with increases in value added (our dependent variable)

The Thomson Financial database is used to retrieve the Standardized Industrial Classification (SIC) codes of the firms in our sample and we use these codes to split our sample into two sub-samples - manufacturing firms and non-manufacturing firms. Table 1 documents the classification of firms based on their SIC codes. We begin with 2,160 non-financial US based firms with assets in excess of US$ 100 million. On average, firms in our sample have total assets of $ 4.2 billion with a median of $ 783 million. The range includes firms with assets from $ 100 million to $462 billion. The sample is then divided into 1,166 manufacturing firms (SIC codes from 20-39) and 994 non-manufacturing firms (SIC codes from 40-59). Manufacturing firms are significantly smaller than the non-manufacturing firms. The sample

254

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

of the manufacturing firms has a mean in total assets of $ 3.9 billion (median of $ 646 million) compared to $4.6 billion (median $1.0 billion) for the sample of nonmanufacturing firms. To compute the change in our variables (as defined below) we require at least two consecutive periods and exclude data without a consecutive period. We also remove negative value added data from the sample. Our final samples consist of 915 firms and 10,882 observations (an average of 12 periods per firm) for the regressions 1 & 2, and 927 firms with 11,244 observations (on average 12 periods per firm) for regression 3, over the 25 year period 1980-2004. Table 20.1 SIC industry classification codes for firms used in our sample. Our sample consists of firms with SIC primary codes from 01-59 01.09 10.14 15-17 20-39 40-49 50-51 52-59 60-67 70-89 91-99

Agriculture, Forestry, And Fishing Mining Construction Manufacturing Transportation, Communications, Electric, Gas, And Sanitary Services Wholesale Trade Retail Trade Finances Services Public Administration

We collect data on our dependent variable and the three independent variables from the annual report of each firm over the 25 year period 1980-1994. Our variables are defined as follows: (1)

VAD = (Sales-COGS) / Number of Employees

VAD is the value added defined as gross profit per employee. Sales are annual sales taken from the firm’s annual income statement. COGS are annual costs of goods sold taken from the firm’s annual income statement. Number of employees is the year end total number of employees as reported in annual report. (2)

DoS = (End of the year inventory / COGS) × 365

DoS is the days of supply. End of the year inventory is taken from the firm’s balance sheet. COGS are the annual costs of goods sold taken from the firm’s annual income statement. A reduction in DoS proxies for improvements in operations and should lead to an increase in VAD. (3)

CAPEX = net capital expenditures on new investments in long term assets reported on the cash flow statement.

20 Measuring the Effects of Improvements in Operations Management

255

CAPEX is the capital expenditure on long term assets. An increase in CAPEX indicates new investments for operations which should lead to an increase in VAD. (4)

R&D = research and development expenses reported on the income statement.

R&D is the annual expenditure on research and development. An increase in R&D includes improvements in operating processes which should lead to an increase in VAD. We specify our regression as follows: (5)

Δ ln (VAD) = Δ ln (DoS) + Δ ln (CAPEX) + Δ ln (RD) + exp

2

We measure the percentage change in value added per employee in relationship to the percentage change in speed of inventory, investments, and research and development. We use the fixed effects regression to analyze the relationship between the variables3 . This method was originally developed and used by economists and has gained broad acceptance and use throughout the social sciences. The fixed effects method eliminates major biases from regression models with multiple observations, especially longitudinal ones as we have here, for each subject. There is a direct link between inventory reduction and gross margin; however, this link is very small. As inventory is reduced there will be profit improvements due to interest savings as well as a reduction in storage fees, handling, and waste. These savings have been estimated by the literature to be on the order of 20-30% (see Brigham and Gapenski, 1993). The interest savings are the reduction in inventory times the firms cost of capital. This will have no effect on gross margin as interest revenue or a reduction in interest expense is reported after operating profits. This leaves an impact of approximately 20% in other savings which could appear as an improvement in gross margin. The impact on value added (gross margin/# of employees) will be on the order of 2%.

20.4 Empirical Analysis, Results and Discussion We begin our examination of the link between days of supply, capital expenditures and R&D costs with value creation across all firms and then separate our sample between manufacturing and non-manufacturing firms. Table 20.2 shows the results 2 Specifying the regression differently does not change our conclusions. Using CAPEX/Total Assets and RD/Sales instead of changes in CAPEX and RD does not change our conclusions nor does using absolute values of VAD and DoS. 3 Our conclusions do not change if we use random effects regression or if we lag value added. Using other definitions of days supply and value added also does not change our conclusions nor does using absolute values of variables.

256

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

for the whole sample and the two sub-samples for the three regression models. The first model (regression 1) is the original model with value added per employee as the dependent variable. To test the sensitivity of our results to the choice of dependent variable and macroeconomic conditions, in the second model (regression 2), we adjust value added per employee for inflation resulting in value added per employee in 1980 US dollars. This adjustment does not change our conclusions. We run the third model (regression 3) with value added divided by total assets (instead of divided by employees) as the dependent variable. Our conclusions remain the same. For the whole sample there is a strong and statistically significant (1% level) link between our three independent variables and value added, regardless of the choice of regression model. A closer look shows that this holds only for manufacturing firms. For non-manufacturing firms, the only statistically significant coefficient is days of supply, and in most cases only at the 10% significance level. This was expected based both on prior research and common sense. Value creation in non-manufacturing operations is by default less related to our chosen independent variables. Many of the non-manufacturing firms have little or no inventory, and others having a constant inventory level, thus making their value added insensitive to changes in inventory management. The detailed results for non-manufacturing industries (industry analysis) are mixed and do not indicate a relationship between the dependent variable and any of the three independent variables4 . N is the number of firms in the sample and the sub-samples. Value Added is defined as (Sales - COGS)/Number of Employees. Days of Supply is defined as (End of the year inventory/COGS)×365. Capital expenditure is defined as the net capital expenditures on new investments in long term assets. R&D cost is defined as the research and development expenses. Values in parentheses are standard errors of the coefficients, while ***, ** and * represent significance levels of 1, 5, and 10% respectively. As noted above, for manufacturing firms the coefficient of speed is significant and negatively related to value added. The coefficients associated with both capital expenditures and R&D expenditures are positive and significant for manufacturing firms. These results are consistent with the argument that improving operational speed, increasing capital investments and/or increasing funding of R&D all lead to an increase in value added. The results are consistent with our model depicted in Fig. 20.1 for manufacturing companies. To analyze manufacturing industries in more depth, we further divide our sample of manufacturing firms into sub-samples based on the two first digits of their SIC code. We eliminate those industries in our sample with less than 20 firms. The results are presented in Table 20.3. While the coefficient associated with days supply remains negative and significant, the capital expenditure and R&D expenditure coefficients differ across industries. The coefficient of capital expenditure is significant and positive for primary metal, machinery and computer equipment, electronics, transportation equipment, and instruments. By contrast, an increase in capital expenditure does not seem to have any relationship to the value added increase in 4

We analyzed all non-manufacturing industries but we do not present the results in this paper. The results are not significant for days of supply in any of the analyzed non-manufacturing industries.

20 Measuring the Effects of Improvements in Operations Management

257

Table 20.2 The table presents fixed effects regression coefficients of the percentage increase in Value Added during the 25 year period 1980 to 2004 on the percentage change in days of supply, capital expenditures and research and development costs All firms Regression 1: (915 firms) Value added per Employee Regression 2: (915 firms) Value added per employee adjusted for inflation Regression 3: (927 firms) Value added per Total assets

Manufacturing Regression 1: (826 firms) Value added per Employee Regression 2: (826 firms) Value added per employee adjusted for inflation Regression 3: (838 firms) Value added per Total assets

Days of Supply

R&D costs

Capital Expenditures

-0.185*** -0.01

0.042*** -0.009

0.061*** -0.006

-0.184*** -0.01

0.042*** -0.009

0.061*** -0.006

-0.216*** -0.01

0.051*** -0.01

0.0214*** -0.006

Days of Supply

R&D costs

Capital Expenditures

-0.192*** -0.01

0.046*** -0.01

0.065*** -0.006

-0.191*** -0.011

0.045*** -0.01

0.064*** -0.006

-0.219*** -0.011

0.056*** -0.01

0.022*** -0.022

food, chemical products or fabricated metal industries. The coefficient of R&D expenses also differs across industries. It remains positive and significant for paper products, machinery and computer equipment, electronics, and instruments. However, it is not significant in food, primary metal, fabricated metal, or transportation equipment. All manufacturing industries with more than 20 firms were analyzed. Value Added is defined as (Sales - COGS)/Number of Employees. Days of Supply is defined as (End of the year inventory/COGS)×365. Capital expenditure is defined as the net capital expenditures on new investments in long term assets. R&D cost is defined as the research and development expenses. Values in parentheses are standard errors of the coefficients, while ***, ** and * represent significance levels of 1, 5, and 10% respectively. Capital expenditures appear to play a major role in capital intensive, some hightech and complex products producing industries. A similar pattern is found on the impact of increasing R&D investments. The underlying message seems to be that

258

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

Table 20.3 The table presents fixed effects regression coefficients of the percentage increase in Value Added during the 25 year period 1980 to 2004 on the percentage change in days of supply, capital expenditures and research and development costs SIC Industry name

# of firms

Days of Supply

Capital Expenditure

R&D costs

20 Food

21

28 Chemical and allied products

141

33 Primary metal

23

34 Fabricated metal

28

35 Machinery and allied products

124

36 Electronics

194

37 Transportation equipment

58

38 Instruments

136

-0.313*** (0.119) -0.110*** (0.025) -0.256** (0.108) -0.374*** (0.051) -0.184*** (0.022) -0.223*** (0.024) -0.425*** (0.034) -0.099*** (0.024)

-0.012 (0.060) 0.015 (0.016) 0.139*** (0.036) 0.019 (0.027) 0.054*** (0.013) 0.116*** (0.013) 0.042** (0.020) 0.047*** (0.012)

-0.062 (0.095) 0.0712** (0.0344) -0.107** (0.054) -0.007 (0.041) 0.077*** (0.027) 0.064** (0.027) 0.037* (0.022) 0.048*** (0.016)

competition in different industries has been increasing over the past two and half decades and survival has required some level of improved operational efficiency. We next track trends in the relationship between value added and days of supply by industry. As noted above, some managers of job-shop and assembly industries began shifting to JIT and other related methods to improve speed and efficiency in the early 1980s. To provide evidence on this shift in managerial focus, we compute industry medians of value added and days of supply for different industries. Figure 20.2 shows 4 different industries with their very different development profiles over the past 25 years. The transportation equipment industry (Fig. 20.2a), SIC 37) has reduced days of supply from around 100 days to 40 days, while at the same time the value added has almost doubled. The improvement in speed occurred primarily during the 1980s. This was the period when JIT and a focus on small lot sizes was introduced to the industry, and is clearly reflected in the graph. This period of speed improvement also witnessed a strong improvement in value creation. Lieberman and Demeester (1999) document, in their profound study on Japanese automakers, that each 10% reduction in inventory leads to approximately a 1% gain in labor productivity, with a lag of about one year. Although their result is from a limited specific sample, our sample of US companies in the transportation equipment industry also finds that a focus on speed results in improved value creation. A more detailed look at this industry reveals that those firms with large relative reductions in the days of supply have higher increases in value creation, regardless of the absolute starting level. Companies which are speedier but do not improve their speed are stagnant

20 Measuring the Effects of Improvements in Operations Management

259

Fig. 20.2 Inflation adjusted value added per employee and days of supply for four industries. Value Added, adjusted to 1980 US dollars, is defined as (Sales - COGS)/Number of Employees. Days of Supply is defined as (End of the year inventory/COGS)×365

260

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

in their value generation. This indicates the important element is to focus on speed improvement as opposed to having the lowest absolute days supply. Effectively, continuous improvement in speed appears to be as vital for valuation creation as quality management. The electronic equipment and component industry (Fig. 20.2b), SIC 36) shows a continuous increase in value added over our sample period. The median of value added increased three times while speed improved (was reduced) by 20%. This can perhaps be explained by a more in-depth examination of the sample. The SIC code 36 holds several different industries. The conventional electronics industries, like household appliances, operate in well established commodity type industries. These firms have maintained a fairly constant speed level throughout the 25 year period. The highly competitive telecommunication industries have had a major drive to improve their speed. Here, the top performing companies halved their speed and increased their value creation ten times. Next, we examine the 25th and 75th percentiles for the industrial machinery and computer equipment industries (SIC code 35, see Fig. 20.3). Here we expect a strong correlation between improvements in operational speed and value creation. The top companies of this industry have more than halved, from 160 to 75 days, their operational speed during the past 25 years, while their value added has increased more than three times. During the same period the worst performing companies have reduced their speed by 20% but their value added remained virtually unchanged with only a 10% increase in 25 years. We note that speed for the worst companies started on a significantly better level than the best performing companies. This means, that in relative terms, the speed improvement is very drastic among the top value creating companies. This gives strong support to the hypothesis that companies with a focus on the speed of their operations have a greater potential to increase value added. The most efficient firm in the machinery and computer equipment industry (SIC code 35) in 2004 was Dell with days of supply of only 4.15, followed by Apple Computers (6.19 days). Dell outperformed its competition in terms of speed in most years since it entered our sample in 1987, resulting in above average value added per employee of $ 73.000 in 2004 (in 1980 dollars). Figure 20.4 shows the speed and performance of Dell in the period of 1987-2004. Days of supply decreased from 80 to 4 while the firm’s value added per employee doubled from $ 36.000 to $ 73.000 (in constant 1980 dollars).

20.5 Summary and Conclusions This paper examines the relationship between a managerial focus on improvements in inventory reduction and value added. We analyze financial information on large non-service US based firms over the 25 year period from 1980 to 2004. Our results show a very strong correlation between the percentage increase in value added (defined as gross margin per employee) and the percentage reduction in days of supply

20 Measuring the Effects of Improvements in Operations Management

261

Fig. 20.3 Companies in SIC 35 belonging to top and bottom 25th percentiles when measured in value added and days of supply. Value Added is defined as (Sales - COGS)/Number of Employees. Days of Supply is defined as (End of the year inventory/COGS)× 365

across all manufacturing industries. The relationship between value added and the capital expenditure and R&D expenses is also strong with differences across industries based on the standard industrial classification codes. The longitudinal analysis shows that speed based competition was especially pronounced in the machine production, transportation, and computer equipment industries. Other industries also displayed improvements in value added but without a similar relationship to inventory. The results strongly support the operations management literature which claims a managerial focus on efficiency, in particular increases to the speed of operations, will result in significant value creation for firms. The results also imply that the concept of competition based on operational speed has not been transferred across all firms and the potential for improvement still exits in most industries. Our results are based on correlations, not on experimental research. This means our results do not prove causality between our variables. However, the many documented cases in the literature on positive turnarounds in manufacturing companies through the reduction of inventories and speeding operations serve as experimental research which through the manipulation of the concerned variables has strongly demonstrated the causal relationship between the variables. Our statistical analysis combined with the visualization of the time series clearly supports our hypotheses. There are several implications of the analyses which provide additional insight to the studied variables and their relationships. The following list summarizes the main

262

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

Fig. 20.4 Inflation adjusted value added per employee and days of supply for Dell Computer. Value Added, adjusted to 1980 US dollars, is defined as (Sales - COGS)/Number of Employees. Days of Supply is defined as (End of the year inventory/COGS) × 365. Dell’s value added per employee (in 1980 dollars) and days of supply in the 1980-2004 period

findings of our study and how manufacturing companies should perceive them in an operational sense: • Slow companies will eventually fail. The analyzed data demonstrate that companies which are unable to improve their operational speed will gradually loose their capability to produce value. Even the worst performing companies show progress in operational speed. Without this improvement they probably would not have been able to remain competitive. Improving operational speed appears to be vital for the survival of a firm, independent of the firm’s current level of performance. Naturally, companies with a monopolistic position are excluded from our discussion as they are removed from the competitive environment. The strong correlation between days of supply and value creation across all manufacturing industries should invite managers to emphasize operational speed and its reduction in their strategies. A managerial focus on operational speed should be perceived to be as important as ROI, market share, and profit. • Each industry sector has companies that exploit speed based operational superiority over others. Our data demonstrates that despite differences across industries, in each industry there are companies who enjoy larger value creation due to their speedier operations. Although some industries are competing based on

20 Measuring the Effects of Improvements in Operations Management

263

factors other than speed, there appears to be additional opportunities for firms to improve relative to the industry by focusing on expediting operations. • Slower companies, in general, have a higher potential to improve their value added, while already faster companies can further increase their lead with additional improvements in operational speed. Our analysis indicates that, in relative terms, all top percentile companies improved their speed in absolute terms relative to the lower percentile companies. However, in some industries the lower level companies improved at a faster rate than the top companies. This demonstrates the importance of both derivatives of speed improvement - the first is the importance of improving the absolute level, the second is the importance of improving the rate of change. As noted above, the savings from inventory costs related to reduced inventory levels have very little impact on COGS and the Gross Margin per employee. This means that companies who improve their speed are doing things better in general. They tap outsourcing, technology, or whatever new operational principles emerge faster than the other companies and incorporate the benefits more efficiently. Clearly, doing things faster is not simply having people work harder and having trucks move faster. Rather, it means a wide range of operations are being done in a different and more intelligent manner.

References Barlow G (2002) Just-in-Time: Implementation within the hotel industryA case study. International Journal of Production Economics 80(2):155–167 Brigham E, Gapenski L (1993) Intermediate financial management. The Dryden Press, New York Bruun P, Mefford R (2004) Lean production and the Internet. International Journal of Production Economics 89(3):247–260 Fisher M (1997) What is the right supply chain for your product? Harward Business Review 75(2):105–116 Fisher M, Raman A (1996) Reducing the cost of demand uncertainty through accurate response to early sales. Operations Research 44(1):87–99 Ford H (1922) My life and work. Ayer Company, Salem, New Hampshire Forrester J (1961) Industrial Dynamics. MIT Press, and John Wiley & Sons, Inc, New York Frohlich M, Westbrook R (2002) Demand chain management in manufacturing and services: Web-based integration, drivers and performance. Journal of Operations Management 20(6):729–745 Goldratt EM, Cox J (1984) The Goal. North River Press, Croton-on-Hudson, New York Hall R (1983) Zero inventories. Homewood, IL, Dow Jones-Irwin Hameri A, Lehtonen J (2001) Production and supply management strategies in Nordic paper mills. Scandinavian Journal of Management 17(3):379–396

264

Vedran Capkun, Ari-Pekka Hameri and Lawrence A. Weiss

Hameri A, Weiss L (2006) Value creation and days of supply in major pulp and paper companies, paper and Timber (forthcoming) Hendricks K, Singhal V (2003) The effect of supply chain glitches on shareholder wealth. Journal of Operations Management 21(5):501–522 Holmstr¨om J (1995) Speed and efficiencyA statistical enquiry of manufacturing industries. International Journal of Production Economics 39(3):185–191 Hopp W, Spearman M (1996) Factory Physics. Irwin, Chicago Houlihan J (1987) International supply chain management. International Journal of Physical Distribution and Materials Management 17(2):51–66 Kaplan R, Cooper R (1998) Cost and Effect: Using integrated cost systems to drive profitability and performance. Harvard Business School Press, Boston, Massachussets Lieberman M, Demeester L (1999) Inventory reduction and productivity growth: linkages in the Japanese automotive industry. Management Science 45(4):466– 485 Little J (1961) A proof for the queuing formula: L= λ W. Operations Research 9:383–387 Monden Y (1983) Toyota production system: An integrated approach to just-intime. Industrial Engineering and Management Press, Institute of Industrial Engineers, Norcross, GA Piszczalski E (2000) Lean vs. Information Systems. Automotive Manufacturing & Production 112(8):26–28 Schmenner R (1988) The merit of making things fast. Sloan Management Review 30(1):11–17 Schmenner R (2001) Looking ahead by looking back: Swift, even flow in the history of manufacturing. Production and Operations Management 10(1):87–96 Schmenner R, Swink M (1998) On theory in operations management. Journal of Operations Management 17(1):97–113 Schonberger R (1982) Japanese manufacturing techniques: Nine hidden lessons in simplicity. Free Press, New York Stalk G (1988) Time-The next source of competitive advantage. Harvard Business Review 66(4):41–51 Stalk G, Hout T (1990) Competing against time: How time-based competition is reshaping global markets. The Free Press, New York Suri R (1994) Common misconceptions and blunders in implementing quick response manufacturing. In: Proceedings of the SME Autofact 94 Conference, Society of Manufacturing Engineers, p 23 Suri R (1998) Quick Response Manufacturing. Productivity Press: Portland, OR Womack J, Jones D (1996) Lean thinking: Banish waste and create wealth in your corporation. Simon & Schuster, New York Womack J, Jones D, Roos D (1990) The machine that changed the world. Rawson Associates, New York

Chapter 21

Managing Demand Through the Enablers of Flexibility: The Impact of Forecasting and Process Flow Management Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

Abstract In recent years increased attention has been paid to integrated demand and supply chain management. The present research study discusses the main concept in this context, i. e., flexibility. In particular, we have learned from existing research work, that is difficult to link the construct “flexibility” with performance (efficiency as well as effectiveness). Therefore, we analyzed important enablers of flexibility. Based on our conceptual flexibility framework, we discussed the impact of layout, process flow management as well as the forecasting performance (error). Our results provided some interesting insights. In particular, only process flow management is linked with effectiveness as well as efficiency performance. On the other hand, layout as well as forecasting performance, is only linked with efficiency. These results demonstrate that, e. g., forecasting is not directly linked with external results like customer satisfaction. Based on these results it is possible to motivate further research activities that should investigate these complex relationships more in detail.

21.1 Introduction In recent years, literature has devoted more and more attention to the problem of supply and demand management in uncertain contexts. On the one hand, literature Matteo Kalchschmidt Department of Economics and Technology Management, Universit di Bergamo – Viale Marconi 5, 24044 Dalmine e-mail: [email protected] Yvan Nieto Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected] Gerald Reiner Institut de l’entreprise, Universit´e de Neuchˆatel – Rue A.-L. Breguet 1, CH-2000 Neuchˆatel e-mail: [email protected]

265

266

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

in the field of demand management has discussed this topic from several points of view; relevant attention has been paid on building better and more performing forecasting techniques and approaches, in order to reduce the uncertainty companies perceive (Hanssens et al, 2003). Other authors have applied models to learn how to reduce demand uncertainty through the adoption of marketing actions, e. g., every day low price strategies (Lee and Tang, 1997). In this context, improvement of information sharing based on partnerships with customers is also of interest (Cachon and Fisher, 2000). On the other hand, in the supply management literature several contributions have been provided on the adoption of flexibility as a mean to cope with uncertainty (Lee, 2002). Limited contributions, however, can be found regarding the interaction among these leverages (i. e. forecasting, information sharing and process flow management) to manage demand in uncertain contexts as well as regarding their joint effect on company performance. In the end, it should be one of the core objectives in OM research to match capacity and inventory management with customer demand management in order to maximize business results. In other words, in terms of our specific research study, what is flexibility and what is the impact of flexibility with regard to dynamic interactions with forecast accuracy/characteristic, process flow, etc.? Based on these results the next research questions could be to identify what is the “right” flexibility level? The aim of this work is thus to study the relationship between enablers (i. e., practices under consideration of contingency factors) of flexibility and company performance.

21.2 The “Concept” of Flexibility Despite flexibility being commonly recognized as a key answer to environmental uncertainty (Ho et al, 2005), understanding the relations existing between flexibility and performance still presents open challenges. Of course, evidence of the positive impact of flexibility on performance have been provided (e. g. Suarez et al, 1996; Das, 2001; Jack and Raturi, 2002; Hallgren and Olhager, 2009). Nevertheless, important knowledge is still assumed to be missing in order to clearly understand flexibility and performance interrelationship. A first explanation for this gap is of course the complexity of the flexibility concept which, for example, still lacks an unanimous definition (Zhang et al, 2002). Due to the vast and multi-dimensional aspect of the concept, numerous difficulties arise when trying to encapsulate flexibility as a whole in a single measure (see also below). Flexibility has been widely studied at the manufacturing level (see e. g. Kara and Kayis, 2004; Koste and Malhotra, 1999, for good reviews), and knowledge is nowadays also increasing regarding supply chain flexibility (see Stevenson and Spring, 2007, for a review). Considering the latter, Reichhart and Holweg (2007) proposed an interesting conceptual framework of responsiveness and flexibility at the supply chain level. In fact, that framework focuses on responsiveness but the proximity of those concepts makes it easy to transpose it to a flexibility framework. Nevertheless, the proximity of both notions leads to con-

21 Managing Demand Through the Enablers of Flexibility

267

siderable ambiguity in the existing literature and both words have even been used interchangeably, even though they are distinct concepts. Reichhart and Holweg’s framework make the distinction between external and internal flexibility, i. e. external flexibility is linked to achieving a competitive advantage (‘what the customer sees’) by opposition to internal flexibility which is the internal means by which external flexibility can be achieved (‘what can we do’). Following Slack (1987) and Upton (1994) a system’s flexibility is based on internal resources that can be used to achieve different types of internal flexibility, which in turn can support the system’s ability to demonstrate external flexibility to its environment. This distinction separates the capabilities of operations resources from the market requirements, the dual influences that need to be reconciled by operations strategy (Slack, 2002). This discrepancy between internal and external flexibility may explain contradicting results concerning the relationship between uncertainty and flexibility. E. g., Swamidass and Newell (1987) found that flexibility improves performance in uncertain environments, in contrast Pagell and Krause (1999) found no relationship between measures of environmental uncertainty and operational flexibility in a survey among North-American manufacturers.

Capacity Management

Inventory management ………..

Flexibility

External Results

Internal Results

Enablers and their Interrelations

Enhanced forecasting

Direct customer requirements and demand characteristic

Average on hand inventory, inventory obsolescence, bullwhip effect measure, utilization (resources), costs, …

Delivery performance, delivery time, sales, lost sales, customer satisfaction, …

Fig. 21.1 Flexibility: conceptual framework

A second dimension of Reichhart and Holweg’s framework is its distinction between potential and demonstrated responsiveness. Based on Upton (1994), this element also applied for flexibility, differentiating the available flexibility (potential) from the flexibility actually needed to fulfill customer requirements (demonstrated). Based on the above mentioned research, it can be summarized that the approaches applied (e. g., inventory management, forecasting) are the flexibility enablers to fulfill the customer requirements and finally increase customer satisfaction. Further-

268

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

more, these approaches will have also an impact on the efficiency, i. e. costs. The total success (effectiveness as well as efficiency) of flexibility can be only evaluated under consideration of both aspects. The overall framework is presented in Fig. 21.1. Flexibility and performance are known to be influenced by contingency factors which also affect their relation. Numerous examples of influences from contingent factors have been provided, including e. g. perceived environmental uncertainty (Swamidass and Newell, 1987) or company’s business strategy (Gupta and Somers, 1996) and, reviewing literature on manufacturing flexibility, Vokurka and O’LearyKelly (2000) identify four general contingent factors. Specifically, the authors mentioned strategy, environmental factors, organizational attributes and technology as being exogenous variables impacting on flexibility and performance, highlighting once more the complexity of the topic. Finally, we can summarize, when considering flexibility as the result of a set of business practices, that contingency is of main importance as a selected practice may not be possible in all settings (e. g. Ketokivi, 2006). In the context of demand management, forecasting is also known to be a strong lever against uncertainty and thus can contribute in gaining better performances. Literature traditionally considers accuracy as the relevant performance to be evaluated in a forecasting process (Mentzer and Bienstock, 1998; Chase, 1999). When forecast accuracy increases, cost and delivery performances consequently improve, as they are typically correlated with forecast error. Inventory levels, and thus related costs, can be reduced; manufacturing systems are better managed, as equipment utilization improves and companies can effectively plan in advance actions to be undertaken (Vollmann et al, 1992; Ritzman and King, 1993; Fisher and Raman, 1996). In turn manufacturing and product costs decrease. Delivery performances (e. g., order fulfillment and delivery speed/punctuality) also improve as, when forecast accuracy is higher, it is more probable that products are available when the customer orders (Enns, 2002; Kalchschmidt et al, 2003). Forecast inaccuracy causes major rescheduling and cost difficulties for manufacturing (Ebert and Lee, 1995) and it may impact on logistic performances such as delivery timeliness and quality (Kalchschmidt and Zotteri, 2007). For these reasons, it is no surprise that several surveys show accuracy as the most important criterion in selecting a forecasting approach (Dalrymple, 1987; Mahmoud et al, 1988). For this reason some authors have even recommended eliminating forecasts entirely (Goddard, 1989). Other possibilities to hedge uncertainty are inventory management as well as capacity management. Traditionally, inventory management is challenging because uncertain demand and uncertain supply and/or production flow times make it necessary to hold inventory at certain positions to provide adequate service to the customers. As a consequence, increasing process inventories will increase customer service and revenue, but it comes at a higher cost. Therefore, management has to resolve this tradeoff by identifying possibilities to decrease inventories by simultaneously improving customer service. A well known management lever in this respect is risk pooling by different types of centralization or standardization, e. g. central warehouses, product commonalities, postponement strategies (see e. g. Tallon, 1993). In this way, it is usually possible to reduce inventory costs to a large extent. However, this reduction of inventory costs often is related with an increase of other costs such as transporta-

21 Managing Demand Through the Enablers of Flexibility

269

tion costs or production costs. If activities are postponed downstream in the process by shifting the customer order decoupling point upstream in the process, the order flow time is affected. E. g., if no additional resources are allocated to the postponed activities, the order flow time and thus the delivery time for a customer will be increased. Therefore, additional resources (labour and/or equipment) have to be taken into account for the evaluation of such process changes and the additional production costs have to be traded off with the reduction in inventory costs (Jammernegg and Reiner, 2007).

21.3 Objectives and Methodology The aim of this work is to study the relationship between enablers of flexibility and performance. In particular, the paper wants to address the following research question: what is the impact of flexibility enablers (forecasting, layout, process flow management, etc.) on companies’ operational performances? In order to analyze this research question, we considered two different performances: efficiency and effectiveness. Analytical literature suggests that enablers may have some impacts on both performances, but as we mentioned, limited empirical evidence can be found on these relationships. Thus the theoretical model we are considering is represented in Fig. 21.2. Empirical analysis is based on data collected from the 4th edition of the GMRG survey. The Global Manufacturing Research Group (GRMG) collects information regarding manufacturing practices in several countries all over the world. Currently, full data sets have been provided by 598 companies in 13 different countries (Austria, Australia, China, Germany, Hungary, Italy, Korea, Mexico, Poland, Sweden, Switzerland, Taiwan, USA), all belonging to manufacturing and assembly industry. /D\RXW

3URFHVVIORZPDQDJHPHQW

)RUHFDVWLQJ

(IILFLHQF\DQG HIIHFWLYHQHVV

Fig. 21.2 The theoretical model

Table 21.1 synthesizes the distribution of the sample in terms of size and Table 21.2, the distribution among the different countries. The sample size shows several medium and large companies, but also some small companies are present in the dataset.

270

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

Table 21.1 Distribution of the sample in terms of size Company size

Frequency

Small (less than 50 employees) Medium (50 – 250 employees) Large (more than 250 employees)

19.6 % 38.0 % 42.4 %

Table 21.2 Distribution of the sample between countries Country

%

Australia Austria China Germany Hungary Italy Korea

5.0 2.8 9.5 0.7 8.8 9.0 19.1

Country Mexico Polonia Sweden Switzerland Taiwan USA

% 13.1 9.5 2.2 5.6 8.3 6.6

In order to analyze the research questions previously mentioned we proceed as follows. • First we define proper items and constructs in order to measure the relevant variables we have considered. Based on the GMRG database we were able to collect information regarding the different variables. Reliability of the constructs is tested through confirmative factor analysis and reliability analysis. • Then we adopt linear regression to evaluate the relationships between the considered variables.

21.4 Empirical Analysis In order to define the different constructs we applied a confirmative factor analysis based on the items that according to current literature should be influenced by these variables. All items considered are measured on a 7 point Likert scale ranging from 1 (not at all) to 7 (a great extent). In order to evaluate flexibility we defined two separate constructs based on previous literature on the enablers of flexibility: layout and process flow management. Since our goal is to study the impact of flexibility on performances, theoretically we should have considered flexibility performances. However, it is very difficult to identify flexibility performances that are strictly related to specific practices. For this reason we decide to evaluate flexibility by means of practices, thus assuming that a relationship exists between what companies do (i. e. practices) and what they gain (i. e., performances). The use of proper layout solutions can influence flexibility capabilities i. e. through the use of cellular manufacturing systems or by leveraging on automation. Coherently with previ-

21 Managing Demand Through the Enablers of Flexibility

271

ous literature, in order to measure the extent of investment on layout for flexibility, we considered the extent to which companies have invested on: (1) cellular manufacturing and (2) factory automation.The two items are correlated with each other (Pearson Correlation index is 0.44 and significant at 0.01 level). To measure the extent of investment on responsiveness, we considered the extent to which companies have invested on: (1) Just-In-Time, (2) Manufacturing throughput time reduction, (3) Setup time reduction and (4) Total Quality Management. The items are correlated with each other (all Pearson Correlation indexes are above 0.40 and significant at 0.001 level). Thus the constructs layout and process flow management are defined by averaging the specific items. We assessed convergent validity and unidimensionality of the defined constructs with a confirmative factor analysis model. Literature recommends using normed fit index (NFI) and comparative fix index (CFI) together in assessing model fit. NFI is 0.98 and CFI is 0.99 which let us consider the model as acceptable (Hu and Bentler, 1999). In addition root mean square error of approximation (RMSEA) is 0.05 which suggests that the model fit is acceptable. Factors loads are all significant and respect the lower suggested value of 0.40 (Gefen et al, 2000). Cronbach’s Alpha was also measured in order to verify reliability of the constructs; constructs were considered reliable if Alpha’s value is above the minimum requirement of 0.60 (Nunnally and Bernstein, 1994). To evaluate forecasting accuracy, companies have been asked to provide the average error for a single product on a two-month period. Thus we evaluate short term forecast performances. In the end, efficiency and effectiveness performances were considered. As for efficiency, three items were examined, as we asked respondents to provide an evaluation of the following performances compared with their competitors on a 7 point Likert scale (1 is for “far worse than” and 7 for “far better than”): (1) direct manufacturing costs, (2) total product costs, (3) raw material costs. As to the effectiveness performances, a similar question was asked for the following: (1) product quality, (2) delivery speed and (3) delivery as promised. It can be noted that, as it is difficult to compare performances between companies operating within different contexts, this research focuses on perceptual and relative measures of cost and delivery performances. Thus the constructs efficiency and effectiveness are defined by averaging the specific items. NFI is 0.99 and CFI is 1.00 which let us consider the model as acceptable. In addition RMSEA is 0.00 which suggests that the model fits well. Factors loads are all significant and Cronbach’s Alpha value is significantly above the minimum requirement of 0.60. When dealing with survey data, common method bias (CMB) can affect statistical results. As suggested by Podsakoff et al (2003), we checked for this problem by means of confirmatory factor analyses (CFA) on competing models that increase in complexity (Podsakoff et al, 2003). If method variance is a significant problem, a simple model (e. g., single factor model) should fit the data as well as a more complex model (in this case a five factor model). The hypothesized model, containing five factors yielded a better fit of the data than the simple model (one factor model: CFI 0.56 and RMSEA 0.17; five factor model: CFI 0.98 and RMSEA 0.04). Furthermore, the improved fit of the six factor model over the simple model was statistically significant: the change in χ 2 is 1030.40 and the change in df is 9 (p <.001). Thus, CMB did not appear to be of concern in our analy-

272

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

sis. In order to study the research questions we adopted linear regression between the different variables considered. In particular we run two different regression analyses by considering efficiency and effectiveness as dependent variables. We also control the regression results by adding the size of the company as a control variable; to evaluate the contribution of the independent variables we consider the coefficient of determination R2 increment (specifically whether it is significant or not) and check for multicollinearity problems through the analysis of the variance inflation factor (VIF). Table 21.3 summarizes the results of the linear regression between forecasting performances and flexibility practices on efficiency performances. Table 21.3 Results of the regression analysis: dependent variable is efficiency Model Variables

Unstd. Stand. Sig. VIF coef. coef.

11

(Constant) Employees

4.36 0.00

22

(Constant) Employees Layout Process flow mgt Forecast error

3.51 0.00 0.11 0.13 0.19 0.12 0.16 -0.59 -0.14

0.000 0.19 0.000 1.00 0.000 0.015 0.000 0.003 0.001

1.08 1.58 1.61 1.01

As we can see from the first regression analysis, the size of the company is positively related to its capability of being efficient. This is no surprise since size is typically related to the ability of gaining economics of scale. However, in the second model, the variables we are considering are all significantly related to efficiency and the model fit is significantly better. In particular both flexibility-related practices are positively related to efficiency, thus the more companies invest in layout and responsiveness the more they are able to improve their performances. Coherently, forecast error is negatively related to efficiency. Multicollinearity doesn’t seem to be a main concern since VIF is lower than 2 for all variables. Table 21.4 provides the same analysis for effectiveness performances. As we can see, size is no longer significant significant any more. Quite interestingly, only process flow management is related to effectiveness performances. This provides evidence that layout investments for improving flexibility and forecasting accuracy mainly have impact on efficiency.

1 2

Regression is significant at 0.000 level; R2 = 0.037; R2ad j = 0.035.

Regression is significant at 0.000 level; R2 = 0.154; R2ad j = 0.147; R2 change is significant at 0.000 level. 3 Regression is not significant 4 Regression is significant at 0.000 level; R2 = 0.102; R2 = 0.094; R2 change is significant at ad j 0.000 level.

21 Managing Demand Through the Enablers of Flexibility

273

Table 21.4 Results of the regression analysis: dependent variable is effectiveness Model Variables

Unstd. Stand. Sig. VIF coef. coef.

13

(Constant) Employees

5.23 0.00

24

(Constant) Employees Layout Process flow mgt Forecast error

4.39 0.000 0.00 -0.02 0.725 1.07 0.04 0.006 0.261 1.59 0.18 0.27 0.000 1.62 -0.20 -0.05 0.236 1.01

0.000 0.07 0.158 1.00

21.5 Discussion A first interesting result is that a relationship between flexibility enablers and performance exists. This is important empirical evidence because we are able to show that flexibility enablers (layout as well as process flow management) can be very effective in gaining better performances both internally (efficiency) and externally (effectiveness). Quite interestingly, however, this relationship is very strong with efficiency performances, while only process flow management has an impact on effectiveness. This means that companies investing on layout should expect to achieve better internal performances, but at the same time, they may have to pay attention to the impact on the customer. A second interesting result relates to forecasting. In fact we can see that forecasting accuracy impacts on efficiency performances (coherently with what other contributions provide (see previous literature discussion). However, this impact disappears when effectiveness is considered, claiming that forecasting accuracy doesn’t directly replicate in customers’ satisfaction. This result is consistent with previous works (see Danese and Kalchschmidt, 2008) that show that this missing link doesn’t mean that forecasting is useless, but rather that this relationship is far more complex than expected. We argue thus that the relationship between forecasting and process flow management deserves more attention. These results do not provide evidence of a clear preference between the two practices. Therefore, it would interesting to understand whether any synergic relationship exists between the two or if process flow management simply compensates forecasting error, thus limiting the negative impact that forecast error may have on performances. In the end the provided results emphasize that the impact of flexibility on performances is very complex: specifically, different levers can have very different impacts on performances. Thus, attention on what companies are doing to increase flexibility should be seriously considered. To conclude, we would also like to address some limitations of this work. First of all, as we mentioned, we compared flexibility enablers (i. e. practices) with forecasting performances. This approach was due to the inability of defining specific measures for flexibility. The comparison is not completely fair since we are not putting together homogeneous variables. Next, studies should take this issue into consideration, by, for example, looking at forecasting

274

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

practices and not simply at the outcome of this process. A second issue relates to contingencies: we didn’t consider specifically any contingent factor that may influence the different variables and relationships here described. Future works should devote attention to those factors that may change how variables are defined. We argue that general results will be not be drastically affected by these variables, also because several degrees of freedom are left to companies in terms of which practices they can adopt. Acknowledgements Partial funding for this research has been provided by the PRIN 2007 fund “La gestione del rischio operativo nella supply chain dei beni di largo consumo” as well as by the project “Matching supply and demand - an integrated dynamic analysis of supply chain flexibility enablers” supported by the Swiss National Science Foundation.

References Cachon GP, Fisher ML (2000) Supply chain inventory management and the value of shared information. Management Science 46(8):1032–1048 Chase C (1999) Sales forecasting at the dawn of the new millennium? Journal of Business Forecasting Methods and Systems 18:2–2 Dalrymple D (1987) Sales forecasting practices: Results from a United States survey. International Journal of Forecasting 3(3):379–91 Das A (2001) Towards theory building in manufacturing flexibility. International journal of production research 39(18):4153–4177 Enns S (2002) MRP performance effects due to forecast bias and demand uncertainty. European Journal of Operational Research 138(1):87–102 Fisher M, Raman A (1996) Reducing the cost of demand uncertainty through accurate response to early sales. Operations Research 44(1):87–99 Gefen D, Straub D, Boudreau M (2000) Structural equation modeling and regression: Guidelines for research practice. Structural Equation Modeling 4(7) Goddard W (1989) Lets scrap forecasting. Modern Materials Handling 39:39 Gupta Y, Somers T (1996) Business strategy, manufacturing flexibility, and organizational performance relationships: a path analysis approach. Production and Operations Management 5(3):204–233 Hallgren M, Olhager J (2009) Flexibility configurations: Empirical analysis of volume and product mix flexibility. Omega 37(4):746–756 Hanssens D, Parsons L, Schultz R (2003) Market response models: Econometric and time series analysis. Kluwer Academic Publishers Ho C, Tai Y, Tai Y, Chi Y (2005) A structural approach to measuring uncertainty in supply chains. International Journal of Electronic Commerce 9(3):91–114 Hu L, Bentler P (1999) Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural equation modeling 6(1):1–55

21 Managing Demand Through the Enablers of Flexibility

275

Jack E, Raturi A (2002) Sources of volume flexibility and their impact on performance. Journal of Operations Management 20(5):519–548 Jammernegg W, Reiner G (2007) Performance improvement of supply chain processes by coordinated inventory and capacity management. International Journal of Production Economics 108(1-2):183–190 Kalchschmidt M, Zotteri G (2007) Forecasting practices: empirical evidence and a framework for research. International Journal of Production Economics 108:84– 99 Kalchschmidt M, Zotteri G, Verganti R (2003) Inventory management in a multiechelon spare parts supply chain. International Journal of Production Economics 81:397–413 Kara S, Kayis B (2004) Manufacturing flexibility and variability: an overview. Journal of Manufacturing Technology Management 15:466–478 Ketokivi M (2006) Elaborating the contingency theory of organizations: The case of manufacturing flexibility strategies. Production and Operations Management 15(2):215–228 Koste L, Malhotra M (1999) A theoretical framework for analyzing the dimensions of manufacturing flexibility. Journal of Operations Management 18(1):75–93 Lee H (2002) Aligning supply chain strategies with product uncertainties. California Management Review 44(3):105–119 Lee H, Tang C (1997) Modelling the costs and benefits of delayed product differentiation. Management science 43(1):40–53 Mahmoud E, Rice G, Malhotra N (1988) Emerging issues in sales forecasting and decision support systems. Journal of the Academy of Marketing Science 16(3):47–61 Mentzer J, Bienstock C (1998) Sales forecasting management. Sage Beverley Hills, CA Nunnally J, Bernstein I (1994) Psychometric theory. New York, NY Pagell M, Krause D (1999) A multiple-method study of environmental uncertainty and manufacturing flexibility. Journal of Operations Management 17(3):307–325 Podsakoff P, MacKenzie S, Lee J, Podsakoff N (2003) Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology 88(5):879–903 Reichhart A, Holweg M (2007) Creating the customer-responsive supply chain: a reconciliation of concepts. International Journal of Operations & Production Management 27(11):1144–1172 Ritzman L, King B (1993) The relative significance of forecast errors in multistage manufacturing. Journal of Operations Management 11(1):51–65 Slack N (1987) The flexibility of manufacturing systems. International Journal of Operations & Production Management 7(4):35 – 45 Slack N (2002) Operations Strategy. Prentice Hall Stevenson M, Spring M (2007) Flexibility from a supply chain perspective: definition and review. International Journal of Operations & Production Management 27(7):685–713

276

Matteo Kalchschmidt, Yvan Nieto and Gerald Reiner

Suarez F, Cusumano M, Fine C (1996) An empirical study of manufacturing flexibility in printed circuit board assembly. Operations Research 44(1):223–240 Swamidass P, Newell W (1987) Manufacturing strategy, environmental uncertainty and performance: a path analytic model. Management Science 33(4):509–524 Tallon W (1993) The impact of inventory centralization on aggregate safety stock: the variable supply lead time case. Journal of Business Logistics 14:185–185 Upton D (1994) The management of manufacturing flexibility. California Management Review 36(2):72–89 Vokurka R, O’Leary-Kelly S (2000) A review of empirical research on manufacturing flexibility. Journal of Operations Management 18(4):485–501 Vollmann T, Berry W, Whybark D (1992) Manufacturing planning and control systems Zhang Q, Vonderembse M, Lim J (2002) Value chain flexibility: a dichotomy of competence and capability. International Journal of Production Research 40(3):561–583

Chapter 22

Threats of Sourcing Locally Without a Strategic Approach: Impacts on Lead Time Performances Ruggero Golini and Matteo Kalchschmidt

Abstract This paper analyses the impact of local sourcing on lead time performances. In particular attention is devoted to the effect of choosing to source locally without having made a proper strategic analysis of the purchasing process. Analyses are based on data provided by the IMSS research project regarding more than 500 companies in different countries around the world. Results show that local sourcing can lead to bad performances if it is adopted without a clear sourcing strategy.

22.1 Introduction During the last twenty years companies have witnessed a considerable expansion of supply chains into international locations (Taylor, 1997; Dornier et al, 1998). This growth in globalization has motivated both practitioner and academic interest in global supply chain management (Prasad and Babbar, 2000). Looking only at the upstream part of the supply chain, global sourcing (i.e. the management of supplier relationships on a global perspective) has been considered and analyzed (e.g., Kotabe and Omura, 1989; Murray et al, 1995). One major issue regarding global sourcing is why companies extend their relationships internationally and to what extent this practice contributes to increase their competitive advantage (e.g., Alguire et al, 1994; Womack and Jones, 1996; Trent and Monczka, 2003). Bozarth et al (1998) identify different motivators for global sourcing: offset requireRuggero Golini, corresponding author Department of Economics and Technology Management, Universit`a degli Studi di Bergamo, Viale Marconi 5, 24044 Dalmine (BG), Italy, Tel. +39 035 205 2360, Fax. +39 035 205 2077, e-mail: [email protected] Matteo Kalchschmidt Department of Economics and Technology Management, Universit`a degli Studi di Bergamo, Viale Marconi 5, 24044 Dalmine (BG), Italy, Tel. +39 035 205 2360 Fax. +39 035 205 2077, e-mail: [email protected]

277

278

Ruggero Golini and Matteo Kalchschmidt

ments, currency restrictions, local content and countertrade, lower prices, quality, technology access, access to new markets, shorter product development and life cycles, competitive advantage. In some cases, internal factors (e.g., company image) can be the principal motivators (Alguire et al, 1994). However, sourcing globalization is still little diffused (Trent and Monczka, 2003; Cagliano et al, 2008). This is because global sourcing can imply longer supply lead times that can bring to higher inventory levels and other hidden costs (Handfield, 1994). Moreover, in a global sourcing context it becomes more difficult to have an integrated and efficient supply chain (Das and Handfield 1997). Nevertheless, sourcing locally can lead to a competitiveness loss, if other companies are able to efficiently exploit globalization opportunities. In fact, thanks to experience and investment in supply chain, companies can achieve better performances, also on lead times and inventories, even with a globalized supply base (Bozarth et al, 1998; Golini and Kalchschmidt, 2008). On the other side, local sourcing allows companies to invest in JIT with suppliers thus allowing improving procurement performances. In fact many companies prefer to source locally, and in some cases they invest in Just-in-Time practices - that require physical proximity - to keep inventories under control (Das and Handfield, 1997; Prasad and Babbar, 2000). Actually, several studies have failed to detect any significant impact on general business success of global supply chains (Kotabe and Omura, 1989; Steinle and Schiele, 2008) and specifically of global sourcing. Only some weak evidences have been found: it seems that global sourcing can improve product and process innovation, but it seems to have no impact on strategic performances (Kotabe, 1990; Murray et al, 1995). Companies, however, have to carefully select their globalization strategy, using, for example, hybrid global/local approaches according to the type of goods purchased (Steinle and Schiele, 2008). In the rather developed literature on global or local sourcing, however, there are limited empirical researches regarding the impact of this practice on lead times. This paper aims at contributing to this issue, by providing evidence on the relationship between global and local purchasing, supply chain management and lead time performances. The remainder of the paper is structured as follows. In the next section literature regarding the relationship between global sourcing, supply chain management and lead time performances is taken into account. Following, research objectives and methodology are detailed and empirical analyses are described. Then discussion of empirical results is provided and, in the end, we draw some conclusions and suggest potential future developments.

22.2 Literature Review Lead times are a major concern for suppliers as these impact directly on cus-tomers’ performances: lower lead times induce lower inventories and allow to be more flexible. Also from a supply chain perspective, lead time reduction positively contributes in reducing the bullwhip effect (Chen et al, 2000) making the entire supply chain

22 Threats of Sourcing Locally Without a Strategic Approach

279

more efficient. On the other side competitive pressures (lower costs, higher quality, innovativeness) drive many companies in scouting suppliers abroad (Alguire et al, 1994; Ettlie and Sethuraman, 2002; Frear et al, 1992; Smith and Reece, 1999; Trent and Monczka, 2003; Birou and Fawcett, 1993; Womack and Jones, 1996). This practice may intuitively cause higher procurement lead times mainly because of geographical distances. This is confirmed by Handfield (1994): among the top five costs problems experienced in using international sources there are long lead time and inventory costs. In the same study it is also shown that an international sourcing systematically causes less on-time deliveries, longer lead times and higher lead times. However the problem is more complex, as at least three aspects of the system dynamic have to be considered. First of all, companies may compensate higher procurement lead time with higher inventories thus not affecting the lead time for the customer. The second aspect is the position of the decoupling point: companies can have higher material inventories if they produce in make-to-stock or make-to-order. Engineer-to-order companies instead have a direct impact of the procurement lead time on the total lead time. Moreover the quality level of the supply may affect the manufacturing lead time, as scrap and reworks can make it longer. Again companies can react to this with higher work-in-progress and finished goods inventories if their production model allows it (i.e. make-to-stock). Finally supply chain practices have to be considered. In fact literature suggests reducing procurement lead time by means of investments in supply chain integration with suppliers (e.g., Droge et al, 2004; Fliedner, 2003). Many contributions (e.g., Frohlich and Westbrook, 2001) show how a higher level of integration provides better operational performances (also in terms of lead time), thus suggesting that firms should invest in this direction. In this area, researchers have identified two complementary ways in which supply chain integration can be applied (Cagliano et al, 2005; Vereecke and Muylle, 2006): information sharing and system coupling. The first regards exchanging information on market demand, inventories, production plans, delivery dates, etc. Lee et al (1997) provide several examples of information sharing as an effective instrument to face the bullwhip effect. Regarding information sharing domain, significant importance has been given to the use of electronic tools for information exchange and integration. The adoption of electronic communication channels between firms has been a relevant issue for several years (e.g., Malone et al, 1987), and the literature contains several contributions on the role of ICT in SCM from different perspectives (for a review, see Gunasekaran and Ngai, 2005; Power and Singh, 2007). More recently, literature has focused on Internet-based electronic communication (i.e. eBusiness). According to this literature, purchasing performances can be improved through the adoption of internet tools (McIvor et al, 2000), and the flow of information along the supply chain can be easily transferred, which helps companies to be more responsive (e.g., Naylor et al, 1999; Aviv, 2001; Devaraj et al, 2007). The second area of integration, system coupling, is represented by coordinating physical activities, through mechanisms as VMI, CPFR or JIT-Kanban to obtain a smooth material flow and a seamless supply chain (see for example Childerhouse

280

Ruggero Golini and Matteo Kalchschmidt

et al, 2002; Disney and Towill, 2003). From this point of view, an integrated supply chain offers the opportunity for firms to compete on the basis of speed and flexibility, while at the same time holding minimum levels of inventory in the chain (Power, 2005). In particular some studies have highlighted that JIT sourcing requires specific conditions (e.g., frequent and fast deliveries, small lots, etc.) that can be difficult to be performed in an international environment. So, even if it is possible to achieve efficiency in a global sourcing context through JIT, they are not yet comparable to what can be gained at domestic level (Das and Handfield, 1997). Finally, some authors have highlighted the importance of designing a proper sourcing process in all its subphases and applying proper tools (Zeng, 2003; Petersen et al, 2005; Quintens et al, 2005; Gelderman and Semeijn, 2006). All this evidence suggests to companies with global sourcing processes to invest in supply chain management and integration, e.g. to keep lead times under control (Bozarth et al, 1998). In conclusion, literature highlights how companies can improve operational performances (and lead time ones specifically) through supply chain integration, i.e. information sharing and system coupling. Nevertheless, integration - and especially system coupling - can be difficult to be performed in a global sourcing context because of suppliers’ distance. This can make more difficult for companies to control sourcing globalization counter effects - mainly longer lead times - and neutralize lower cost seeking strategies.

22.3 Research Objectives and Methodology The objective of this paper is to explore the relationship between global sourcing, supply chain investments and lead time performances. Literature suggests that global sourcing can have an impact on lead time performances, however this relationship is not completely straightforward. For example, from one side global sourcing means that companies purchase from suppliers that are far from the plant compared to what companies can do in a local sourcing situation. This should increase the procurement lead time due mainly to transportation over longer distances. However companies sourcing locally will have to choose among a limited set of potential suppliers while when global sourcing is adopted companies are able, at least potentially, to choose the best suppliers, thus being able to gain better performances. For this reason, the first research question this work wants to address is: RQ1 What is the impact of local sourcing on lead time performances? Some companies often source locally simply because they don’t consider other alternatives, either because they are too small or because they rely on consolidated relationships with local suppliers. The impact of local sourcing on performances may not be completely straightforward also because some companies choose to

22 Threats of Sourcing Locally Without a Strategic Approach

281

source locally through a clear and rational analysis. Other companies, on the contrary, choose to purchase locally because that’s enough to achieve their business goals. For this reason we aim at considering also the impact of how the choice of local sourcing is made, whether it is likely the result of a structured analysis or not. We argue that companies that source locally based on a strategic analysis of their context have better performances compared to those that simply choose to purchase locally without taking into account strategic issues. Thus we formulate the following research question: RQ2 What is the impact of local sourcing on lead time performances? Literature suggests that improvements on procurement lead time can be achieved by leveraging on supply chain investments such as JIT, information sharing with suppliers, coupling between customer and supplier production systems, etc. Limited evidence however can be found regarding the extent to which these investments are related to local sourcing approaches. In particular some investments (e.g., JIT) are positively influenced by local sourcing, but previous works (e.g., Golini and Kalchschmidt, 2008) have shown that other investments are negatively related to local sourcing (e.g., purchasing process improvement programs). Thus our third research question is: RQ3 What is the impact of local sourcing on supply chain investments that influence lead time performances? In order to investigate the above research questions, data have been collected within the fourth edition of the International Manufacturing Strategy Survey (IMSS), a research project carried out in 2005 by a global network. This project, originally launched by London Business School and Chalmers University of Technology, studies manufacturing and supply chain strategies within the assembly industry (ISIC 2835 classification), through a detailed questionnaire that is administered simultaneously in many countries by local research groups; responses are gathered in a unique global database (Lindberg and Trygg, 1991). The sample consisted of 660 companies from 21 countries, with an average response rate of 34 %. The usable sample included 620 companies, which provided enough information for the purpose of this study. Among these companies we limited our analyses only on those companies that don’t rely on Engineering to order manufacturing systems. This is due to the fact that this kind of companies face different challenges compared to assemble/make-to-order or make-to-stock ones (e.g. inventories have a different role). The distribution of the sample in terms of country, size and ISIC code is shown in Table 22.1 and Table 22.2 In order to measure the extent of localization of sourcing activities, we collected information regarding the percentage of purchases inside the country where the plant is based. On the other side, to evaluate the extent to which companies decide to purchase locally for strategic reasons we collected the information regarding the extent to which companies consider the proximity of suppliers a key element in selecting a supplier. This item is measured on a 1-5 Likert scale where 1 equals to “not at all” and “to a great extent”.

282

Ruggero Golini and Matteo Kalchschmidt

Table 22.1 Sampe distribution in terms of country (a) and size (b) - Small: less than 250 employees, Medium: 251-500 employees, Large: over 501 employees

Country

N % Country

N

%

Size

Argentina Australia Belgium Brazil Canada Denmark Estonia Germany Hungary Ireland Israel

36 8 27 11 12 26 19 17 48 9 15

Italy Netherlands New Zealand Norway Portugal Sweden Turkey UK United States Venezuela

39 49 21 15 8 74 28 13 30 22

7.4 9.3 4 2.8 1.5 14 5.3 2.5 5.7 4.2

Total

527 100

6.8 1.5 5.1 2.1 2.3 4.9 3.6 3.2 9.1 1.7 2.8

N

%

Small Medium Large NA

300 100 124 3

56.9 19 23.5 0.6

Total (b)

527 100

(a)

Table 22.2 Distribution of the sample in terms of ISIC code ISIC Code 28 29 30 31 32 33 34 35 NA

Industry description

N

%

Manufacture of fabricated metal products, except machinery and equipment Manufacture of machinery and equipment not elsewhere classified Manufacture of office, accounting and computing machinery Manufacture of electrical machinery and apparatus not elsewhere classified Manufacture of radio, television and communication equipment and apparatus Manufacture of medical, precision and optical instruments, watches and clocks Manufacture of motor vehicles, trailers and semi-trailers Manufacture of other transport equipment

213 95 11 63 31 26 51 32 5

40.4 18 2.1 12 5.9 4.9 9.7 6.1 0.9

Total

527 100

We created three clusters based on these two variables (a detailed methodological description is reported in the following) and next different lead time performances have been compared among clusters. In particular we considered: i) Procurement lead time compared to competitors; ii) Manufacturing lead time compared to competitors; iii) Delivery speed compared to competitors iv) Throughput time efficiency compared to competitors. In order to control our results we also considered inventories since they are typically influenced by lead time performances. Specifically we measured: i) Number of days of production in raw materials inventory; ii) Number of days of production in work in progress inventory; iii) Number of days of production in finished products inventory.

22 Threats of Sourcing Locally Without a Strategic Approach

283

Then, we considered manufacturing conformance performance (relative to competitors and in terms of scrap and rework costs), since it could depend from the quality of the purchased products and it can negatively affect the manufacturing lead time. In the end to evaluate investments in supply chain management, we considered the following variables1 : • Adoption of Just In Time for managing procurement (measured by the percentage of deliveries from suppliers managed through JIT). • Extent of information sharing with suppliers. We defined a latent variable based on the degree of adoption with suppliers of: Share inventory level knowledge, Share production planning decisions and demand forecast knowledge, Order tracking/tracing, Agreements on delivery frequency; • Extent of system coupling with suppliers. We defined a latent variable based on the degree of adoption with suppliers of: Vendor Management Inventory systems, Collaborative Planning, Forecasting and Replenishment, Physical integration; • Extent of use of Internet for managing suppliers. We define a latent variable based on the degree of use with suppliers of: scouting/pre-qualify, Auctions, Rfx, Data analysis, Access to catalogues, Order management and tracking, Content and knowledge management, Collaboration support services. • Purchasing process improvements. We defined a latent variable based on the company’s investments on: Rethinking and restructuring supply strategy and the organization and management of suppliers portfolio through (e.g. tiered networks, bundled outsourcing, and supply base reduction), Implementing supplier development and vendor rating programs, Increasing the level of coordination of planning decisions and flow of goods with suppliers including dedicated investments (in e.g. Extranet/ EDI systems, dedicated capacity/tools/equipment, dedicated workforce, etc.)

22.4 Results In order to analyze our research questions we proceeded as follows. First we applied cluster analysis in order to identify different groups based on the use of local sourcing and whether strategic decision making is applied. Then we analyzed the differences among clusters on some contingent variables in order to check for clusters’ reliability. In the end, we compared the different clusters on performances and supply chain practices.

1 Reliability of these latent variables were checked by controlling Cronbach’s Alpha values (for all variables over 0.6) and items’ factor loads (for all variables over 0.5)

284

Ruggero Golini and Matteo Kalchschmidt

22.4.1 Cluster Analysis The cluster analysis has been performed on two variables: degree of local sourcing (measured as percentage of purchases within the same country where the plant is located - scale from 0% to 100%) and the extent to which physical proximity is important in the suppliers’ selection. Through the analysis of the dendrogram, it is possible to identify that the best solution is three or four clusters. For interpretability sake we selected three clusters because the two variables are positively correlated (sig. 0,000); by looking at how the points are jointly distributed in a scatter diagram it is possible to identify a priori these three high density areas: • Low local sourcing (i.e. global sourcing) and relatively low importance to physical proximity (named in the following Globals) • High local sourcing and high importance to physical proximity (named Patriots). • High local sourcing and low importance to physical proximity (named Idlers). Companies with global sourcing and high importance to physical proximity are, as one could expect, very few and we put them together with the Globals. Because of this data structure, when we perform a Two Step cluster analysis (loglikelihood distance on standardized variables) specifying three as the cluster number we obtain results reported in Table 22.3. Table 22.3 Cluster analysis results. For each cluster the average values and the significant dissimilarities respect to other clusters are represented (** Mann-Whitney U sig.< 0.000) N. Local Sourcing Physical proximity 1 Patriots 233 81.725 (3)** 2 Idlers 105 78.933 (3)** 3 Globals 189 26.060 (1,2)** Average 56.558

3.624 (2,3)** 1.762 (1,3)** 2.476 (1,2)** 2.746

In order to check for the reliability of the defined clusters we tested differences among clusters on different contingency variables related to company size, business objectives, manufacturing globalization, industrial sector and position of decoupling point (see Table 22.4 for details). As we can see there are no differences related to the business objectives and industrial sector. Global sourcers are larger companies and have a more globalized manufacturing network than local ones. This is quite intuitive, even if it is not always confirmed by literature (e.g., Cavusgil et al, 1993; Quintens et al, 2005). Finally it is interesting to notice that companies sourcing locally have higher direct salaries/wages costs and lower direct materials/parts/components. This suggests that Globals, since they have a higher incidence of direct materials and components on the total costs, are more driven to scout best and most convenient suppliers around the world. Given the relevance of these purchases, a longer lead times or higher supply chain investments can be acceptable. We can summarize results by stating that no relevant differences arise from this analysis

22 Threats of Sourcing Locally Without a Strategic Approach

285

and thus we can argue that the clusters are not biased. After the contingency analysis, we looked for differences among clusters on lead time and other related performances and, separately, on SC practices. Since analyzed variables are not normally distributed (based on Kolmogorov-Smirnov test) we adopted non-parametric tests. Table 22.4 Contingent factors analysis results Contingency

Results

Business Objectives Company size Manufacturing globalization Industry (ISIC code) Cost structure

No differences Globals are bigger than Patriots and Idlers Globals have a more globalized production network No significant patterns. Globals have a lower incidence of direct salaries/wages and higher incidence of direct materials/parts/components

22.4.2 Performances For what concerns performances we took into account lead time performances (procurement, manufacturing and delivery lead times), manufacturing conformance, throughput time efficiency, scrap and rework costs, raw materials/components inventory, WIP and final products inventory. In particular we made a two-by-two comparisons among clusters using Mann-Whitney’s U test. Here results are reported for each comparison. Globals vs Idlers. Looking at Table22.5, Globals have superior performances compared to Idlers on Procurement lead time, Manufacturing lead time, Delivery speed and Manufacturing conformance. This claims that even if Globals have more distant suppliers they are still able to have competitive lead times without carrying higher inventories. The shorter manufacturing lead time can be in part explained by the higher manufacturing conformance even if there is no difference on scrap and rework costs. Quite interestingly there is no difference on the throughput time efficiency. Globals vs Patriots. Patriots don’t show particular differences with Globals apart from superior throughput time efficiency (Table22.6). This shows that a strategic approach to local sourcing can give the opportunity to remain competitive on lead time performances (not only procurement, but also manufacturing and delivery lead times). On the other side, global sourcing doesn’t always imply worse lead times. Patriots vs. Idlers. Finally we compared Patriots with Idlers (Table22.7). Results show almost no differences between the clusters a part from a superior throughput time efficiency of the Patriots. Because of that Patriots have probably other reasons rather than lead times to select suppliers locally.

286

Ruggero Golini and Matteo Kalchschmidt

Table 22.5 Average rank of the analyzed clusters on the different performances Globals Idlers Sig. Procurement lead time Manufacturing lead time Delivery speed Manufacturing conformance Throughput Time Efficiency Scrap and rework costs Material/ components inventory WIP inventory Finished products inventory

130 132 135 136 134 149 160 153 159

113 107 115 118 138 163 158 172 158

0.043 0.004 0.027 0.045 0.720 0.217 0.814 0.077 0.940

Table 22.6 Average rank of the analyzed clusters on the different performances Globals Patriots Sig. Procurement lead time Manufacturing lead time Delivery speed Manufacturing conformance Throughput Time Efficiency Scrap and rework costs Material/ components inventory WIP inventory Finished products inventory

163 165 167 170 151 183 204 192 203

153 148 159 161 194 203 186 203 187

0.252 0.061 0.394 0.311 0.000 0.077 0.121 0.350 0.146

Table 22.7 Average rank of the analyzed clusters on the different performances Idlers Patriots Sig. Procurement lead time Manufacturing lead time Delivery speed Manufacturing conformance Throughput Time Efficiency Scrap and rework costs Material/ components inventory WIP inventory Finished products inventory

104 103 106 110 99 131 143 139 138

112 113 118 119 123 133 131 131 129

0.290 0.194 0.142 0.251 0.008 0.793 0.262 0.411 0.344

22.4.3 Supply Chain Management Practices We also took into consideration the supply chain management practices put in place by companies. Namely we considered: Just in time adoption, Information sharing, System coupling, eBusiness, Purchasing process improvements, as defined in the Methodology chapter. Following the same approach of the previous chapter, we compared the clusters two-by-two on the different practices.

22 Threats of Sourcing Locally Without a Strategic Approach

287

Globals vs Idlers. Globals use information sharing, system coupling and eBusiness more than Idlers while there are no significant differences on just in time and supply chain investments (Table22.8). Globals vs Patriots. Globals tend to adopt less JIT compared to Patriots, while there are no significant differences on the other dimensions (Table22.9). Patriots vs Idlers. Patriots use all supply chain practices to a more extent compared to Idlers. The only exception is supply chain investments where no difference arises (Table22.10). Table 22.8 Average rank of the analyzed clusters on the different supply chain practices Globals Idlers Sig. Just in time Information Sharing System Coupling eBusiness Supply chain investments

156 174 178 170 164

155 144 139 134 144

0.980 0.007 0.001 0.001 0.069

Table 22.9 Average rank of the analyzed clusters on the different supply chain practices Globals Patriots Sig. Just in time Information Sharing System Coupling eBusiness Supply chain investments

180 207 211 204 206

208 203 200 201 187

0.013 0.739 0.343 0.816 0.084

Table 22.10 Average rank of the analyzed clusters on the different supply chain practices Idlers Patriots Sig. Just in time Information Sharing System Coupling eBusiness Supply chain investments

124 130 127 121 135

144 153 153 152 139

0.043 0.028 0.010 0.003 0.648

22.5 Discussion The previous results highlight several issues we discuss here.

288

Ruggero Golini and Matteo Kalchschmidt

A first interesting result comes from the clustering procedure itself. Several companies adopt local sourcing without any strategic reason for that, but probably due to common practices. Even if a large extent of our sample considers strategy a key element to decide how to manage sourcing, some companies still don’t pay relevant attention to this topic (20% of our sample). Moreover, even if global sourcing is an ever growing practice (see Cagliano et al, 2008), several companies decide to keep sourcing local and prefer not to extend their supply chains abroad (Patriots constitute 36% of our sample). We argue that this result contributes to the debate on the impact of globalization on managerial practices and the extent of the phenomenon. A second important result relates to the impact on performances. Analyses show clearly that no significant differences emerge between those companies that source global and those that clearly decide to stay local. This lack of difference may be for sure due to several contingency factors (e.g., product characteristics, production organization, strategic goals, etc.), even if contingency analysis doesn’t provide any relevant difference in some of these items. We argue that this result highlights that there is no “one best way” in how to manage sourcing activities (i.e., local vs. global), but that different options are available to companies. Quite interestingly, however, the only difference between those companies that choose to stay local (Patriots) and all others (Idlers and Globals) is that the former ones are capable of having a better throughput time efficiency compared to the latter. This result provides evidence that local sourcing, if it is properly managed, can have significant impacts in terms of internal efficiency. Companies should then take in proper consideration the impact of going global not only for their sourcing activities, but also for their manufacturing processes. Quite interestingly no significant differences (besides that on through put time efficiency previously mentioned) arise between Patriots and Idlers. This may be due to the fact that companies may decide to stay local for different reasons either related to performances that they want to improve directly (e.g., transportation cost, delivery lead time, flexibility etc.) or to access specific local competences. This means that each Patriots may focus its attention only on some specific performances thus leading variance in the sample to increase and limit the difference between the two clusters. A third important result arises again from the comparison of Idlers and the other groups. Quite interestingly, besides the differences on performances, Idlers are also characterized by lower investments on supply chain practices. In fact, compared to both Globals and Patriots, these companies tend not to invest on information sharing with suppliers, system coupling and eBusiness tools. This result explains, at least partially, why these companies have also lower performances compared to the others. It rather seems that sometimes it is not so important from where companies source but how this is done. The only exception to this observation is the adoption of JIT deliveries with suppliers that, due to its characteristics, is less adopted by Globals compared to those that source locally. In the end the paper provides evidence that sourcing locally is not going to provide companies with better performances if this choice isn’t part of the purchasing

22 Threats of Sourcing Locally Without a Strategic Approach

289

strategy of a company. Company should thus spend proper attention in defining a structure approach to sourcing by means of analysis tools and proper investments.

22.6 Conclusion This paper contributes in the understanding of the impact of global and local sourcing on both companies’ performances and behavior. The paper provides evidence of the importance of choosing carefully a sourcing strategy and to consider how it can be supported by leveraging on supply chain management. In the end, we would like to highlight the limitations of this work. First of all, statistical analyses were based on a specific sample, thus in the future replication of this study should be considered in order to verify the reliability of the results provided. Second, due to limitations in terms of space, we didn’t consider at any rate why companies choose to adopt different sourcing approaches. We argue that this element should help in understanding better differences and similarities among the considered clusters. In the, we clustered companies based on the extent of local or global sourcing. However some companies adopt hybrid approaches between these two alternatives according to the type of good purchased. This aspect has not been considered here and future studies should consider this situation and how companies manage hybrid configurations. Acknowledgements Financial support for this research was provided by the PRIN 2007 fund “La gestione del rischio operativo nella supply chain dei beni di largo consumo”.

References Alguire M, Frear C, Metcalf L (1994) An examination of the determinants of global sourcing strategy. Journal of Business and Industrial Marketing 9(2):62–74 Aviv Y (2001) The effect of collaborative forecasting on supply chain performance. Management Science 47(10):1326–1343 Birou LM, Fawcett SE (1993) International Purchasing: Benefits, Requirements, and Challenges. International Journal of Purchasing & Materials Management 29(2):28 – 37 Bozarth C, Handfield R, Das A (1998) Stages of global sourcing strategy evolution: An exploratory study. Journal of operations management 16(2-3):241–255 Cagliano R, Caniato F, Spina G (2005) E -business strategy: How companies are shaping their supply chain through the internet. International Journal of Operations & Production Management 25(12):1309–1327

290

Ruggero Golini and Matteo Kalchschmidt

Cagliano R, Golini R, Caniato F, Kalchschmidt M, Spina G (2008) Supply chain configurations in a global environment: A longitudinal perspective. Operations Management Research DOI 10.1007/s12063-008-0012-0 Cavusgil S, Yaprak A, Yeoh P (1993) A decision-making framework for global sourcing. International Business Review 2(2):143–156 Chen F, Drezner Z, Ryan J, Simchi-Levi D (2000) Quantifying the bullwhip effect in a simple supply chain: The impact of forecasting, lead times, and information. Management Science 46(3):436–443 Childerhouse P, Aitken J, Towill D (2002) Analysis and design of focused demand chains. Journal of Operations Management 20(6):675–689 Das A, Handfield R (1997) Just-in-time and logistics in global sourcing: An empirical study. International Journal of Physical Distribution and Logistics Management 27(3/4):244–259 Devaraj S, Krajewski L, Wei J (2007) Impact of eBusiness technologies on operational performance: The role of production information integration in the supply chain. Journal of Operations Management 25(6):1199–1216 Disney S, Towill D (2003) The effect of vendor managed inventory (VMI) dynamics on the Bullwhip Effect in supply chains. International Journal of Production Economics 85(2):199–215 Dornier P, Ernst R, Fender M, Kouvelis P (1998) Global operations and logistics: Text and cases. John Wiley & Sons, Inc., New York, NY Droge C, Jayaram J, Vickery S (2004) The effects of internal versus external integration practices on time-based performance and overall firm performance. Journal of Operations Management 22(6):557–573 Ettlie J, Sethuraman K (2002) Locus of supply and global manufacturing. International Journal of Operations and Production Management 22(3):349–370 Fliedner G (2003) CPFR: an emerging supply chain tool. Industrial Management and Data Systems 103(1-2):14–21 Frear C, Metcalf L, Alguire M (1992) Offshore sourcing: Its nature and scope. International Journal of Purchasing and Materials Management 28(3):2–11 Frohlich M, Westbrook R (2001) Arcs of integration: An international study of supply chain strategies. Journal of Operations Management 19(2):185–200 Gelderman C, Semeijn J (2006) Managing the global supply base through purchasing portfolio management. Journal of Purchasing and Supply Management 12(4):209–217 Golini R, Kalchschmidt M (2008) Moderating the impact of global sourcing on inventories through supply chain management. ISIR 2008 Conference Proceedings, Budapest Gunasekaran A, Ngai E (2005) Build-to-order supply chain management: A literature review and framework for development. Journal of Operations Management 23(5):423–451 Handfield R (1994) US global sourcing: Patterns of development. International Journal of Operations and Production Management 14(6):40–51

22 Threats of Sourcing Locally Without a Strategic Approach

291

Kotabe M (1990) The relationship between offshore sourcing and innovativeness of US multinational firms: An empirical investigation. Journal of International Business Studies 21(4):623–638 Kotabe M, Omura G (1989) Sourcing Strategies of European and Japanese Multinationals: A Comparison. Journal of International Business Studies 20(1):113–130 Lee H, Padmanabhan V, Whang S (1997) Information distortion in a supply chain: The Bullwhip Effect. Management Science 43(5):546–558 Lindberg P, Trygg L (1991) Manufacturing strategy in the value system. International Journal of Operations & Production Management 11(3):52–62 Malone T, Yates J, Benjamin R (1987) Electronic markets and electronic hierarchies. Communications of the ACM 30(6):484–497 McIvor R, Humphreys P, Huang G (2000) Electronic commerce: Re-engineering the buyer-supplier interface. Business Process Management Journal 6(2):122–138 Murray J, Kotabe M, Wildt A (1995) Strategic and financial performance implications of global sourcing strategy: A contingency analysis. Journal of International Business Studies 26(1):181–202 Naylor B, Naim M, Berry D (1999) Leagility: Integrating the lean and agile manufacturing paradigms in the total supply chain. International Journal of Production Economics 62(1-2):107–118 Petersen K, Handfield R, Ragatz G (2005) Supplier integration into new product development: Coordinating product, process and supply chain design. Journal of Operations Management 23(3-4):371–388 Power D (2005) Supply chain management integration and implementation: A literature review. Supply chain management: An international journal 10(4):252–263 Power D, Singh P (2007) The e-integration dilemma: The linkages between Internet technology application, trading partner relationships and structural change. Journal of Operations Management 25(6):1292–1310 Prasad S, Babbar S (2000) International operations management research. Journal of Operations Management 18(2):209–247 Quintens L, Matthyssens P, Faes W (2005) Purchasing internationalisation on both sides of the Atlantic. Journal of Purchasing and Supply Management 11(2-3):57– 71 Smith T, Reece J (1999) The relationship of strategy, fit, productivity, and business performance in a services setting. Journal of Operations Management 17(2):145– 161 Steinle C, Schiele H (2008) Limits to global sourcing? Strategic consequences of dependency on international suppliers: cluster theory, resource-based view and case studies. Journal of Purchasing and Supply Management 14(1):3–14 Taylor D (1997) Global Cases in Logistics and Supply Chain Management. Information distortion in a supply chain: The bullwhip effect, Boston, MA Trent R, Monczka R (2003) Understanding integrated global sourcing. International Journal of Physical Distribution and Logistics Management 33(7):607–629 Vereecke A, Muylle S (2006) Performance improvement through supply chain collaboration in Europe. International Journal of Operations & Production Management 26(11):1176–1198

292

Ruggero Golini and Matteo Kalchschmidt

Womack J, Jones D (1996) Lean Thinking: Banish Waste and Create Wealth in Your Corporation. Simon & Schuster New York, NY Zeng A (2003) Global sourcing: Process and design for efficient management. Supply Chain Management: An International Journal 8(4):367–379

Chapter 23

Improving Lead Times Through Collaboration With Supply Chain Partners: Evidence From Australian Manufacturing Firms Prakash J. Singh

Abstract Whilst it is well recognized that lead time reductions can be of strategic benefit to firms, most existing methods for generating these outcomes are seen as being too complex and difficult to implement. In this paper, the possibility of using supply chain collaboration for the purpose of reducing lead times was examined. Data from a study involving 416 Australian manufacturing plants showed that there were strong albeit indirect links between collaborative practices that firms develop with key customers and suppliers, and lead time performance. From this, it is suggested that firms consider, amongst other strategies, developing strong collaborative relationships with their trading partners if they wish to reduce lead times.

23.1 Introduction Benefits of reducing lead times associated with new product development and manufacturing are well documented with publications such as Quick Response Manufacturing (Suri, 1998), Competing Against Time (Stalk and Hout, 1990) and Clockspeed (Fine, 1998) bringing sharp attention to this issue. However, a significant challenge continues to persist. This challenge is in how firms can achieve significant reductions in lead time. Although the literature provides some prescriptions on how lead times can be reduced, many firms continue to struggle to reduce lead times. The existing ideas appear to be too steeped in mathematical modeling, leading managers to believe that lead time reduction methods are too difficult and costly to implement (Suri, 1998; De Treville et al, 2004; Tersine and Hummingbird, 1995; Suri, 1999). There is a need to develop methods for lead time reduction that are simpler and practically achievable. One such idea emanates from the supply chain management (SCM) body of knowledge. More specifically, the strong emphasis that SCM thePrakash J. Singh Department of Management & Marketing, University of Melbourne-Parkville, 3010, Australia e-mail: [email protected]

293

294

Prakash J. Singh

ory places on close inter-organizational relationships between key trading partners would appear, prima facie, to provide the opportunity for firms to achieve significant reductions in lead times, along with many other benefits. A number of terms are used to describe the close relationships between supply chain partners. These include cooperation, coordination and collaboration. Although these terms are sometimes used interchangeably, there are subtle but important differences between them. In this paper, the focus is on collaboration. An attempt is made to shed light on what characterizes successful collaboration in order enable firms to reduce lead times associated with new product development and manufacturing. Since much of the SCM literature is replete with the idea that firms should work closely with their key customers and suppliers, this collaboration model is essentially based on this principle. To inform the model, both the demandside and supply-side literature were examined for practices on how firms should collaborate with customers and suppliers respectively. This led to the development of three key collaboration constructs: internal organizational processes, relationships with customers and relationships with suppliers. It was predicted that relationships with customers and relationships with suppliers affected performance (measured in terms of reductions in new product development and manufacturing lead times), both directly and through internal organizational processes. This model was empirically tested with survey data from 416 Australian manufacturing firms

23.2 Literature Review 23.2.1 Lead Time Reduction Lead time can generally be defined as the time period between the initiation of a task and its completion. More specific definitions depend on context. For example, in the new product development area, lead time is conceived in terms of the time it takes to identify a market need, design and test the product, and develop the processes for manufacturing (Tennant and Roberts, 2001; Yazdani and Holmes, 1999). Manufacturing lead time, on the other hand, is the “elapsed time between releasing an order and receiving it” (Hsu and Lee, 2008, p.1). A number of different terms are used to describe manufacturing lead time. Examples include “manufacturing throughput time” (Johnson, 2003) and “order-to-delivery time” (Zhang et al, 2007). From a SCM perspective, supply chain lead time is the “time spent by the supply chain to process the raw materials to obtain the final products and deliver them to the customer” (Bertolini et al, 2007, p.199). For the purpose of this study, only new product development and manufacturing lead times will be considered. There are two reasons for this. Firstly, these lead times are more directly under the control of firms; supply chain lead times can often be too difficult for individual firms to control and influence. Secondly, the interest is in more closely examining the concepts that come under “time-based competition”

23 Improving Lead Times Through Collaboration With Supply Chain Partners

295

(Stalk and Hout, 1990; Stalk, 1988). These include “fast-to-market” and “fast-toproduct”. Firms that compete with fast-to-market strategy emphasize reductions in new product development lead times. On the other hand, fast-to-product firms emphasize speed in responding to customer demands for existing products. This involves reducing the time it takes to manufacture products (throughput time) as well as the ability to reduce the time between taking a customer’s order and actually delivering the product (delivery speed). Studies have shown that lead times can involve tremendous amounts of waste, with about 85 percent of time spent waiting between value-adding steps (Holweg and Pil, 2004). Therefore, there are many benefits of reducing lead times. Ceteris paribus, firms that are able to reduce new product development lead times can gain a market edge over others that are not able to do same (Tennant and Roberts, 2001). Also, these firms move along the learning curve faster than their competition. Both these factors increase barriers to competitors. Methods for reducing this form of lead time include application of concurrent engineering practices (Yazdani and Holmes, 1999; Wilding and Yazdani, 1997), careful project management policies (Tennant and Roberts, 2001) and formal stagegate processes (Cooper, 1995). Similarly, reductions in manufacturing lead times can generate numerous benefits, including lower inventory levels, improved quality, reduced forecasting errors and increased flexibility (Johnson, 2003; Ouyang et al, 2007). These then improve customer service and satisfaction levels, which in turn contribute to competitive advantage accruing to the firms (Tersine and Hummingbird, 1995). Practical methods for generating manufacturing lead time reductions include process reengineering (Bertolini et al, 2007), reducing product variety (Zhang et al, 2007), implementing lean and JIT manufacturing practices (De Treville et al, 2004; Ouyang et al, 2007), using cellular manufacturing arrangements (Suri, 1998), and using ICT tools (Bertolini et al, 2007). A number of researchers have complained that many of these methods that have been suggested in the literature have not been taken up and implemented in firms for the purpose of reducing lead times (Suri, 1998; De Treville et al, 2004; Tersine and Hummingbird, 1995; Suri, 1999). Reasons for these include the complexity of some of these ideas, and the perceived difficulties associated with them. Hence, there is a need to develop more simpler and easy-to-implement ideas for reducing new product development and manufacturing lead times. Ideas from the SCM area have some potential.

23.2.2 Lead Time Reductions through Collaboration between Supply Chain Partners Collaboration involves firms working together and sharing resources and benefits, with the expectation of generating joint improvements in customer service, and achieving competitive advantage (Simatupang and Sridharan, 2002; Foster, 2005). SCM proponents claim that firms that collaborate with their customers and suppliers

296

Prakash J. Singh

are able to generate many benefits, including reduced new product development and manufacturing lead times (Christopher, 2005; Lee, 2004). The empirical evidence for these claims is in the form of case studies and survey type multi-organizational cross-sectional studies. For example, case studies of high profile firms such as Nokia (Heikkil¨a, 2002) and Toyota (Stalk, 1988) demonstrate the ability of these firms to achieve lead time reductions through, inter alia, pursuing close collaborative arrangements with their trading partners. In a similar vein, survey based studies show that collaboration has a positive impact on the operational (including lead time reductions) and financial performance of firms (Vickery et al, 2003; Wisner, 2003). Despite collaboration’s inherent attractiveness, many firms have found it is not all that easy to sustainably develop these types of relationships. There appears to be two main reasons for this. Firstly, collaboration is a relatively ’hard’ concept to implement because it requires a lot more effort from firms relative to other forms of inter-organizational relationships such as cooperation and coordination. Secondly, firms have predominantly focused on technical aspects of collaboration, with the belief that these systems would provide the infrastructure for strong relationships (Narasimhan and Kim, 2001; Patterson et al, 2003). As such, large investments have been made into establishing ICT systems (Bendoly and Kaefer, 2004). Relatively less investment has gone into establishing appropriate social systems (Burgess and Singh, 2006; Nahapiet and Ghoshal, 1998; Tsai and Ghoshal, 1998). But, as Spekman et al. Spekman et al (1998) and Golicic et al. Golicic et al (2003) contend, collaboration is as much to do with the technical system as it is with the social system. The relative neglect of social issues has led to firms not achieving the purported benefits of collaboration. For firms to be able to establish and benefit from collaboration, it is self-evident that the above mentioned issues need to be resolved. We attempt to (at least partially) address this in this paper. We focus on one key element of collaboration: the nature and extent of integration that firms need to enter into with their trading partners in order to realize significant benefits in the form of lead time reductions. Frohlich and Westbrook Frohlich and Westbrook (2002) separate this into supply and demand integration. Supply integration includes JIT (frequent, small lots) delivery, small supply base, suppliers selected on the basis of quality and delivery performance, longterm contracts with suppliers and elimination of paperwork. Demand integration, on the other hand, includes increased access to demand information throughout the supply chain to permit rapid and efficient delivery, coordinated planning and improved logistics communication.

23.2.3 Hypotheses Based on above discussions and purpose of this study, we hypothesize that firms that are able to develop high levels of collaboration (through integration) with supply chain partners would be able to generate significant reductions in both new product development and manufacturing lead times. Presenting this formally, it is hypothesized that:

23 Improving Lead Times Through Collaboration With Supply Chain Partners

297

• H1: Collaboration based internal organizational processes are positively related to lead time performance of firms. • H2a - H2b: Collaboration based relationships with customers are positively related to lead time performance of firms, both directly and through their internal organizational processes. • H2c - H2d: Collaboration based relationships with suppliers are positively related to lead time performance of firms, both directly and through their internal organizational processes. • H2e: Collaboration based relationships with suppliers and relationships with customers that firms develop are positively inter-related.. These hypotheses can be summarized in diagrammatic form as shown in Fig.23.1.

Fig. 23.1 Theoretical model

23.3 Research Method 23.3.1 Study Participants Data for the empirical testing of the above hypotheses was obtained through a postal survey targeting firms in the manufacturing industry in Australia. The JASANZ Register (Standards, 2004) was used for selecting the sample of firms. The unit of analysis was at the plant level. A target list of 1,053 unique plants was selected from the database. The respondents were senior managers (general, operations, quality,

298

Prakash J. Singh

production, etc). The survey was carried out in two stages. The final usable response rate was 41 percent (n=416). The study participants were predominantly small plants with almost half having fewer than 100 employees and $A10 million in annual revenue. Also, the plants were mainly from the machinery and equipment manufacturing (26 percent) and metal products (17 percent) manufacturing industry sub-categories.

23.3.2 Measurement Instrument The items used in this study was from a measurement instrument that was developed for a large study of quality and operations management practices (Singh, 2003).Since this instrument was original in many respects, a full set of tests for reliability and validity were performed to ensure that the various types of errors were within acceptable levels. These included pretest with eight practitioners and academicians and pilot test within 21 firms. A total of 146 items were present. Each of the items was measured on a five-point Likert scale.For this paper, a subset of the items relevant to the key constructs of internal organizational processes, relationships with suppliers, relationships with customers, and lead time performance was used. These constructs along with their associated items, and the scales that were used to measure the items, are shown in Table 23.1.

23.4 Data Analysis Procedures and Results 23.4.1 Psychometric Properties of Measurement Models Content Validity. The lists of items assigned to the constructs were based on literature (summarized in the Literature Review section earlier). This provided evidence that the items associated with the four constructs had sufficient grounding in relevant literature and therefore had content validity. Correlation Coefficients and Descriptive Statistics. The inter-item Pearson correlation coefficients were low to moderate in magnitude, suggesting that multicollinearity related problems were not present, with all coefficients being less than the threshold value of 0.9 (Hair, 2006). Further, the mean and standard deviation values of all the items suggested that the item measures did not suffer from excessive nonnormality. Reliability. The Cronbach’s alpha reliability coefficients for the constructs internal organizational processes, customer relationship, supplier involvement and firm performance constructs were 0.830, 0.654, 0.634 and 0.774 respectively. These coefficients exceeded the minimum threshold level of 0.6 for acceptable reliability (Nunnally and Bernstein, 1978) for all the constructs. Therefore, the selected items

23 Improving Lead Times Through Collaboration With Supply Chain Partners

299

Table 23.1 Constructs and associated items 1.

Internal organizational processes

IOP1 The quality assurance processes ensure that the organization’s own output requirements are consistently met IOP2 The actual manufactured products are checked against customer orders before they are delivered IOP3 The equipment to carry out tests and inspections are available when needed IOP4 Products and processes are inspected and/or tested IOP5 Processes that produce products that cannot be tested or inspected are continuously monitored IOP6 Everyone is aware of what needs to be done with raw materials, work-in-progress and finished products that fail inspections IOP7 The handling, storage, packaging and delivery methods help to minimize quality-related problems IOP8 It is possible to identify clearly when a raw material, work-in-progress or finished product has been inspected IOP9 It is possible to establish relevant details (such as parts suppliers, place and date of manufacture, persons-in-charge) of all finished products 2.

Relationships with customers

RC1 The organization is aware of the requirements of its customers RC2 Processes and activities of the organization are designed to increase customer satisfaction levels RC3 Misunderstandings between customers and the organization about customer orders are rare RC4 All contracts are systematically reviewed before acceptance, even if they are routine ones RC5 The organization has systematic processes for handling complaints RC6 Changes made to contracts lead to confusion in the organization (Reverse coded) 3.

Relationships with suppliers

RS1 The organization seeks assurance of quality from suppliers RS2 The main criterion for choosing suppliers is the quality of their products RS3 Misunderstandings between suppliers and the organization about orders placed with them are rare RS4 The quality of supplied products and services are assessed RS5 Materials provided by the customer for incorporation into products are treated the same as materials from any other suppliers 4.

Operating performance

OP1 Total lead-time of manufacturing products OP2 Time taken for new product development OP3 Delivery performance

reliably estimated the constructs. Convergent and Discriminant Validities. Convergent validity (i.e., the assigned items yield roughly the same results) and discriminant validity (i.e., the items estimate only the assigned construct and not any others) were both assessed by using a confirmatory factor analysis (CFA) model testing approach. The CFA model

300

Prakash J. Singh

is a structural equation model (SEM) where the constructs are all co-varied with R each other. The SEM analysis was conducted using the AMOS5.0 (Arbuckle and Wothke, 2004) software package. The maximum likelihood (ML) estimation technique was used to fit the model to the data because it is a reasonably scale- and distribution-free procedure (Hair, 2006). A number of commonly reported indices for assessing the goodness-of-fit of SEM models with data were obtained for the CFA model. These were as follows: χ 2 (246) = 876 with p-value ¡ 0.001; normed χ 2 = 3.567; goodness-of-fit index (GFI) = 0.844; adjusted goodness-of-fit index (AGFI) = 0.810; Tucker-Lewis index (TLI) = 0.764; comparative fit index (CFI) = 0.790; root mean square residual (RMR) = 0.045; and, root mean square error of approximation (RMSEA) = 0.079. The χ 2 fit measure has a tendency to produce negative results with sample sizes are greater than 200, and so was be disregarded. For all other fit indices, applying the cutoff criteria proposed by researchers (Hair, 2006; Marsh et al, 2004; Sharma et al, 2005; Schermelleh-Engel et al, 2003), the ’acceptable’ descriptor reasonably accurately captures the level of fit that has been obtained here. The parameters associated with the CFA showed the convergent validity of the constructs was generally supported; all the estimated factor loadings of items on constructs were significant (at p-values ¡ 0.001), the signs were all positive and only one was below 0.4, with the minimum being +0.346. Further, from the squared multiple correlation coefficient values, the variances of the items explained by their constructs were reasonably high (with the average being 34 percent). As for discriminant validity, correlations between the constructs were mostly moderate, suggesting that items assigned to one construct were not significantly highly loading on others.

23.4.2 SEM Results for Structural Model 23.4.2.1 Evaluation of Goodness-of-Fit Indices As with the CFA model, the SEM analysis procedure was used to assess the hypothesized relationships in the theoretical model presented in Fig.23.1. The fit indices for the hypothesized model were as follows: χ 2 (df=246) = 876 with p-value = 0.000; χ 2 /df = 3.567; GFI = 0.844; AGFI = 0.810; TLI = 0. 764; CFI = 0.790; RMR = 0.045; and, RMSEA = 0.079. These indices suggest that the theoretical model has adequate level of empirical support.1

1

To further confirm this result, a χ 2 difference test is traditionally used whereby the χ 2 values for the hypothesized model is compared with the CFA model (Anderson and Gerbing, 1988). However, in our case, the number of parameters in the hypothesized model is the same as that in the CFA, resulting in all fit indices being the same for the two models. Hence, this χ 2 difference test could not be meaningfully performed.

23 Improving Lead Times Through Collaboration With Supply Chain Partners

301

23.4.3 Evaluation of Theoretical Model Parameter Estimates Figure23.2 shows all the structural model parameters (regression, relevant squared multiple correlation, and correlation coefficients) within the theoretical model. These are in standardized form. The results have several noteworthy aspects. In terms of the magnitude and sign of the relationships, as Fig.23.2 shows, only two out of six relationships (H2a and H2c) are statistically insignificant in magnitude, having p-values greater than 0.05. The other relationships are all statistically significant in magnitude (p-values less than 0.05) and positive in sign. Also, the intercorrelation between the two exogenous constructs is positive, statistically significant and moderate in magnitude. Finally, the squared multiple correlation values for the two endogenous constructs, internal organizational processes and lead time performance, were 0.726 and 0.334 respectively. The exogenous constructs therefore accounted for large proportions of variances in these constructs. We further analyzed the regression and correlation data presented in Fig.23.2 by examining the standardized effect sizes between the constructs. Effect size is the increase/decrease in the endogenous construct (in standard deviation units) when there is a one standard deviation increase in the exogenous construct. The standardized direct effects, indirect effects (calculated using the path analysis tracing rules described by Kline (2005)) and total effects of all the exogenous constructs on the endogenous constructs of the model are shown in Table 23.2.

Fig. 23.2 Theoretical model, showing maximum likelihood estimates of standardized regression coefficients (on straight lines with single-arrowheads), squared multiple correlation coefficients (on constructs) and correlation coefficients (on curved lines with double-arrowheads.* 0.05 < p-value ≤ 0.1; ** 0.01 < p-value ≤ 0.05; *** p-value ≤ 0.01

302

Prakash J. Singh

Table 23.2 Effects decomposition of paths in the hypothesized model Exogenous construct:

Endogenous construct: Internal organizational processes Direct effect

Relationships with customers Relationships with suppliers Exogenous construct: Relationships with customers Relationships with suppliers Internal organizational processes

0.521 0.431

Indirect effect

Total effect

0.251 0.304

0.772 0.735

Endogenous construct: Operating performance Direct effect

Indirect effect

Total effect

0.156 0.065 0.398

0.783 0.38 -

0.939 0.445 0.398

23.5 Discussion The SEM model-data fit results suggested that in an overall sense, there was empirical support for SCM-based collaboration - lead time performance model as described in Fig.23.1. However, in terms of the specific hypothesized relationships, the same analysis showed that two direct hypothesized relationships (H2a: Relationship with customers → lead time performance, and H2c: Relationship with suppliers → lead time performance) were not supported. All other hypothesized relationships had empirical support. One could therefore conclude that SCM based collaboration strategies such as development of greater integration with supply chain partners are not so effective. This however would be an erroneous conclusion. Effects analysis results in Table23.2 show that the indirect effects are strong and more than compensate for the weak direct effects. As Table23.2 shows, the total effect of relationships with customers on lead time performance was 0.939, indicating that one standard deviation improvement in the relationships with customers construct is associated with 0.939 standard deviation change in lead time improvement. This is a very strong effect. Similarly, the total effect of relationships with suppliers construct on lead time performance was 0.445, indicating strong effect. These strong total effects indicate that collaboration practices combine in a synergistic manner to affect lead time performance. As a result, it would be myopic if each individual corroboration construct is assessed and evaluated for its effect in isolation. The literature on SCM collaboration has shown a number of benefits that firms can generate. To date, however, it has not been clear if lead time performance could also improve through supply chain collaboration. This study has shown that there is a strong albeit indirect link between supply chain collaboration practices and lead time performance. As such, firms can pursue such strategies with the understanding that a wide range of benefits that include lead time performance improvements can be achieved. Due to the manner in which the key supply chain collaboration constructs have been defined, measured and empirically validated, there is guidance to managers as to how these constructs can be operationalised in practice. From the items in

23 Improving Lead Times Through Collaboration With Supply Chain Partners

303

Table23.1, it is clear that much of the actual practices relating to supply chain collaboration are quite straightforward and simple in nature, and therefore realistic to achieve. One could indeed consider most of these items as being commonsensical and generally good practices that could be found in most well managed firms. As such, a key contribution of the paper is the articulation of a simple set of practices that enable and facilitate supply chain collaboration. Firms can be confident that if they put into practice the items listed in Table23.1, then by the empirical links established in the study, there will be a good chance that lead time performance improvements would follow. In doing so, this would overcome the belief that many managers have that the existing methods for lead time reductions are too complex and difficult to implement.

23.6 Conclusion In this study, the aim was to establish if it is possible to use ideas from the SCM field, specifically those relating to collaboration between firms and their trading partners, for the purpose of reducing lead times for new product development and manufacturing. The empirical data from a sample of 416 Australian manufacturing plants showed that this may indeed be possible, with data showing that relationships that firms develop with key customers and suppliers, acting directly and indirectly through internal organizational processes, strongly affected lead time performance. Given that lead time reduction has been recognized as playing a vital strategic role in enabling firms to successfully and sustainably compete, managers should therefore consider supply chain management collaboration as one more tool available to them, amongst other existing methods, for the purposes of generating lead time reductions.

References Anderson J, Gerbing D (1988) Structural equation modeling in practice: A review and recommended two-step approach. Psychological bulletin 103(3):411–423 R 5.0 users guide. Chicago: Smallwaters Arbuckle J, Wothke W (2004) Amos Bendoly E, Kaefer F (2004) Business technology complementarities: impacts of the presence and strategic timing of ERP on B2B e-commerce technology efficiencies. Omega 32(5):395–405 Bertolini M, Bottani E, Rizzi A, Bevilacqua M (2007) Lead time reduction through ICT application in the footwear industry: A case study. International Journal of Production Economics 110(1-2):198–212 Burgess K, Singh P (2006) A proposed integrated framework for analysing supply chains. Supply Chain Management: An International Journal 11(4):337–344 Christopher M (2005) Logistics and supply chain management: Creating valueadding networks, 3rd edn. London, Prentice Hall

304

Prakash J. Singh

Cooper RG (1995) Developing new products on time, in time. Research Technology Management 38(5):49–57 De Treville S, Shapiro R, Hameri A (2004) From supply chain to demand chain: The role of lead time reduction in improving demand chain performance. Journal of Operations Management 21(6):613–627 Fine C (1998) Clockspeed: Winning Industry Control in the Age of Temporary Advantage. Perseus Books Foster S Fand Srikanth (2005) Seven imperatives for successful collaboration. Supply Chain Management Review 9(1):30–37 Frohlich M, Westbrook R (2002) Demand chain management in manufacturing and services: Web-based integration, drivers and performance. Journal of Operations Management 20(6):729–745 Golicic S, Foggin J, Mentzer J (2003) Relationship magnitude and its role in interorganizational relationship structure. Journal of Business Logistics 24(1):57–76 Hair Ja (2006) Multivariate Data Analysis, 5th edn. New Jersey: Pearson PrenticeHall Heikkil¨a J (2002) From supply to demand chain management: Efficiency and customer satisfaction. Journal of Operations Management 20(6):747–767 Holweg M, Pil F (2004) The second century: Reconnecting customer and value chain through build-to-order. MIT Press Cambridge, MA Hsu SL, Lee CC (2008) Replenishment and lead time decisions in manufacturerretailer chains. Transportation Research Part E: Logistics and Transportation Review 45(3):398 – 408 Johnson D (2003) A framework for reducing manufacturing throughput time. Journal of Manufacturing Systems 22(4):283–298 Kline R (2005) Principles and practice of structural equation modeling, 2nd edn. Guilford Press, New York Lee H (2004) The triple-A supply chain. Harvard Business Review (10):102–112 Marsh H, Hau K, Wen Z (2004) In search of golden rules: Comment on hypothesistesting approaches to setting cutoff values for fit indexes and dangers in overgeneralizing Hu and Bentler’s (1999) findings. Structural Equation Modeling 11(3):320–341 Nahapiet J, Ghoshal S (1998) Social capital, intellectual capital and the organizational advantage. Academy of management review 23(2):242–266 Narasimhan R, Kim S (2001) Information system utilization strategy for supply chain integration. Journal of Business Logistics 22(2):51–76 Nunnally J, Bernstein I (1978) Psychometric theory, 2nd edn. New York: McGrawHill Ouyang L, Wu K, Ho C (2007) An integrated vendor–buyer inventory model with quality improvement and lead time reduction. International Journal of Production Economics 108(1-2):349–358 Patterson K, Grimm C, Corsi T (2003) Adopting new technologies for supply chain management. Transportation Research Part E: Logistics and Transportation Review 39(2):95 –121

23 Improving Lead Times Through Collaboration With Supply Chain Partners

305

Schermelleh-Engel K, Moosbrugger H, M¨uller H (2003) Evaluating the fit of structural equation models: Tests of significance and descriptive goodness-of-fit measures. Methods of psychological research online 8(2):23–74 Sharma S, Mukherjee S, Kumar A, Dillon W (2005) A simulation study to investigate the use of cutoff values for assessing model fit in covariance structure models. Journal of Business Research 58(7):935–943 Simatupang T, Sridharan R (2002) The Collaborative Supply Chain. International Journal of Logistics Management 13(1):15–30 Singh P (2003) What Really Works in Quality Management: A Comparison of Approaches. Consensus Books, Sydney Spekman R, Kamauff J, Myhr N (1998) An empirical investigation into supply chain management: A perspective on partnerships. International Journal of Physical Distribution & Logistics Management 28(8):630–650 Stalk G (1988) Time–The next source of competitive advantage, vol 66. Harvard Business Review Stalk G, Hout T (1990) Competing against time: How time-based competition is reshaping global markets. New York : Free Press Standards A (2004) Joint Accredited Systems - Australia and New Zealand (JASANZ) Register [cited Oct 2004]. URL http://www.jas-anz.com.au Suri R (1998) Quick response manufacturing: A companywide approach to reducing lead times. OR., Productivity Press, Portland, OR, Portland Suri R (1999) How Quick Response Manufacturing Takes the Wait Out. Journal for Quality & Participation 22(3):46–49 Tennant C, Roberts P (2001) A faster way to create better quality products. International Journal of Project Management 19(6):353–362 Tersine R, Hummingbird E (1995) Lead-time reduction: The search for competitive advantage. International Journal of Operations and Production Management 15(2):8–18 Tsai W, Ghoshal S (1998) Social capital and value creation: The role of intrafirm networks. Academy of management journal 41(4):464–476 Vickery S, Jayaram J, Droge C, Calantone R (2003) The effects of an integrative supply chain strategy on customer service and financial performance: An analysis of direct versus indirect relationships. Journal of Operations Management 21(5):523–539 Wilding R, Yazdani B (1997) Concurrent Engineering in the Supply-Chain. Logistics Focus 5:16–22 Wisner J (2003) A structural equation model of supply chain management strategies and firm performance. Journal of Business Logistics 24(1):1–25 Yazdani B, Holmes C (1999) Four models of design definition: sequential, design centered, concurrent and dynamic. Journal of Engineering Design 10(1):25–37 Zhang X, Chen R, Ma Y (2007) An empirical examination of response time, product variety and firm performance. International Journal of Production Research 45(14):3135–3150

Appendix A

International Scientific Board

The chair of the international scientific board of the 1st rapid modelling conference ”Increasing Competitiveness - Tools and Mindset” consisted of: • Gerald Reiner (University of Neuchˆatel, Switzerland) Members of the international scientific board as well as referees are: • • • • • • • • • • • • • •

Djamil A¨ıssani (LAMOS, University of B´ejaia, Algeria) Michel Bierlaire (EPFL, Switzerland) Bnamar Chouaf (University of Sidi Bel Abes, Algeria) Lawrence Corbett (Victoria University of Wellington, New Zealand) Krisztina Demeter (Corvinus University of Budapest, Hungary) Suzanne de Treville (University of Lausanne, Switzerland) Gerard Gaalman (University of Groningen, The Netherlands) Petri Helo (University of Vaasa, Finland) Werner Jammernegg (Vienna University of Economics and Business Administration, Austria) Matteo Kalchschmidt (University of Bergamo, Italy) Doug Love (Aston Business School, UK) Jose Antonio Dominguez Machuca (University of Sevilla, Spain) Jeffrey S. Petty (Lancer Callon, UK) Boualem Rabta (University of Neuchˆatel, Switzerland)

307

Appendix B

Sponsors

The sponsors of the 1st rapid modelling conference ”Increasing Competitiveness Tools and Mindset” are: • • • • • • • •

Journal of Modelling in Management http://www.emeraldinsight.com/jm2.htm LANCER CALLON http://www.lancercallon.com REHAU http://www.rehau.at SOFTSOLUTION http://www.softsolution.at SWISS OPERATIONS RESEARCH SOCIETY http://www.svor.ch THENEXOM http://www.thenexom.net University of Lausanne, Faculty of Business and Economics (HEC) http://www.hec.unil.ch University of Neuchˆatel, Faculty of Economics http://www.unine.ch/seco

309