Alfred Angerer The Impact of Automatic Store Replenishment on Retail
GABLER EDITION WISSENSCHAFT
Alfred Angerer
Th...
47 downloads
952 Views
8MB 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
Alfred Angerer The Impact of Automatic Store Replenishment on Retail
GABLER EDITION WISSENSCHAFT
Alfred Angerer
The Impact of Automatic Store Replenishment on Retail Technologies and Concepts for the Out-of-Stocks Problem
With a foreword by Prof. Dr. Daniel Corsten
Deutscher Universitats-Verlag
Bibliografische Information Der Deutschen Bibliothek Die Deutsche Bibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet iJber abrufbar.
Dissertation Universitat St. Gallen, 2005
I.Auflage April 2006 Alle Rechte vorbehalten © Deutscher Universitats-Verlag/GWV Fachverlage GmbH, Wiesbaden 2006 Lektorat: Brigitte Siegel / Sabine Scholler Der Deutsche Universitats-Verlag ist ein Unternehmen von Springer Science+Business Media. www.duv.de Das Werk einschlieBlich aller seiner Telle ist urheberrechtlich geschiitzt. Jede Verwertung auSerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustlmmung des Verlags unzulassig und strafbar. Das gilt insbesondere fur Vervielfaltlgungen, Ubersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Die Wiedergabe von Gebrauchsnamen, Handelsnamen, Warenbezeichnungen usw. in diesem Werk berechtigt auch ohne besondere Kennzeichnung nicht zu der Annahme, dass solche Namen im Sinne der Warenzeichen- und Markenschutz-Gesetzgebung als frei zu betrachten waren und daher von jedermann benutzt werden diirften. Umschlaggestaltung: Regine Zimmer, Dipl.-Designerin, Frankfurt/Main Druck und Buchbinder: Rosch-Buch, ScheBlitz Gedrucktauf saurefreiem und chlorfrei gebleichtem Papier Printed in Germany ISBN 3-8350-0302-X
dedicado a las dos mujeres mas importantes de mi vida: mi madre y Anne
VII
Foreword Fast moving consumer goods retailing is a highly competitive market. European retailers are continuously aiming to improve customer loyalty by offering good service. At the same time, they are struggling to reduce costs in order to stay competitive. One technology that promises to decrease the number of out-of-stocks while simultaneously reducing store handling costs is automatic store replenishment (ASR). At the heart of ASR systems lies software that automatically places an order to replenish stocks. Many European grocery retailers have started to implement such decision support systems.
Surprisingly, although several retailers have automated their order process in the last few years, there is almost no academic source examining this topic at the level of the store. It is worth noting that other technologies In retail, such as RFID (Radio Frequency Identification) and the introduction of the barcode, have received far greater attention from the public and from researchers. Furthermore, while the topic of extent and root-causes of retail out-of-stock has received substantial interest over the course of the last years, the question to what extent existing and new practices remedy OOS is largely unanswered. In particular, there is a debate whether ASR improve or worsen OOS. Therefore, Dr. Alfred Angerer has well chosen a topic of both managerial and academic relevance.
Although there are many success stories from practitioners describing the enormous advantages of introducing automatic store replenishment systems there has been limited empirical proof of this. To the best of my knowledge no conceptual framework exits that can help practitioners to choose an adequate automatic replenishment system. In order to develop such a model research on relationship between replenishment
performance
(e.g. OOS rate, inventory levels) and contextual
variables (such as store and product characteristics) is required. Finally, it is not clear how retailers have to adapt its organization and processes to best support the chosen ASR system.
Dr. Angerer confidently identifies and covers several research gaps and manages to give answers to this research gaps by a skilful combination of quantitative and qualitative research methodologies. In a first part an exhaustive data set of a European retailer is examined. With this data analysis the performance of replenishment system before and after the introduction of ASR systems is compared.
VIII
Dr. Angerer is able to statistically prove and quantify the positive impact of such systems on inventory levels and out-of-stock rates. In the second part, several case studies illustrate how ASR systems are implemented In practice. The given recommendations on store processes help retailers to benefit most from automatic replenishment systems.
Overall, this thesis makes an important contribution to the field of retail operations in practice and theory. I personally wish Dr. Angerer's work wide attention in both academic and practitioner circles.
Prof. Dr. Daniel Corsten
IX
Acknowledgment Rarely is a doctoral thesis the contribution of a single person. Many people supported and consulted me during my three years of research at the University of St.Gallen. Therefore, I would like to express my thanks to everyone who supported me in finalising this work. I am greatly indebted to my two advisors, Prof. Dr. Daniel Corsten and Prof. Dr. Fritz Fahrni. They guided me through the inevitable ups and downs that characterise such a long research project. I want to specially thank Daniel Corsten who supported my research from the very outset. Without his never ending striving for improvement, the present results would not have been gained. I also cordially thank my second adviser, Fritz Fahrni, who always helped me to look for the big picture in my work. Further, I am thankful to Prof. Dr. Frank Straube and Prof. Dr. Wolfgang Stolzle for their backup as directors of the KLOG. A special thanks goes to my colleagues Lars Dittmann and Christian Tellkamp, with whom I shared several "research camps". They decisively influenced my research. Jens Hamprecht deserves a special thank for his numerous advices as well as Dorothea Wagner does. Her extensive knowledge about the consumer goods industry was always a valuable contribution. My time at the University of St.Gallen would only have been half the fun without my colleagues. I want to thank Gunther Kucza, Marion Peyinghaus, Jorg Hofstetter, Jan Felde, Jan Frohn, Elias Halsband, Florian Hofer, Petra Seeger, Dirk Voelz and all other colleagues at the Kuehne-lnstitute for Logistics and the Institute for Technology Management for being such good colleagues and for all the good moments we shared. Throughout the last years, I received valuable contributions from researchers and students. Especially, I want to thank Johanna Smaros and Michael Faick from the HUT for the enriching discussions and research projects I shared with them. I would also like to express my thanks to all the students whose bachelor and master thesis I coached. Their interviews provided a basic foundation for my research. Without the support from practitioners, this research would only have been a theoretical contribution. A very warm thank you to Roland S., who invested his time to provide me the data for the quantitative research. Further, I am very grateful to
Marianne S., Daniel B. and all the interview partners for the time they invested in my research project. Finally, I want to thank my mother and father, Toni, Lydia, Nic and Anne for their never-ending moral support. Despite the distance, I always felt their affection throughout my education and career. St. Gallen, November 2005
Alfred Angerer
XI
Content Overview 1. Introduction 1.1. 1.2. 1.3. 1.4. 1.5. 1.6.
Logistics Contribution to Retail Excellence Excellence in Store Operations New Technologies Enable Automatic Store Replenishment Systems Research Deficit Research Questions Thesis Structure
2. Research Framework and Design 2.1. Research Framework 2.2. Research Methodology 2.3. Research Process 3. Literature Research 3.1. 3.2. 3.3. 3.4. 3.5.
Inventory Management Perspective Logistics and Operations Management Perspective Business Information Systems Perspective Contingency Theory Perspective Literature Research Overview
4. Development of Models 4.1. A Descriptive Model of Replenishment Systems 4.2. Classification of Automatic Replenishment Systems 4.3. Explanatory Model 5. Quantitative Analysis 5.1. 5.2. 5.3. 5.4.
1 1 3 6 8 12 14 16 16 18 23 26 26 30 41 47 51 53 53 64 68 85
Sample and Methodology Hypothesis Testing: Datasetl Dataset2: Pretest/Posttest Quantitative Research-Conclusions
85 99 122 728
6. Field Research and Managerial Implications
132
6.1. 6.2. 6.3. 6.4. 6.5.
Research Sample Replenishment Processes Organizational Changes and Personnel Issues ASR Performance Lessons Learned and Recommendations for Management
7. Conclusion 7.7. Theoretical Contributions 7.2. Contribution for Practitioners 7.3. Further Research Fields
732 737 757 757 767 181 787 783 786
8. Appendix and References
189
8.7. Statistical Appendix 8.2. References 8.3. List of Interviews
789 793 270
XIII
Table of Contents 1. Introduction 1.1. 1.2. 1.3. 1.4. 1.5. 1.6.
Logistics Contribution to Retail Excellence Excellence in Store Operations New Technologies Enable Automatic Store Replenishment Systems Research Deficit Research Questions Thesis Structure
2. Research Framework and Design 2.1. Research Framework 2.2. Research Methodology 2.3. Research Process 3. Literature Research
1 1 3 6 8 12 14 16 16 18 23 26
3.1. Inventory Management Perspective 26 3.1.1. Optimization in Inventory Management Research 27 3.1.2. Theoretical Sources on OOS 28 3.1.3. Contributions and Deficits of an Inventory Management Perspective 29 30 3.2. Logistics and Operations Management Perspective 3.2.1. Supply Chain Management and ECR 31 3.2.2. Automatic Replenishment Programmes 33 3.2.3. Operations Management in Retail 39 3.2.4. Contributions and Deficits of a Logistics and Operations Management Perspective 40 3.3. Business Information Systems Perspective 41 3.3.1. Characteristics of ERP Systems 42 3.3.2. ERP Implementation and Selection 42 3.3.3. ERP and Human Agency 44 3.3.4. Contributions and Deficits of a Business Information Systems Perspective 46 3.4. Contingency Theory Perspective 47 3.4.1. Contingency Theory at the Organizational Level 47 3.4.2. Contingency Theory on Information Technology and Processes 49 3.4.3. Contributions and Deficits of a Contingency Perspective 50 3.5. Literature Research Overview 51 4. Development of Models 4.1. A Descriptive Model of Replenishment Systems 4.1.1. Inventory Visibility 4.1.2. Replenishment Logic 4.1.3. Order Restrictions 4.1.4. Forecasts 4.2. Classification of Automatic Replenishment Systems 4.3. Explanatory Model 4.3.1. Purpose and Structure of the Explanatory Model 4.3.2. Hypothesis Development: Product Characteristics 4.3.3. Hypothesis Development: Store Characteristics
53 53 56 57 60 61 64 68 68 71 77
XIV
4.3.4. Hypothesis Development: ASR Characteristics 5. Quantitative Analysis 5.1. Sample and Methodology 5.1.1. Datasetl: Testing of Out-of-Stock Hypotheses 5.1.2. Dataset2: Pretest/Posttest Analysis 5.2. Hypothesis Testing: Datasetl 5.2.1. Influence of Product Characteristics 5.2.2. Influence of Store Characteristics 5.3. Dataset2: Pretest/Posttest 5.4. Quantitative Research-Conclusions 6. Field Research and Managerial Implications 6.1. Research Sample 6.1.1. Company Selection 6.1.2. Market Characteristics 6.1.3. Supply Chain Structure 6.1.4. Chains and Store Formats 6.1.5. Delivery Frequency and Order-to-Deliver Lead Times 6.2. Replenishment Processes 6.2.1. Inventory Visibility 6.2.2. Forecasts and Replenishment Logic 6.2.3. Order Restrictions 6.3. Organizational Changes and Personnel Issues 6.3.1. Structural Changes and Setup 6.3.2. Personnel and Change Management 6.4. ASR Performance 6.4.1. Performance Measurement 6.4.2. Inventory Level Performance 6.4.3. OOS Reduction and Overall Performance 6.5. Lessons Learned and Recommendations for Management 6.5.1. The Adequate Automation Level: Recommendations 6.5.2. ASR Introduction 6.5.3. Technical and Organizational Requirements 6.5.4. Store Operations: Recommended Action 6.5.5. Cost-Benefit Analyses 7. Conclusion 7.7. Theoretical Contributions 7.2. Contribution for Practitioners 7.3. Further Research Fields 8. Appendix and References 8.7. Statistical Appendix 8.1.1. Calculation of the Inventory Level 8.1.2. ANOVA Considerations and Prerequisites 8.2. References 8.3. List of Interviews
81 85 85 85 93 99 99 113 722 728 132 732 132 134 134 136 136 737 138 146 151 757 152 154 757 157 158 159 767 161 165 166 170 175 181 787 783 786 189 789 189 191 793 270
XV
List of Abbreviations and Acronyms ANOVA ARP ASR ASRx CD CU CU/TU CRP CPFR BAN ECR EDI ERP DC DSD DSS HQ IS IT ITEM KLOG KPI MAD MAPE COS OR OSA PC PDA POS QR SC SCM SKU TU VMI
Analysis Of Variance Automatic Replenishment Programme Automatic Store Replenishment Automatic Store Replenishment System level x Cross-Docking Consumer Unit Consumer Unit per Trading Unit (=case pack size) Continuous Replenishment Planning Collaborative Planning Forecasting and Replenishment European Article Numbering Efficient Consumer Response Electronic Data Interchange Enterprise Resource Planning Distribution Centre Direct Store Delivery Decision Support System Headquarters Information System Information Technology Institute for Technology Management Kuehne-lnstitute for Logistics Key Performance Indicator Mean Absolute Deviation Mean Absolute Percent Error Out-Of-Stock Operations Research On-Shelf Availability Personal Computer Personal Digital Assistant (handhelds) Point Of Sales Quick Response Supply Chain Supply Chain Management Stock Keeping Unit Trading Unit Vendor Managed Inventory
XVII
Figures Figure 1: The importance of logistics for different industries
2
Figure 2: Percentage of logistics costs on total costs by industry (in %)
3
Figure 3: Summary of OOS root causes
5
Figure 4: Thesis structure
15
Figure 5: Focus of research
16
Figure 6:
19
Integrative research procedure
Figure 7: Case study research as iterative process between theory and empiricism
22
Figure 8: Research activities in this research project
23
Figure 9: Spectrum of misfit resolution strategies
44
Figure 10: Descriptive model of replenishment systems
54
Figure 11: Exemplary time dependent course of the inventory stock level
56
Figure 12: Qualitative and quantitative forecasting techniques
62
Figure 13: Classification of automatic replenishment systems
65
Figure 14: Overview of hypotheses, product characteristics
83
Figure 15: Overview of hypotheses, store characteristics
84
Figure 16: Overview of hypotheses, ASR characteristics
84
Figure 15: Distribution of the 84 products in Datasetl
88
Figure 16: Comparison of the replenishment systems by store
92
Figure 17: Estimated OOS (order-related) rate by sales coefficient of variance
100
Figure 18: Estimated inventory range of coverage by sales coefficient of variance Figure 19: Estimated OOS (order-related) rate by speed of turnover
102 103
Figure 20: Estimated inventory range of coverage by speed of turnover
104
Figure 21: Estimated OOS (order-related) rate by price
105
Figure 22: Estimated inventory range of coverage by price
106
Figure 23: Estimated OOS (order-related) rate by CU/TU group
107
Figure 24: Estimated Inventory range of coverage by case pack
108
Figure 25: Estimated OOS (order-related) rate by product size
109
Figure 26: Estimated inventory range of coverage by product size
110
Figure 27: Estimated OOS (order-related) rate by shelf life
Ill
Figure 28: Estimated inventory range of coverage by shelf life
112
Figure 29: Estimated OOS (order-related) rate by store
114
Figure 30: Estimated inventory range of coverage by store
115
Figure 31: Estimated inventory coefficient of variance by store
116
Figure 32: Relationship between shrinkage and OOS
117
Figure 33: Relationship of OOS and SKU density
118
XVIII
Figure 34: Relationship of OOS and number of personnel perm^
118
Figure 35: Relationship of OOS and number years of the store manager working in the store
119
Figure 36: Relationship of OOS and the size of the backroom
120
Figure 37: Relationship of OOS and customer satisfaction
121
Figure 38: Mean inventory range of coverage in days of dairy products and Controll
123
Figure 39: Means of the repeated ANOVA on the coefficient of variance of the stock level, ASR3 group and Controll Figure 40: Mean
inventory
range
of
coverage
125 in
days
of
non-food
products and Control2
126
Figure 41: Estimated means of the repeated ANOVA on the inventory range of coverage, ASR2 group Figure 42: Estimated
means
of the
127 repeated
ANOVA
on the
inventory
range of coverage, ASR2* group
127
Figure 43: Mean inventory coefficient of variance in days of ASR2, ASR2* group andControl2
128
Figure 44: Supply chain structure of sample
135
Figure 45: Delivery frequency of sample
137
Figure 46: Inventory storage places and product flow processes
138
Figure 47: Comparison of inventory records and real inventory in one store
142
Figure 48: Decision tree for practitioners
162
Figure 49: Cost of forecasting versus cost of inaccuracy
168
Figure 50: Overview of store operations recommendations
170
Figure 51: Comparison of inventory on shelf and total store inventory for a glue stick
172
Figure 52: Costs in relation to replenishment level
177
Figure 53: Theoretical contribution of thesis
181
Figure 54: Contribution for practitioners
183
Figure 55: Relative inventory level curve without zero line
189
Figure 56: Absolute inventory level curve after the correction
190
XIX
Tables Table 1: Overview of research deficits
12
Table 2: Overview of basic theoretical sources reviewed (excerpt)
26
Table 3: Implementation of ARP-related items
35
Table 4: Effectiveness in achieving automatic replenishment-related goals
36
Table 5: Information systems capabilities
37
Table 6: Summary of research streams perspectives
52
Table 7: Inventory notations
55
Table 8: Basic inventory decision rules
57
Table 9: Exemplary order restrictions
60
Table 10: Characteristics of automatic replenishment levels
68
Table 11: Overview of hypotheses concerning product characteristics
77
Table 12: Overview of hypotheses concerning store characteristics
81
Table 13: Overview of the utilization of the two datasets for hypothesis testing
85
Table 14: Overview of variables used in the analysis
90
Table 15: Product characteristics of Datasetl by replenishment system
91
Table 16: OOS rates (order-related)of the sample
92
Table 17: Inventory range of coverage of Datasetl
93
Table 18: Dataset for the pretest/posttest
94
Table 19: Descriptive statistics of the dairy products (ASR3) and Controll group (ASRO)
95
Table 20: Descriptive statistics of the beauty and household group (ASR2) and Control2 (ASRO)
97
Table 21: Descriptive statistics of the non-food group (ASR2*) and Control2 (ASRO)
97
Table 22: ASR level and sales coefficient of variance ANOVA on OOS (order-related)
100
Table 23: ASR level and sales coefficient of variance ANOVA on inventory range of coverage Table 24: ASR level and speed of turnover ANOVA on OOS (order-related)
101 103
Table 25: ASR level and speed of turnover ANOVA on inventory range of coverage
104
Table 26: ASR level and price ANOVA on OOS (order-related)
105
Table 27: ASR level and price ANOVA on inventory range of coverage
106
Table 28: ASR level and CU/TU ANOVA on OOS (order-related)
107
Table 29: ASR level and CU/TU on inventory range of coverage
108
Table 30: ASR level and product size ANOVA on OOS (order-related)
109
Table 31: ASR level and product size ANOVA on inventory range of coverage ...110
XX
Table 32: Regression of shelf life and shelf life squared on OOS
112
Table 33: Correlation between OOS per week and store characteristics (Datasetl)
113
Table 34: ASR level and Store ANOVA on OOS (order-related)
114
Table 35: ASR level and Store ANOVA on inventory range of coverage
115
Table 36: ASR level and store ANOVA on inventory coefficient of variance
116
Table 37: ASR level ANOVA on OOS (order-related)
121
Table 38: ASR level ANOVA on inventory range of coverage
122
Table 39: Performance of ASR3 group compared to the Control 1 (ASRO)
124
Table 40: Repeated ANOVA on inventory range of coverage, ASR3 group
124
Table 41: Repeated ANOVA on the coefficient of variance of the stock level, ASR3 group Table 42: Repeated
ANOVA
125 on
mean
inventory
range
of
coverage,
ASR2 group
126
Table 43: Results overview: product characteristics hypotheses
129
Table 44: Results: store characteristics hypotheses
130
Table 45: Results: ASR hypotheses
130
Table 46: Overview of the results of the hypotheses tested
131
Table 47: Selected companies for the field research
133
Table 48: Inventory range of coverage of European grocery retailers in days
158
Table 49: Technical requirements and recommendations on operations and organization structure Table 50: Overview
of
possible
169 benefits
and
system introduction Table 51: Overviewof further research opportunities
costs following
an
ASR 177 186
XXI
Abstract European fast moving consumer goods retailers face a mature market with low margins and high competition. To improve their situation, retailers are looking for technologies and concepts to increase consumer satisfaction while at the same time reducing costs. One technology that promises to increase the availability of the products on the shelf while simultaneously reducing store handling costs is automatic store replenishment (ASR). At the heart of ASR systems lies software that automatically places an order to replenish stocks of a certain product. A majority of European grocery retailers have implemented such decision support systems. Yet research in this area is practically non-existent. Therefore, this thesis aims to investigate the impact of this technology on retail, taking into account financial, organizational and personnel aspects. To answer this main research question, a quantitative and a qualitative methodology was chosen. First of all, based on theoretical sources and more than 50 interviews, a descriptive model and an ASR classification system is developed. Next, an explanatory model is developed with a view to enabling identification of the characteristics of products, stores and replenishment systems that influence the replenishment performance of retail stores. To be able to test the hypothesis derived from this explanatory model, exhaustive data from a grocery retailer is examined. The quantitative analysis clearly shows that even simple automatic replenishment systems are able to dramatically reduce the average shelf out-of-stock rate and at the same time lower inventory level. In addition, a major advantage of automatic systems over manual ones is that they show constant results, independently of product characteristics. Yet the analysis also shows that badly-parameterised automatic systems will fail to deliver the desired results. In order to better understand how ASR systems are best implemented in practice, four major grocery retailers are analysed in detail. These case studies illustrate the necessary technological and organizational changes and highlight the influence of ASR systems on the working behaviour of employees. Overall, this thesis makes contributions to both practice and theory. On the one hand, the results presented are a first stepping stone towards the creation of a basic theory of ASR systems. A descriptive model enables further researchers to make differentiated statements on the impact of ASR based on the classification developed. Another contribution is the explanatory model which tests existing and demonstrates
new
relationships
hypothesised
in
inventory
and
operations
management research. On the other hand, practitioners receive an overview of the
XXII
existent
systems
by which they
may automate
store
replenishment.
The
determination of ASR benefits and necessary requirements help them to make a cost-benefit analysis. In addition, the several implications of the automation of their replenishment system for the organization and for human working patterns are illustrated. Practical recommendations on store processes help retailers to benefit most from automatic replenishment systems. And finally, a decision tree helps practitioners to identify the best-suited ASR system for each product category.
1. Introduction
1. Introduction Grocery retailing is a highly competitive market (e.g. Keh and Park 1997). European retailers are continuously aiming to improve customer loyalty by offering good service. At the same time, they are struggling to reduce costs in order to stay competitive. The effort to achieve customer service excellence has only been partly successful, as the low average product shelf availability rates of 92-95% (Gmen, Corsten et al. 2002; Roland Berger 2003b) and a sunk store loyalty underline. The major part of retailer costs are personnel costs, and in particular it is the operations In the store that require intensive staff dedication (Broekmeulen, van Donselaar et al. 2004a). The German retailer Globus has calculated that the logistics costs of the last 50 meters in the store, i.e. from the backroom to the shelf, are three times as expensive as the first 250 kilometres from the producer to the store gate (Shalla 2005). A technique that promises to reduce the out-of-stock (OOS) rate by simultaneously reducing the store handling costs are so-called automatic store replenishment (ASR) systems, the main research subject of this thesis. This chapter provides an introduction to the business challenges faced by retailers and the valuable role of logistics in retail, followed by a short introduction to ASR systems. Later, research deficits in the literature are identified and the research questions of this thesis are derived. Finally, an overview of the structure of this research study is given.
1.1. Logistics Contribution to Retail Excellence The major market developments that make retail challenging started in the 1990s and still are prevalent today, namely high cost pressure, shorter innovation cycles, increasing consumer expectations and globalization (Baumgarten and Wolf 1993; Lee 2001). The common response of retailers has been a so-called quantity strategy: They introduced more product variants, invested in new channels of distribution, diversified store formats and expanded into new countries. However, the benefits harvested from such a strategy seem to have come to an end, as the market has become saturated. The fraction of private consumption that flows into food and nearfood retail has decreased continuously in the last two decades. In Germany, for example, it sank from 44.2% in 1990 to 29.3% in 2004 (Korber 2003), and this trend is typical for many developed countries. Nevertheless, a small group of retailers was able to defy this trend and outperformed the market. As a study by Accenture (2000) reports, approximately one-third of 63 examined retailers outperformed the other two-thirds by far and showed a yearly revenue increase of at least 10% coupled with
1. Introduction
a higher-than-average Increase In stock price. According to the study, this group had developed the right strategy by focusing their Investments In areas where the most efficiency potentials were located. One of the areas with such potential Is without doubt logistics, as effective and efficient logistics Is the fundamental to successful retailing. Hans Joachim Korber (2003), CEO of Metro AG, describes logistics as "the physical accomplishment of the concern strategy." Figure 1 depicts the great Importance of logistics for retail and various industry sectors under the aspects "differentiation" (I.e. logistics as a marketing tool) and "rationalisation" (i.e. logistics as a method of saving costs).
Figure 1: The importance of logistics for different industries^
The importance of logistics for the retail sector Is based on the nature of the products sold. Most consumer goods, for example dally food Items, are relatively cheap and the consumer generally buys without lengthy quality or price comparisons. Nevertheless, the Importance of logistics In other sectors Is Increasing as well, as Pfohl (2004) stresses.
^ Source: Kowalski (1992).
1. Introduction
A precise estimation of the logistics costs is rather difficult. Pfohl compared studies measuring the logistics costs as a percentage of turnover. The large differences in the results can often be explained by geographical differences between countries and their infrastructure levels. Yet even within a single country like Germany, there are several studies with significantly divergent figures. This Is the result of the varying definition of logistics costs. One of the most cited studies is that by Baumgarten and Thoms (2002). They estimate the retailers' logistics costs at up to 27% of total costs (see Figure 2).
26.7% Figure 2: Percentage of logistics costs on total costs by industry (in %)
Even if other researchers have clearly lower estimations (e.g. Klaus 2003), there is a common agreement that there exists a large savings potential. Two studies from the year 1999 estimate the savings potentials at about 12-25% (Baumgarten and Wolf 1993; European Logistics Association and A.T. Kearney 1999). In order to achieve these savings, new advanced logistics-technology is employed. But logistics should never be reduced to its cost-reducing effect, as logistics concepts can also be utilized to improve service and consequently increase sales (Angerer and Corsten 2004). The next section deals with one of the most important measures used to quantify customer-service levels: the on-shelf availability rate.^
1.2. Excellence in Store Operations A high availability rate of products on the shelves is of utmost importance for retailers. All the efforts made to improve the supply chain are futile if, in the end, the consumer is unable to buy the product because it is not available on the shelf. There Source: Baumgarten and Thoms (2002). ^ The on-shelf availability rate is the percentage of products that are available for purchasing on the store's shelves at a particular moment in time.
1. Introduction
exist studies that show that out-of-stocks (OOS) in stores are the most frequently mentioned cause of frustration for dissatisfied customers in retail (Sterns, Unger et al. 1981). Interviews with practitioners confirm the importance of high shelf availability: "The three criteria that decide the success of a product are the right price, the right forms of advertisement and high on-shelf availability. (...) In particular, if there is a promotion, there is nothing more important than having the goods on the shelfl"^ Obviously, the impact of an OOS depends on the reaction of the customer: "The reaction of customers [on OOS] differs a great deal. If the customer buys a different brand, we are happy. If he or she does not buy anything at all, then we are not content. And if the customer buys the product in a competitor's store, that is a catastrophe! Seventy percent of customers change to the competition for good if they experience repeated OOS; and that is a complete catastrophe!'^ Furthermore, there is a strategic component to high shelf availability, as it ensures an advantage in increasingly competitive markets: "If we want to compete with new aggressive retailers such as LIDL which are planning to enter the Swiss market, we have to increase the turnover per square meter. For that, we need to increase the on-shelf availability (...) to make our stores more interesting for customers.'^ The importance of a high availability is underlined by the research of Dr6ze, Hoch et al. (1994) among others. They show that the total amount of money spent on any store visit is an elastic quantity and is highly dependent on product presentation and quantity on the shelf. Although the on-shelf availability rate plays such an Important role in the business of retailers, it seems that only a minority of European grocery retailers systematically measures this important KPI (key performance indicator). A case study of 12 leading European grocery retailers has shown that only four companies have established a process for daily availability check (Sm^ros, Angerer et al. 2004a). Only one retailer had implemented an electronic-based system for automatic checks. The magnitude of the OOS problem still appears not to have been identified by many retailers. They tend to derive the availability rate in their stores
"* Source: Arthur Mathys, Director Logistics, Denner, 04.08.2003. ^ Source: Wolfgang MShr, Director IT, Spar Switzerland, 16.02.2004. ® Source: Wolfgang Mahr, Director IT, Spar Switzerland, 16.02.2004.
1. Introduction
from the service level at their distribution centres (DCs). Their argunnent is that if the DC can fulfil 99% of the store orders, then one can expect an on-shelf availability rate of 99%. This thought is not quite correct, as the work of Gruen, Corsten et al. (2002) demonstrates. Their meta-study proves that in the last few decades the OOS rate has not decreased. It seems to have remained rather stable at a level of about 8%. This high figure is rather surprising for manufacturers and retailers, as they expected a far better rate considering the progress made in technology and new logistic concepts Introduced in the last few years. In order to tackle this problem, the study further examines the reasons for OOSs, as depicted in Figure 3.
Figure 3: Summary of OOS root causes^
Surprisingly, almost three-quarters of stock-outs are the direct result of retail store practices and shelf restocking processes. A very similar result was found in a KLOG project carried out with the European grocery retailer MYFOOD^. Sixty percent of the OOS situations at this retailer were caused by the ordering behaviour in the stores. In 10% of the cases the goods were in the store but not on the shelves. Here lies a possible answer for the Ineffectiveness of existing ECR (Efficient Consumer Response) activities on the OOS rate. Many of these ECR activities concentrate on the smooth transportation of items up to the store gate. How replenishment orders are placed, how order quantities are determined and processes in the so-called last 50 metres in the store still remains an area for research. To see the financial implications of OOS incidents, Gruen, Corsten et al. (2002) conducted an estimate of the overall effect of OOS on sales that takes into account ^ Source: Gruen, Corsten et al. (2002). The name of this retailer and of three others that are examined in the case study section 6 have been made anonymous for confidentiality reasons.
1. Introduction
the response of consumers. The result is that on average, retailers might lose up to 4% of their turnover due to SKUs (stock keeping units) being absent from the shelves.
1.3. New Technologies Enable Automatic Store Replenishment Systems As half of all OOSs arise from incorrect ordering and forecasting processes, it is sensible to have a closer look at stores' replenishment processes and systems. Some decades ago, there was no alternative to manual store replenishment systems. A planner, for example the store manager, was responsible for deciding the two main parameters of replenishment systems, namely the amount to be ordered and the when to place the order. In order to do this, the planner had to check manually the quantity in stock. In the last decade, there has been an impressive diffusion of large-scale information packages such as ERP (Enterprise Resource Planning) in organizations (Kallinikos 2004). In addition, identification technology (such as barcodes and scanners) and communication tools (such as EDI®) have become very cheap, their implementation and use is nearly routine (Kuk 2004). Today, these and other new technologies make it increasingly possible to automate the replenishment decision-making. The interviews conducted with practitioners as well as other surveys (Bearing Point 2003; Sm^ros, Angerer et al. 2004a) clearly reveal a trend in retailing towards automating store replenishment processes: "The normal replenishment process has been until now consisted of store personnel deciding what quantity to order by looking on the shelf. Now, retailers want to let the systems take this decision. "^^ Semi-automatic systems merely support the planner in his decision, for example, by showing him electronically the inventory and order restrictions. Advanced automatic store replenishment systems are IT-based software systems that automatically decide when to order which quantity. Nevertheless, there are several differences in the complexity and performance of such ASR systems. The simplest systems just place an order as soon as an article is sold or when a certain minimum stock level is reached. No forecasts are made; the quantity to be ordered is calculated with a very simple algorithm (e.g. fill up to a certain level). This kind of automatic system is, for example, used by the Swiss retailers Mobile Zone, Marionnaud and Fust. One example of a complex, state-of-the art ASR system comes from the company SAP Electronic Data Interchange. ^ Source: Wolfgang Mahr, Director IT, Spar Switzerland, 16.02.2004.
1. Introduction
AG (Switzerland). The main advantage of their ASR software "Superstore" is that it makes a separate forecast for every item in every store. This is in contrast to other software programs, which make their calculations at SKUs/stores clusters due to IT performance restrictions (Beringe 2002). Furthermore, such forecasts do not rely only on historic sales. Their sophisticated causal models also consider price, promotion, seasons, holiday and other events when predicting demand. The introduction of this product in the German over-the-counter chemist retailer dm-drogeriemarkt
resulted
in a
70-80%
reduction
in OOS
incidents
and
simultaneously reduced the inventory stock level by 10-20% (Beringe 2002). A detailed classification of the various existing ASR systems is provided in section 4.1. Several technological developments were necessary to enable the implementation of such sophisticated replenishment systems: •
Electronic inventory systems
•
Identification technologies (barcodes, scanners)
•
Data warehousing capacities (for historical sales data)
•
Electronic data interchange (EDI)
•
IT computation power (for forecasts at SKU level)
•
Enterprise Resource Planning (ERP) systems
First, inventory management systems were introduced that made it possible to manage quantities of a product in the electronic systems. These electronic inventory systems profited markedly from identifying technologies such as the barcode. The order process was simplified by using fax and electronic connections (e.g. EDI) between companies. IT systems' storing capacity increased, making it possible to handle larger amounts of data. The storing of huge quantities of POS (point of sales) and inventory data became feasible with new data warehouses and storage mediums. Furthermore, not only was it internal data that was more easily accessible; thanks to larger communication bandwidths, it has became possible to access large quantities of external data as well. This new external data includes competitive information (e.g. the price of a competitor's products), market data (e.g. from marketing institutes) and collaborative data (e.g. collaborative forecasts with suppliers) (Beringe 2002). The increased IT-power performance has made it possible to calculate in fractions of a second increasingly complicated forecasts at SKU level. In a nutshell, IT-capabilities do not seem to be the decisive restriction anymore. This statement is underlined by a study by Sabath, Autry et al. (2001) which shows that the information system capabilities (such as timeliness of information or compatibility of the IT) of the surveyed manufacturers and retailers are on average on a rather
1. Introduction
high level.^^ The authors conclude that these companies already have the basic requirements to operate automatic replenishment programmes. Yet, IT is only one step towards ASR systems; other questions concerning the organization and logistical processes arise. The introduction of ASR systems goes hand in hand with the implementation of large-scale ERP systems, and thus has radical
implications
for
the
organizations
and
processes
of
firms.
For
Kallinikos (2004, p. 8), the Introduction of these packages marks "a distinctive stage In the history of computer-based information technology's influence in organizations." Their main achievement is the new possibility for integrating operations and information
across
functions,
departments
and
modules.
Therefore,
the
organizational and behavioural implications have to be considered. One example is the role of employees; the introduction of such automatic systems can result in a dramatic change of their working habits: "The changes resulting from the introduction of ASR systems are enormous. (...) It is a change of paradigm. Who has today the same job as 5 years ago? (...) Especially elderly employees have problems with the changes. Our planners have been doing their jobs for 25 years; one has to take this into account. "^^ The importance of ordering for the store employees can be seen on the following statement of a grocery employee, which altered Ren6 Descartes famous statement: 7 order, therefore I am."
1.4. Research Deficit What is the contribution of academic research in the field of ASR? Surprisingly, although several retailers have automated their order process in the last few years (Sm^ros, Angerer et al. 2004a), there is almost no academic source examining this topic at the level of the store. It is worth noting that other technologies In retail, such as RFID (Radio Frequency Identification) and the introduction of the barcode, have received far greater attention from the public and from scholars.^^ One explanation could be that ASR is a technology working in the background. If it works properly, consumers should notice it only indirectly, such as through higher availability in the stores. Yet this would not explain the interest received by other technologies, such as
^^ See Table 5. ^^ Source: Stefan GSchter, DC-director. COOP, 16.02.2004. ^^ This statement can be illustrated by a look at the agenda of European ECR initiatives. There exist several working groups dealing with RFID and barcodes, but none that deals with automatic replenishment.
1. Introduction EDI, which also work in the background. To sum it up, the questions surrounding ASR systems arising for practitioners and researchers can only be partially answered through a review of the literature. The few sources related to this topic are presented in the following.
General Inventory Management Literature The contributions that come from general Inventory management literature are rather basic. Existing academic sources of inventory modelling sources seek to answer the two primary questions that arise when dealing with replenishment systems, namely when should which quantity be ordered (Wagner 2002). Many papers in the operations research (OR) field concentrate on the modelling of replenishment systems, and try to identify an optimum under certain conditions (see Groote 1994; Silver, Pyke et al. 1998; Bassok 1999; Gudehus 2001). Algorithms calculate minimal inventory levels by choosing the right order quantity and order point so that certain a priori set objectives are fulfilled (e.g. a certain percentage of service level, a maximum out-of-stock rate, etc.). In general, much of the inventory management literature remains very theoretical. The implementation of these models in daily business is rather difficult (Wagner 2002). Many simplifications are made. When calculating the optimum, the specific situation of retailers at store level is not taken into account. The critical costs of retailers at store level are not inventory holding costs, but handling costs, which can be between 3-5 times as large as the former (Broekmeulen, van Donselaar et al. 2004b). Yet, store handling costs are seldom taken into account in these mathematical models (cf. van Donselaar, van Woensel et al. 2004).
ASR Related Literature One of the few sources dealing directly with ASR systems is a dissertation published by Norman Gotz (1999). His main achievement has been to develop software that enables the automation of order placement. This program uses existing forecasting heuristics and combines them into a new one. The benefits of Gotz's program are shown with a simulation based on real data from two stores of a German drug retailer. The theoretical benefit compared to the old system is an average cost reduction of 14.5%, mostly from a reduction in the inventory holding costs. GGtz's simulation shows a strong effect of the automation on the OOS rate: Out-of-stocks are reduced by almost 80%. The time savings for the stores are calculated at about 5 hours per ordering day. The remarkable contribution of this thesis is that for the first
10
1. Introduction
time the benefits of such systems were calculated, at least in theory.^^ Gotz stresses that one of the main advantages of the system is that retailers have the power to realize the described benefits on their own, independently from the rest of the market. Nevertheless, his work is only the very beginning of the research on this topic. The IT systems described in Gotz's work are no longer state-of-the art.^^ The overall focus of his work is rather mathematical; the main goal is the development of an optimal forecasting heuristic. Consequently, many aspects in the context of ARP systems such as optimal implementation and organizational influences are not considered.^^
Supply Chain Management As valuable as Gotz and other contributions from Inventory management research are, they show the limits of focusing strongly on mathematical or IT aspects when dealing with ASR systems. An approach to the topic from a more abstract level could be helpful, as the implementation of ASR systems is a fundamental change in the way the flow of materials Is triggered in a supply chain. Therefore, SCM research, which deals with the importance of having demand-based replenishment systems (pull systems), is examined in this thesis.""^ One effect of such a pull-supply chain is a major increase in efficiency and performance (Fiorito, May et al. 1995; Cottrill 1997; Closs, Roath et al. 1998). Because the competition between grocery retailers is so fierce. Bell, Davies et al. (1997) regard pull-supply chains as a necessity for every retailer. A practical implementation of the idea of a demand driven supply chain are automatic replenishment programmes (ARP). Common ARPs are Vendor Managed Inventory (VMI), Continuous Replenishment Planning (CRP), Quick Response (QR) and Collaborative Planning, Forecasting and Replenishment (CPFR).''® Overall, the sources on this topic (e.g. Daugherty, Myers et al. 1999; Ellinger, Taylor et al. 1999; Myers, Daugherty et al. 2000; Sabath, Autry et al. 2001) show the major benefits of such programmes, because OOS and handling costs are reduced while the inventory turn increases.
^* GOtz (1999) used real data for his analysis. Yet, he did not prove that such systems would also work under real life conditions. ^^ For example, in order to save computing power, the products are clustered into 4 groups that have a common forecast function. Today's systems have evolved rapidly in the last few years so that this constraint is no longer relevant for today's ERP systems. ^^ Only once does GOtz acknowledge that performance could depend on the satisfaction of employees with the new software and their commitment to it (GOtz 1999, p. 186). ^^ A pull system is driven by demand at the lowest point of the chain (Christopher 1998). In the context of this thesis this would be the shopper in the store. ^® See for an explanation of these concepts Christopher 1998, chapter 7; Seifert 2002; Alicke 2003, pp.168-169
1. Introduction
11
Although the research methodology and findings of these researchers concerning the performance and context influences of ARPs are remarkable, the transfer of their results to the ASR systems context is limited. The research deficit in this field is that the theoretical sources have concentrated up to now on pull systems from the manufacturer to the retailer's distribution centres. In the context of ASR systems, it is also necessary to consider the replenishing of stores. In ASR systems the demand is being driven by the shopper in the store.^^ Myers, Daugherty et al. (2000) recommend focusing future research on replenishment automation in a single industry. Sabath, Autry et al. (2001, p. 103) further state that "the issue of information systems capabilities is vital as well and deserves further study." Furthermore, the authors point out the importance of making additional investigations concerning organizational structure for developing decision guidelines. All these research recommendations are considered in the conception of this thesis.
Overview of Research Deficits To sum up, more research on this topic is required because the subject of ASR has not received in theory the attention it deserves considering its importance for practitioners. The first deficit is that there is no academic source describing the different systems in use; thus, a descriptive model and a classification has to be developed. Furthermore, an examination of retailers has revealed that the introduction of ASR was often part of changes taking places in the entire ERP system, therefore significant financial and managerial inputs are necessary (Keh and Park 1997). Although there are many success stories from practitioners describing the enormous advantages of introducing automatic store replenishment systems (see, for example, Beringe 2002; Anderson 2004; Hopp and Arminger 2005) there has been only limited examination of such statements from academic sources. Consequently, it is not surprising that some of the retailers interviewed are sceptical about the sense of such systems. Even if practitioners are convinced as to the utility of ASR, they remain insecure on the question of which system to choose for different types of products. There is no conceptual framework available at the moment which would help practitioners to choose an adequate replenishment system. In order to be able to develop such a model more needs to be known about the relationship between replenishment performance (e.g. OOS rate, inventory levels) and contextual
For a detailed explanation of the limitations of ARP see section 3.2.2.
12
1. Introduction
variables (such as store and product characteristics). Finally, it is not clear how retailers have to adapt its organization and processes to best support the chosen ASR system. The research deficits In the context of ASR systems are summarized in Table 1. Research Deficits Concerning ASR Systems Missing: • descriptive model and classification a qualitative study on benefits u knowledge about relationship between store and product characteristics and the performance of replenishment systems • method determining necessary replenishment system for each product category a knowledge on the change retailers' organization and processes Table 1: Overview of research deficits
1.5. Research Questions In the last sections it was demonstrated that numerous unknowns exist in the context of ASR systems. Therefore, this research aims to answer following main question: Q: Under which conditions can retailers benefit from automatic store replenishment systems? Practitioners (cf. Bearing Point 2003) often speak of automatic replenishment systems without taking into account that these systems can vary from very simple heuristic- based decision systems to highly sophisticated ones with self-optimization and complex forecasting models. This thesis aims to help practitioners choose the right system for their business. This implies a need for identification as well as categorization of the automatic replenishment systems available. For that, this thesis first develops a descriptive model from which a classification is derived. Consequently, the first sub-question, which provides support in answering the main research question is: Qi: How can automatic store replenishment systems be classified?
1. Introduction
13^
The kind and magnitude of benefits discussed in theory and practice is very broad; therefore, they are addressed in this thesis in particular. For three of the possible benefits (fewer OOS incidents, lower inventory levels and lower inventory variability) quantitative examinations are carried out. The second sub-question in this thesis is: Q2: What benefits can a company expect from the implementation of replenishment systems? Practitioners want to know if the replenishment systems they are using are best suited for them. To be able to choose the right systems for a given business and product category, it is necessary to perceive the interrelations between the elements of such systems and to understand the influence of environmental setting on performance. Therefore, knowledge about the influence of store and product characteristics on the ASR system outcome is necessary. Consequently, the next sub-question Is: Q3: How is the performance of ASR systems influenced by product and store characteristics? A new automatic replenishment system does not only influence the performance of replenishment it can also influence the entire distribution system, delivery frequencies and employee working behaviour. Consequently, the choice of a new replenishment system with all its implications for an organization is a strategic decision and will be one focus of this thesis. Practitioners need a methodology for choosing the right system. For this reason, the next sub-question is: Q4: Which ASR system is recommended given the characteristics of a certain product and retailer? It can be assumed that some companies will not have a system that adequately meets their needs. Consequently, the recommendation will be to implement another type of ASR system. For an automatic replenishment system to reach its full potential, next to technical requisites it is necessary to adapt the retailers' organization and Internal processes. Therefore, the last question that arises and which will be examined is: Q5: Which intra-organizational aspects of a company have to be changed to adapt a new ASR system?
14
1. Introduction
1.6. Thesis Structure This thesis is structured as follows: a Chapter 1 highlights the major contribution of logistics for retailers and describes current challenges in retail. This chapter ends with the derivation of the principal research questions from the research deficits. The methodology for addressing these research questions is depicted in • Chapter 2. This chapter sets out the research framework, the methodologies used to address the research questions and illustrates the activities undertaken by the author to accompany and guide the research process. • Chapter 3 highlights the contribution of theoretical sources for this thesis. Four research perspectives are examined, namely inventory management, logistics and operations management, business information systems and contingency theory. The theoretical findings from this chapter, together with evidence from interviews, are the foundation of • Chapter 4. In this chapter, hypotheses and models concerning automatic store replenishment systems are developed. First, a descriptive model and a classification structure for ASR systems are presented (Qi). Second, an explanatory model is developed that explores the correlation between the performance of retailers and the store, product and ASR system characteristics. The created hypotheses are tested in a Chapter 5, in a quantitative analysis with the help of real inventory and sales data from a grocery retailer. One part of the statistical analysis shows how the OOS rate and the stock levels of the automated goods changed compared to a control group (Q2). Another part of the chapter illustrates the correlation between store/product characteristics and ASR performance (Q3). A major finding described in this chapter is that ASR systems are Indeed beneficial for retailers, yet their contribution depends strongly on how they are implemented. Therefore, in Q Chapter 6 there is a description of how four European retailers have implemented sophisticated ASR systems. The case studies show the required organization and processes for successful ASR implementation (Q5). This chapter ends with several action recommendations for mangers as regards store operations and choice of system (Q4). • Finally, Chapter 7 summarizes contributions to theory and practice. Figure 4 illustrates the structure of this thesis.
15
1. Introduction
1. Introduction The challenge of store replenishment for retailers Research deficits and research questions
2. Research Framework and Design How can the research questions be answered? Result: Research framework and methodology defines main focus of 3. Literature Research Four theoretical perspectives: inventory management, logistics/operations management, business information systems and contingency theory Result: Theoretical foundation of thesis combined with interview results is tiie basis of 4. Development of Models Descriptive model and classification of ASR systems Explanatory model Result: Models and hypotheses to describe and explain different ASR systems developed hypotheses are tested in 5. Quantitative Analysis Sample and statistical methodology Testing of hypotheses Result: Proof of the benefits of ASR systems
QQQJ
how to introduce ASR systems in practice is shown in 6. Field Research and Managerial Implications Case studies from four European retailers Result: Required changes for successful ASR implementation Decision tree: adequate ASR system Recommended store operations
7. Conclusion Contribution to practice and theory Research limitations 8. References & Appendix Figure 4: Thesis structure
Ms
16
2. Research Framework and Design
2. Research Framework and Design The outcome of any research is strongly affected by the choice of the research methods and strategies. As Scandura and Williams state (2000, p. 1249) "[a]ny research method chosen will have inherent flaws, and the choice of that method will limit the conclusions." This means that design choices about instrumentation, data analysis and construct validation may affect the types of conclusions that are drawn (Sackett and Larson 1990). Therefore, this chapter gives an oven/iew of the research methodology chosen for this thesis. First, a research framework narrows the research field down. Second, the qualitative and quantitative research methods are depicted before, finally, the research activities that led to this thesis are presented.
2.1. Research Framework The research framework mainly focuses on the role of automatic replenishment systems for European grocery retailers. The main research field of this thesis is not the distribution system to the store itself, but the ordering logic that lies behind it (see also Figure 5). This means that the focus of the thesis is the logic of the system that decides at what time and in what quantity the goods are replenished in a store. The routes taken by the goods or their mode of transportation are only secondary in this context. Consequently, whenever in this thesis the term "supplier" is used, there is no differentiation between products that have come from a retailer's own distribution centre or directly from the manufacturer. The shop-floor logistics (i.e. how the goods are transported from the backroom to the shelves) are a relevant part of the research framework, especially when talking about changes in process due to ASR implementation. Focus of thesis Flow of information (order-, inventory level-, POS-data...)
I supplier J
Store ordering processes
Flow of material (replenishment of products)""
Figure 5: Focus of research
(major European grocery retailers)
" I Customer \
2. Research Framework and Design
17
The focus on Europe does not mean that such systems are not interesting for other retailers from other world regions as well. The OOS studies mentioned in the introduction show a similar level of operational problems all over the world (Gruen, Corsten et al. 2002; Roland Berger 2003b), and replenishment systems play an important role for North American retailers as well. A Bearing Point technology study (2003) shows that although 61% of the North American retailers studied use automatic replenishment systems, only 38% have a well-documented and established inventory management strategy and thus a clear view of how such systems are best implemented within the organization. As there are no fundamental differences in the logistics or store operations between these two regions, one can presume that the findings will be significant for American retailers as well.^° The same holds true for retailers from other regions so that the result of the European research will be most probably globally relevant. The grocery retailers group is among the most Interesting ones In retail, as the leading grocery retailers have in the last decades broadened their categories to products that are not related to the food sector at all. Guptill and Wilkens (2002) describe a grocery store as a retail store with at least 1,500 different food items and/or $2 million in annual sales that sells dry grocery, canned goods and non-food items plus some perishable items. A grocer is, according to the Oxford English Dictionary, a "trader who deals in spices, dried fruits, sugar and, in general, all articles of domestic consumption except those that are considered the distinctive wares of some other class of tradesmen."^^ This last definition does not seem to be precise enough nowadays. The leading retailers sell through their distribution channels almost all kinds of consumer goods that were some decades ago only available at certain speciality stores (e.g. apparel, computers, household goods, garden articles, etc.). The term modern grocery^^ is sometimes used to separate the core business of groceries (food and near food) from more exotic products (e.g. financial services). This large variety of articles force retailers to have separate logistics strategies for their items, depending on various factors such as speed of turnover, value, demand volatility, etc. Some retailers have adopted different strategies and channels to distribute their products. There exists a large variance of stores, ranging from rather small supermarkets (200 m^) to huge hypermarkets (10,000 m^ and larger). This breadth of product range and store size among grocery ^° This is a so-called analytical generalization as described by Punch (1998). ^^ Source: definition found on the site http://w\Aw.oed.com (accessed 01.09.2005). ^^ The term modern grocery is taken from the definition of Planet Retail: modern grocery distribution includes both grocery and non-food sales from modern grocery distribution formats. It excludes sales from independent specialist formats and wet markets. Source: http://www.planetretail.org (accessed 01.09.2005).
18
2. Research Framework and Design
retailers will facilitate the generalisation towards other retailers with similar product categories or distribution channels. Another advantage of this group is the high variance that exists within it: some grocery retailers still rely on completely manual systems, while others have implemented very sophisticated systems with complex forecasting algorithms (Sm^ros, Angerer et al. 2004a). Grocery retailers have to fulfil consumers' wishes immediately, as supermarket visitors are not normally willing to wait for the delivery of their goods. This business has further a strong demand volatility as many products are easily substituted, making SKU-level forecasts difficult. Therefore, all the grocery retailers have adopted a make-to-stock strategy.^^ Another benefit of studying grocery retailers Is that they are rather well organized in initiatives such as ECR-Europe. There are several project groups analysing their supply chains and the benefit of collaboration, standardization, electronic data interchange, identification and more so that additional information sources are more easily available for researchers. And last but not least, the choice of grocery retailers follows the research tradition of the Institutes ITEM and KLOG at the University of St.Gallen, where this subgroup of retailers has been examined for several years.
2.2. Research Methodology The long tradition of the University of St.Gallen to focus on topics that are relevant to business practice is fully continued by this work. One of the main advocates of this statement is without doubt Hans Ulrich. For him, business science is understood as a leading and managing-science and has thus the central objective of giving practitioners the ability to act and to make decisions in a scientific way (Ulrich 1981). The starting point for each research project in business science is an analysis of existing practical problems. First of all, interesting situations, correlations and contexts are observed from a practical point of view and then are conceptualised (Ulrich 1981). The concepts that are developed are tested in practice again and again and become gradually more refined. This iterative learning process will finally generate at the end of the research process theoretical and practical solutions to the identified problems that can again be tested in practice (Kromrey 2002). Ulrich's design of a 7-step research process was the base for the structural approach of this thesis (see Figure 6). The main characteristics of this procedure are the iterative approach and the deep contact of the researcher with the practice.
For an explanation of make-to-stock see Alicke (2003, p. 50).
2. Research Framework and Design
19
step 1: Identifying and staicturing problems and their potential solutions that are relevant to business reality
Empirical social sciences
Formal sciences
r^
Step 2: Identifying and interpreting theory and hypotheses of the empirical social sciences relevant for the targeted problem
Step 3: Identifying and specifying formal scientific procedures that are relevant for the targeted problem
LI
T Step 4: Identifying and assessing the relevant application context
i r-M Step 5: Deriving rules and models
T
Practice
•—I Step 6: Testing rules and models in the application context
i Step 7: Documenting research results and consulting of practitioners
Figure 6: Integrative research procedure
In the first part of the research process (steps 1 to 4), a mix of qualitative research methodologies is used. These research steps are documented in this thesis in chapter 2 and 3. Campbell and Fiske (1959) have stressed the importance of using several methodologies to overcome the main deficiency of qualitative research: the limited generalisation. Denzin (1978; 1989) in turn develops the term "multiple triangulation" that applies when researchers combine multiple observers, theoretical perspectives, sources of data and methodologies in one investigation. He further states that "all the advantages that derive from triangulating single forms are combined into a research perspective that surpasses any single-method approach" (Denzin 1978, p. 304). The term triangulation is a metaphor from navigation which "use[s] multiple reference points to locate an object's exact position" (Smith 1975, p. 273). The triangulation of data collection settings affects the external validity of the results (McGrath 1982). Several researchers from social and business sciences, such as Webb, Campbell et al. (1966), Smith (1975) and Jick (1979), recommend triangulation. The research process used in this thesis adopts this methodology by simultaneously combining different research methods and data collection forms.
* Source: adapted and translated from Ulrich (2001, p. 222).
20
2. Research Framework and Design
Mc Grath (1982) defines eight research strategies used in management research. The one strategy corresponding closest to the methodology used in this thesis is the field study. Field studies investigate behaviour in its natural setting. The data is collected by the researchers themselves on site. Scandura and Williams (2000, p. 1251) give the advantages and risks of this method: "This strategy maximizes realism of context, but it can be low on precision of measurement and control of behavioural variables (there is lack of experimental control). It can also be low on generalizability to the population, with the study population not representative of the target population." Despite these drawbacks, this research strategy is extremely popular. In a review of 385 papers of the journals Academy of Management Journal, Administrative Science Quarieriy and Journal of Management from the years 1995-97, Scandura and Williams (2000) found out that 67.5% of all papers had chosen field studies as the research strategy. Ten years previously the percentage was only 54.1%, therefore a significant increase had taken place. An example of this strategy in this thesis is the KLOG OOS-project conducted for the grocery retailer MYFOOD. This was not an experiment, as no variable was manipulated, and the researchers acted merely as observers. The data collection and its analysis comprise qualitative and quantitative research strategies. Friedii, Billinger et al. (2005) state that the three most frequently used methods of qualitative research are action research, grounded theory and case study. The main methodology used in this thesis is the last. The basic idea behind case studies is to investigate only a small number of cases, sometimes even only one, yet these in a great detail. As Punch (1998, p. 150) states, "the general objective is to develop as full an understanding of the case as possible." According to Yin (1988, p. 23), a case is an empirical inquiry that: " • Investigates a contemporary phenomenon within its real-life context; when • the boundaries between phenomenon and context are not cleariy evident; and in which • multiple sources of evidence are used." As demonstrated in section 1.4, there are almost no existing academic sources that investigate the topic of this thesis. The focus on the case study method seems therefore to be most appropriate for this research project, as this methodology is recommended for researching topics in new areas (Eisenhardt 1989). The external validity of qualitative studies is sometimes considered to be limited. It is true that
2. Research Framework and Design
21
case study research is not able to give statistical generalisation, hence the external validity comes from analytical generalisation (Gassmann 1999). In case study research, however, generalisation is not always the goal. Some cases might be so important or interesting that they deserve a study in their own right (Punch 1998). To achieve generalisation, it is necessary to conduct the research at a sufficient level of abstraction: "The more abstract the concept, the more generalisable it is. Developing abstract concepts and propositions raises the analysis above simple descriptions, and in this way a case study can contribute potentially generalisable Findings" (Punch 1998, p. 155). The main advantage of case studies is that they are most appropriated for describing and understanding complex social systems {Marshall and Rossmann 1995). Another argument for the case study approach is that the phenomenon observed in this thesis happens in the presence of but cannot be influenced by the researcher (Yin 1988). The main reason for choosing the case study method is that case studies have a holistic focus and aim to preserve and understand the wholeness and unity of a case (Punch 1998). More than 50 interviews with practitioners were conducted, with the aim of reaching an in-depth understanding of retailers' logistics (see the interviews listed in section 1.1). These interviews are conducted orally, as the actuality and explorative character of this research phase would not support a written interview technique (Lamnek 1993). The intimate connection with empirical reality that is demanded by Glaser and Strauss (1967) in order to develop a sound, relevant and testable theory is guaranteed with this approach. There are, nevertheless, also elements from another two popular qualitative methods. The main idea from grounded theory is that qualitative research has the goal of generating new theories (Punch 1998; Friedii, BIHinger et al. 2005). And from action research the idea is adopted that the starting point for a research project is not a theory but a problem in practice (Coughlan and Coghlan 2002). The aim of the action researcher is to attempt to change the examined system and obtain a desired new status (Ulrich 1981; Coughlan and Coghlan 2002). The author was not personally involved in this change; nevertheless the last chapter of the thesis gives recommendations to managers as regards technical and organizational changes that a retailer may undergo to reach a new level of automation.
22
2. Research Framework and Design
In order to derive and test rules and models (steps 5 and 6), four case studies of retailers using advanced ASR systems are described (chapters 4 and 5 in this thesis). This is the part of the research project with the greatest interaction with practice; several iterative loops between theory generation and theory verification are carried out. On the one hand, the information gained from the interaction with the chosen companies is used to verify and refine the descriptive model developed in sections 4.1 and 4.2. On the other hand, the information gathered is the main source for developing the explanatory model. Closure is achieved when the differences between collected data and developed theory is small (Bansal and Roth 2000). A good illustration of this iterative process can be seen in Figure 7.
Figure 7: Case study research as iterative process between ttieory and empiricism^^
For the creation and verification of the models, the author does not rely only on qualitative data. As stated by Yin (1988) and Eisenhardt (1989), the data collection method and the case study process can also include more qualitative elements. While experiments have great internal validity because of their precise control of variables, an additional external validity can be achieved by accompanying surveys (Scandura and Williams 2000). Therefore, next to several in-depth interviews (qualitative data), archival documents review, participant observation and analysis of the data gained from the ERP system of one retailer is Included. This process allows the researcher to gain an even broader understanding of the company, or as Jick (1979, p. 603) states: "a more complete, holistic and contextual portrayal of the unit(s) under study." In addition, the quantitative analysis is a critical part of the testing of the models and hypotheses created in section 4.3.
Source: translated from Gassmann (1999).
23
2. Research Framework and Design
Some researchers criticise the simultaneous use of quantitative and qualitative methods, as every method has different aims and purposes (Dey 1993). The traditional distinction between two schools of social sciences, one oriented towards the qualitative development of theories, the other directed at the quantitative testing of theories, Is criticised by Baumard and Ibert (2001). The authors point out the importance of not being dogmatic, as both forms of data are useful for constructing and testing theories. Or, as Glaser and Strauss (1967, p. 17) formulate: "There is no fundamental clash between the purpose and capacities of qualitative and quantitative methods or data. (...) Each form of data is useful for both verification and generation of theory." The final step in this research project (step 7) is the developing of recommendations for managers (chapter 6 of this thesis) and the documentation of the findings in form of the presenting thesis.
2.3. Research Process The research process was guided by several activities which can be seen in Figure 8. Summer 2005 (7) Synthesis of the results: writing of thesis
(6) Field research: qualitative and quantitative analysis
(2) Activities with ECR-Europe
October 2002
(3) Supervision] (4) European of Bachelor grocery and Master research theses study
(5) Consulting projects
(1) Literature review
Figure 8: Research activities in this research project
The investigation began, as expected, with a thorough review of the available literature and other information sources (Activity 1). Due to the lack of academic papers dealing with ASR systems, the research started with more general inventory
24
2. Research Framework and Design
theory and logistics literature. Later, the research concentrated on papers targeting typical retail problems, such as data accuracy, inventory visibility and other logistics challenges. A new perspective came from the study of papers dealing with business IT and contingency theory. The involvement in the ECR Europe community was critical to obtain first hand information from practitioners (Activity 2). In several dialogues and meetings, the author deepened his understanding of business practices and current challenges in the retail and consumer goods industry. The Involvement of the author in the ECR community was: • Personal involvement in the ECR-Switzerland working group "On-shelf availability" • Participation in the quarterly ECR-Switzerland working meetings and the official ECR Europe Conferences in Berlin (2003), Brussels (2004) and Paris (2005). • Participation in various activities of the ECR Europe Academic Partnership, including the publishing of the ECR Journal and the organization of the annual ECR Student Award. The supervision of bachelor and master theses was an additional important source of first hand information from practitioners in the industry (Activity 3). All supervised works focused on actual problems of enterprises in the consumer goods sector and gave a broad overview of the performance of logistics within retailers: • CPFR in the sports industry. Simon Steiner (2003) • CPFR in the fashion and apparel sector. Marion Bragger (2003) • CPFR in the German grocery retail market. Juerg Neuenschwander (2004) • CPFR in the consumer electronics industry. Thomas Kohl (2004) • Chances and limits of CPFR in the Swiss grocery retail. Diego Rutsch (2004) a Limits and possibilities of CPFR in the Swiss grocery retail. Dominic Loher (2004) • Supply Chain Management in the fashion industry: critical factors. Benjamin Brechbuhler (2004) • The operational control of stores with inventory management systems. Remo Maggi (2004) a Efficient Consumer Response. Impacts on consumer and welfare. Gabriella Todt (2005)
2. Research Framework and Design
25
The supervision of this work resulted in over 45 documented interviews with directors and managers in charge of logistics and supply chain management, category management, or information technology. The chosen companies were mainly retailers, followed by logistics and technology services providers. This more explorative approach led to a deep understanding of the current logistics processes within retailers. The result of this Inductive qualitative approach was the creation of the descriptive and parts of the explanatory model (see chapter 4.3). The next information source for the creation of the models and hypotheses was the pan-European study "Logistics processes of European grocery retailers" (Activity 4). It was launched in autumn 2003 and lasted for about a year. The aim of the study was to investigate the current logistics processes and performance of leading European grocery retailers. The study was conducted as a collaborative project involving researchers representing five different European universities. The consulting projects conducted together with industrial partners were another important source of information (Activity 5). The most valuable project for the research into on-shelf availability was conducted with the grocery retailer MYFOOD. The data gained through this project is the main data source for the quantitative analysis (see chapter 5). In addition, close contact with employees from MYFOOD gave helpful Insights into retailers' replenishment operations. The last activity before finishing the thesis (Activity 7) was the creation of field research of European retailers that have ASR systems in use (Activity 6). First, in short interviews (about 15 minutes), some 30 German, Swiss and Austrian retailers from very different consumer goods sectors were asked about their replenishment practices. The information gained in this survey was used to choose the four companies for the case study and also to design a semi-structured interview for the analysis. The selected companies and the outcome can be seen in chapter 6.
26
3. Literature Research
3. Literature Research There exists almost no academic source that deals directly with automatic store replenishment systems at retail store level. Yet, even if theory cannot provide direct answers to the research questions, It is fruitful to consider the presented practical problem from different theoretical perspectives in order to obtain new impulses for this research project. Therefore, four theoretical research sources are addressed below in detail to verify, what contribution further theoretical streams can offer. The literature research focuses on sources dealing with inventory management, logistics and operations management, business IT and contingency theory (see Table 2).
Perspective
Contribution to the Thesis/ Research Question
Inventory Management
• Definition of decisions an ASR system has to tal^e (descriptive model) a Existing replenisliment logics and strategies for replenishment systems (classification) • Performance of different replenishment logics a c o s research: reasons and inten-elations D Forecasting methods and typologies
Logistics and Operations IVIanagement
• Benefits of well specified and organized logistics systems and programmes • Best practice in store operations • Coordination of structures, processes and systems to increase the efficiency of the SO
Q2
Business IT Theories
• Role of Enterprise Resource Planning (ERP) systems • Organizational and human agency changes due to new technology
a.
Contingency Theory
• Importance of the choice of the right ASR system for a given environment • Organizational perspective: influence of the contextual factors within an organization on performance of ASR systems
Q3
Qs
Qs 04 Qs
Table 2: Overview of basic theoretical sources reviewed (excerpt)
3.1. Inventory Management Perspective The first perspective that Is used to examine the research problems focuses on the challenge of right inventory management. Existing academic sources in inventory modelling seek to answer the two primary questions that arise when dealing with a replenishment system, namely when should which quantity be ordered (Wagner 2002). Many sources in the operations research (OR) field concentrate on the modelling of replenishment systems, and try to identify an optimum under certain restrictions (see Groote 1994; Silver, Pyke et al. 1998; Bassok 1999; Gudehus 2001). They seek to obtain a minimal inventory level by choosing the right order quantity and order point so that certain a priori set objectives are fulfilled (e.g. a
3. Literature Research
27
certain percentage of service level, a nnaximum rate of out-of-stock, etc.). This approach has been widely used in practice for many years (cf. Chang 1967). More sophisticated models seek to optimize a specific utility function. The challenge here is to identify and quantify all the relevant logistics costs that should be Included in this function. Inventory costs Include factors such as the cost of carrying stock, order costs, safety stock costs, transport costs and an estimate of the out-of-stocks costs. For example, Galliher, Morse et al. (1959) and Dalrymple (1964) explicitly consider OOS-costs in their inventory control systems.^^ The financial importance of optimal inventory control for companies is stressed by many authors such as Vollmann, Berry et al. (1992) and Dubelaar, Chow et al. (2001). Some sources even investigate the importance of the inventory for entire economies. In the United States, for example, the inventory value as a percentage of the GDP was in 1993 as high as 17.7% (Silver, Pyke et al. 1998). In the following sections, first theoretical inventory management sources dealing with mathematical models are highlighted. Papers and studies that in particular examine OOS situations in retail form the second part of the inventory management perspective.
3.1.1. Optimization in Inventory IVIanagement Research Inventory control systems have the aim of balancing demand and supply, reducing overall inventory costs and assuring an adequate service level (Wegener 2002). There are countless papers dealing with the modelling of sophisticated mathematical solutions for such inventory management systems (see e.g. Chang 1967; Inderfurth and Minner 1998; Ketzenberg, Metters et al. 2000).^^ Wagner (2002) criticises the rather theoretical approach of these papers; their implementation in real life applications is problematic. Several other authors concentrate on retailers and their specific inventory management challenges. An identified central strategic issue that Influences the success of a retailer is the setting of the right service level (Balachander and Farquhar 1994; Gudehus 2001; Zinn, Mentzer et al. 2002). Other papers deal directly with the optimal ordering policy for retailers from a marketing point of view, i.e. they look for the optimal assortment and shelf combination to maximize sales. In the 1960's and 70's, several experiments were conducted to measure the effect of shelf A detailed discussion of this cost function can be found in section 4.1.2. ^^ For an overview of the history of OR research on inventory control, see Wagner (2002).
28
3. Literature Research
facings and inventory quantity on sales (e.g. Kotzan and Evanson 1969; Krueckenberg 1969; Cox 1970; Curiian 1972). In the following decades, other models sought to calculate an optimum in order to minimize inventory, shelf space costs and backorder (c.f. Corstjens and Doyle 1981; Cachon 2001). In these models, the handling costs are normally Ignored. An exception is the paper by Broekmeulen, van Donselaar et al. (2004a). These last researchers state that the inventory carrying costs are low compared to the handling costs, therefore they suggest a replenishment logic that takes shelf space and package restrictions more strongly into account.
3.1.2. Theoretical Sources on OOS A group of inventory management papers reviewed deals directly with OOSs. These papers can be split into two sub-groups. The first group empirically studies the extent and causes of OOSs. The first OOS study, conducted nearly 40 years ago, reports an average OOS rate of 12.2% (Progressive Grocer 1968). More recent studies report an OOS rate between 7% and 10% (Andersen Consulting 1996; Gruen, Corsten et al. 2002; Roland Berger 2003b; Stolzle and Placzek 2004). The second sub-group takes a marketing and behavioural perspective and studies the reaction of consumers towards out-of-stock that crucially influence retailers' sales (e.g. Emmelhainz, Stock et al. 1991; Dr^ze, Hoch et al. 1994; Zinn and Liu 2001; Sloot, Verhoef et al. 2002). Retailers try to increase their sales with two groups of market-driven tactics. On the one hand, there are "out-of-stores" tactics, which try to bring more consumers into the stores. Avoiding OOSs is, by contrast, an "in-store tactic." The latter tactics generally attempt to extract more surplus from shoppers once they are in the store. An attractive, full shelf attracts the attention of the consumer, making a purchase more probable. As shelf space is expensive^®, retailers have to decide whether to place another facing of a certain product (to increase its visibility and/or reduce the OOS probability) or to place an additional SKU (Dreze, Hoch et al. 1994). Several studies prove the value of such store specific micro-merchandising; consumer decision-making can be strongly influenced. Only one third of purchases are specifically planned in advance of a shopping trip (Dagnoll 1987). Many buying decisions are made on a low level of involvement and very quickly (Hoyer 1984). In addition, the average shopper shops in 3-4 supermarkets each week (Coca-Cola Retailing Research Council 1994). With these facts in mind It is easy to see the magnitude of the impact OOS still has today for grocery retailers'
In the USA store occupancy costs range between $20 per square foot for dry grocery and $70 per square foot for frozen goods (Drdze, Hoch et al. 1994).
3. Literature Research
29
sales. The knowledge of consumer behaviour is necessary to calculate the losses connected with OOSs.
3.1.3. Contributions and Deficits of an Inventory {Management Perspective The Inventory management perspective has a significant contribution to this thesis: • Basis of descriptive model • Foundation of relationships in explanatory model • OOS magnitude and impact The descriptive and explanatory models developed in chapter 4 are based on contributions from inventory management sources.^^ In particular, the sources that deal with challenges of replenishment in retail are of major value for this thesis. The papers and studies dealing with OOS highlight the magnitude of this problem in practice, illustrate the underlying reasons for low product availability and demonstrate possible solutions. Nevertheless, there are also some limits to the potential contribution of the theoretical inventory approach to this research field: • Simplified view of reality • Context parameters are regarded as given • Missing organizational and contextual aspects First, all mathematical inventory models assume a very simplified view of reality so that it is often very difficult to apply such systems a given real-life situation. Second, it is often assumed that the parameters in such systems are given. What is overlooked is that many parameters can be changed by the organization (e.g. delivery frequency, case pack size) so that the strategic dimension of such decisions are not examined. And third, these OR papers normally neglect the organizational and contextual aspects of replenishment systems, which are nonetheless critical for the performance of such systems. An exception to the last statement is the paper by Zomerdijk and Vries (2002), as the authors place their research focus on environmental influences of inventory control systems. The basic message of the authors is that, beside the traditional points of attention such as order quantities and replenishment strategies, it is critical to take care of contextual and organizational factors. The authors identify four significant aspects in the organizational context of inventory management: task allocation, decision-making and communication
For an overview of authors see Table 6.
30
3. Literature Research
processes as well as the behaviour of personnel. The notion of examining the contextual focus of replenishment systems is a fundamental concept that is incorporated into the research questions (see section 1.5). Overall, a
purely mathematical approach to inventory management
is not
appropriate, as Wagner (2002) states. He sees a major problem in the fact that classical inventory research is blind to all the "dirty data" issues that challenge companies (I.e. the data in the system is not accurate enough to be used for control and calculations). As Wagner further states, inventory modelling research is far removed from the entrenched software that now drives supply chain systems. Therefore, it is necessary to implement a comprehensive discussion of ERP and inventory holding systems, as will be accomplished in section 3.3.
3.2. Logistics and Operations iVIanagement Perspective Academic sources dealing with logistics and operations management form the second perspective addressed in this thesis. Logistics research can be defined as the systematic and objective search for and analysis of information relevant to the identification and solution of any problem in the field of logistics (Chow and Henriksson 1993). The basic assumption behind logistics research is that a particular course of action will be correlated with logistics performance (Chow, Heaver et al. 1994). The problem starts with the definition of performance, sometimes hard measures are meant (such as delivery time or net income), and sometimes more soft measures are in the focus (such as consumer happiness ratings or flexibility). Both perspectives have their strengths and weaknesses. A comprehensive overview of the literature on this topic is made by Chow, Heaver et al. (1994). Their main criticizing point is that none of these studies examines logistics performance in the context of supply chains. Yet this statement has to be revised today in the light of the comprehensive literature on this topic (cf. Stolzle, Heusler et al. 2001; Karrer, Placzek et al. 2004; Stolzle 2004). The logistics research cannot be completely separated from another research stream that is of great relevance for this thesis: operations management. Operations management is an area of business that is concerned with the production of goods and services, and involves the responsibility of ensuring that business operations are efficient and effective.^°
^ Definition by Wil Dk(i+L+p); PLACE Order Qk(i); 0) An example of a multidimensional forecast formula used in an replenishment system is provided by Achabal, Mclntyre et al. (2000). Their model is as follows: Sales = Baseline_sales * Seasonal_effects * Merchandising_effects This multidimensional model takes into account merchandising effects, effects from the elasticity of markdown prices, promotional frequency and length, advertisement size, among other events. However, the authors stress the importance of having as few parameters as possible ("parsimonious forecasts") as too many parameters can be counterproductive. In their experiment, with an increasing number of parameters the adjusted R-square increased in the estimation phase. This means that the models were quite good in explaining historical data. But these over-sophisticated forecasts were less accurate in prediction of future periods.
4.2. Classification of Automatic Replenishment Systems A classification of the different ASR systems was developed based on the examination of dozens of retailers. Based on the degree of automation, the following classification system was developed, as seen in Figure 13.
4. Development of Models
Figure 13: Classification of automatic repienisfiment
65
systems
In the following, the differences between the single levels of the presented classification will be discussed.^^
ASRO: Manual Replenishment Systems This is the level where all replenishment systems were before retailers introduced IT systems. A manual replenishment system relies solely on the information and intelligence of the employees. This kind of system is still in use nowadays; in grocery retailers (e.g. VOLG), bakeries (Kamps), discounters (Denner) and opticians (McOptic). But only three out of 12 major European grocery retailers still rely solely on such manual ordering systems (Smaros, Angerer et al. 2004a).
ASR1: Electronic Inventory-Based Replenishment Systems As soon as retailers have their Inventory levels administered by an IT system, one can say that a state of semi-automation has been attained.^^ In the descriptive model, this step means that the first module "Inventory Visibility" is represented electronically in the IT system. With this electronic information, it is for the first time possible to move replenishment decisions away from the store. A centralized system is now thinkable, where a central organizational unit monitors inventories at all the outlets and makes decisions concerning deliveries. In order to do so, a certain degree of inventory records accuracy has to be reached. It is worth noting that decisions about the time and quantity of the order are still made manually; therefore, the module "Replenishment Logic" is still not IT-based. Order restrictions can be displayed on an IT systems, but the planner still has to take them
See Table 10 for an overview of ASR differences. ' The term semi-automation is used because the inventory records are now automatically administered by the IT system, yet the ordering itself still takes place manually.
66
4. Development of Models
into account "manually" when placing the order. Fresh products, such as fruits and vegetables, are often ordered this way. Examples of companies using predominantly ASR1 systems are consumer electronics retailers (e.g. Media Markt CH) and book stores (Orell Fussll).
ASR2: Simple Replenishment Heuristics Basically, this is the first level where true automation starts. To reach this level, only a small step is necessary from ASR1 systems. At that level, the system already stores electronically inventory quantities. Consequently, just by implementing a simple heuristic it Is possible to make the replenishment system work without any employee intervention. One of the most common heuristics used in retail is the order-point order-quantity (minimum-maximum) heuristic. The inventory management system is updated after each transaction or at least before a new order is to be placed. The ordered amount is either a fixed quantity (Q) or the difference between the actual inventory level and a desired inventory level (S). These replenishment parameters are set beforehand and are valid over a long period (months to years). This kind of replenishment system is called consumption-based, as the consumption of the items at the POS releases the order (pull-system). With such a replenishment logic no forecasts are necessary. This ASR level is the first at which basic order restrictions can be automatically accounted for (e.g. the system follows the quantity restrictions when placing an order). An example of companies with ASR2 systems are grocery retailers, such as Coop and Migros, along with specialised retailers dealing with cosmetics (Body Shop), mobile phones (Orange Zone), household electronics (Fust), among others. Later in this thesis an ASR2 system with a special difference is examined. The replenishment parameters are set at category level and not at SKU product level. The label ASR2* is introduced to make the difference clear.^^
ASR3: Time Series-Based Forecasts At this level, for the first time the forecasting module is regularly in use. The system makes an estimate of the next period's demand and orders according to it. This system basically checks whether it is necessary to place an order in this period, keeping the actual stock level in mind, or whether it suffices to place the order in the next ordering cycle. As the demand is only a prediction, decisions are made taking
The difference between ASR2 and ASR2* is described in detail in section 6.2.2 in the case of MYFOOD.
4. Development of Models
67
into account a defined risk of OOS. The order quantity is either a cost-optimal amount Q* or determined by a preset order-up-to-level point S. The main difference to the former levels is that the placing of the order becomes demand driven. The forecast is created by observing the behaviour of demand data in the past and making assumptions about future consumption. On this level, forecasts are computed with time series methods (one-dimensional techniques). More sophisticated programs take into account trends, cyclical variations, seasonal patterns and irregular random fluctuations (cf. Silver, Pyke et al. 1998). However, even such complex algorithms remain one-dimensional, as the only independent variable is time. As the IT systems used in ASR3 systems are more advanced, it is theoretically possible for them to take into account more order restrictions when ordering. Retailers such as Spar Switzerland, Orange SA and Spengler use these systems for some of their products.
ASR4: Causal Model-Based Forecasts The difference between ASR4 systems and the former level is the complexity of the forecasts used. Causal-based forecasts are obtained by taking into account not only past patterns, but also by anticipating the effects that future events will have. These systems claim to be able to predict the effect of promotions, price reductions, actions of the competition, the weather and more, on the demand. For this, the systems have to understand and compute the underlying effects influencing consumer demand. Demand is consequently treated as the dependent variable, and explained within a causal model with several, independent variables. This kind of system represents at the moment the most sophisticated software tools available on the market. These causal models have been successfully implemented, for example, at retailers such as dm-drogeriemarkt, Woolworth and Metro Group.
ASR5: Multi-Echelon Systems This level somehow breaks with the linear order of Figure 13. It is possible to enhance any of the ASRO-4 systems towards a multi-echelon replenishment system. This means that a system simultaneously controls the inventory stock and makes replenishment decisions on several levels of the supply chain. An example would be an ASR system that automatically and simultaneously calculates the optimal orders for retailers' stores and DCs. Such systems would need even more intelligence than normal systems, as the complexity of a multi-echelon system increases. The optimization becomes more difficult due to more restrictions. As stated on section
68
4. Development of Models
2.1, multi-echelon systems are not included in the focus of this thesis, therefore they will not be further examined here. The following table summarizes the differences of ASR systems.
Table 10: Characteristics of automatic replenisiiment levels
4.3. Explanatory Model 4.3.1. Purpose and Structure of the Explanatory Model The explanatory model in this paper is intended to be a representation of predicted relationships between factors that influence ASR performance. The explanatory model helps to quantify ASR1-ASR4 system benefits (research question Q2) and to find the adequate level of automation for each retailer (Q4). The definition of the unit of analysis is critical for the further understanding of the model. In theory, retailers will face an optimal automation level for each SKU in every single store. Yet for practical reasons, a retailer can only have a certain number of ASR systems running in parallel. Therefore, retailers will have to define product clusters of items with similar characteristics that will be managed through a single system that fits the characteristics of that cluster best. For instance, a cluster could comprise slow-moving canned products with good demand predictability and low value. The explanatory model is also the basis for giving recommendations as to how to change the organization and replenishment processes (Q5).
4. Development of Models
69
A question that has occupied many practitioners is the question of reasons underlying OOS Incidents. In the literature, there are many sources dealing with this question (eg.Gruen, Corsten et al. 2002; Roland Berger 2003b). Although these sources examine the underlying reasons, they remain descriptive. Furthermore, a question that has not been discussed in full detail in theory is the quantitative correlation between OOS and other variables such as product, store and replenishment system characteristics. In the model constructed in this thesis, it is postulated that it is possible to explain, at least partly, the OOS and inventory level performance of retailers by having the product, store and replenishment characteristics as independent variables. Therefore, the additive model that is explored in this thesis is:
Y= M + ASR + P + S + (ASR x P) + (ASR x S) + e where OOS rate or inventory performance measure Y: M: Overall mean P: Effect from the product characteristics S: Effect from the store and personnel characteristics ASR: Effect from the replenishment system in use ASR X P: Interaction effect replenishment system and product ASR X S: Interaction effect between replenishment system and store e: Error term, unidentified effects The model consists of three main effects and two interaction terms. The first effect implemented is the replenishment system characteristics (ASR). It is the main hypothesis of this thesis that the inventory performance and the OOS rate^^ can be influenced decisively by ASR system choice. The second main effect In the model consists of product characteristics. There are several statements found in literature and practice that product characteristics will influence the OOS rate and Inventory performance (e.g. Andersen Consulting 1996; Grocery Manufacturers of America and Roland Berger 2002). For example, from the
In this thesis, the OOS definition formulated by Gruen, Corsten et al. (2002) is employed: the COS rate is measured as a percentage of SKUs that are out-of-stock on the retail store shelf at a particular moment in time (i.e., the consumer expects to find the item but it is not available).
70
4. Development of Models
inventory management literature it is known that products with higher demand uncertainty need higher stock levels to achieve the same availability percentage (Alicke 2003).^® The third main effect implemented in the model consists of store characteristics. It is self-evident that this variable can play an important role in inventory performance. Several studies have stressed that OOS rates vary widely between stores (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b). In the case of MYFOOD-the retailer that will be studied in detail in the quantitative analysis-all stores have exactly the same distribution system; therefore logistics should not play a role. But the stores have different sizes, assortments, regions and managers. These differences most probably influence store performance. Furthermore, another difference in the results could stem from the personnel, as they still have a significant influence on the ordering and the quality of store operations (Grocery Manufacturers of America and Roland Berger 2002). Therefore, it is sensible to Include this variable in the model, as a significant influence on the dependent variable can be expected. In addition to these three main effects, two interaction effects are included in the model. If the characteristics of a product have an influence on replenishment system performance, the ASR x P variable should be statistically significant. This would be the case, for instance, if store employees are influenced by the product's price when ordering while the ASR systems are not. One of the interviewees stressed that some store managers tend to have higher availability rates for products which they personally like. A machine doing the order is emotionless and is therefore less influenced by product characteristics. A similar result can be expected from the ASR X S interaction. It can be assumed that for retailers, such as MYFOOD, where store employees have greater influence on replenishment parameters, ASR performance will differ from store to store, making this interaction variable significant. In the following, hypotheses concerning possible interrelations between the dependent variables OOS and inventory performance and other independent variables are constructed. These hypotheses are constructed based on results from KLOG projects, interviews with practitioners and literature research. The hypotheses are tested in chapter 5 with the help of data from the grocery retailer MYFOOD. As Sometimes the correlation is not as intuitive. For instance, Kotler and Bliemel (1999) emphasise that only 5% of unsatisfied customers complain to store employees. Products where consumers complain more often will probably tend to have a higher availability, as the employees are aware of the importance of the problem. As customers will tend to complain for certain products more than for others, there is clearly a strong relation between the product characteristics and the OOS rate.
4. Development of Models
71
this company has mostly ASRO and ASR2 systems in use, the hypotheses are based on a comparison of these two systems. The quantitative analysis of the hypotheses will be carried out with the help of different types of analysis of variance.®^
4.3.2. Hypothesis Development: Product Characteristics Sales Variance For a retailer it is more challenging to have the right amount of Items on the shelves if there is a large sales variance in a store. There are two problems that arise when facing irregular selling behaviour. First, the replenishment system must be flexible enough to cope with the increased complexity. For example, I must be able to order on one day 50, on another only 5 items of a certain product. The case pack size (CU/TU ratio) will play a role here as it determines the minimum possible order quantity. Nevertheless, much more difficult to handle is the second effect. If, in addition to the high sales variance of a certain product, there is also low predictability, the replenishment becomes very complex.®^ The amount of safety stock is strongly influenced by this parameter: higher sales variance leads to higher inventory levels (Krane 1994; Fafchamps, Gunning et al. 2000; Alicke 2003). The higher the demand uncertainty, the higher the OOS rates will be. For store employees, it is very difficult to keep track of inventory levels for products that have high sales variance. Besides, some store managers prefer regular ordering patterns, i.e. they order every second Monday a case pack (cf. Broekmeulen, van Donselaar et al. 2005). It is obvious that the bigger the sales variance, the worse such simple ordering patterns will work. For automatic systems, such as ASR2, inventory visibility is not affected by the sales rate or variance, the system always knows the quantity in the stores (inventory records accuracy aside). Besides, ASR2 systems decide every replenishment period (i.e. every day in the case of MYFOOD) whether it is necessary to place the order. For these reasons, one can expect that store personnel will have more difficulties replenishing correctly high-sales-variance products compared to ASR2 systems. Consequently, the following relationships are expected:
This procedure goes hand in hand with Backhaus, Erichson et al.'s (2003) recommendations, as the authors emphasise the necessity of having a priori hypotheses about the effects of the independent variables on the dependent variable before using ANOVAs. For the products analysed in this thesis it is not possible to say whether a product has a good predictability or not. To be able to compare the variance of slow and fast moving products the sales coefficient of variance is used. This coefficient is calculated by dividing the daily sales' standard deviation of each product by its mean sales.
72
4. Development of Models
H1a: ASRO products^^ with high sales variance have more OOSs than products with a low sales variance. H1b: For ASR2 products, the influence of sales variance on the OOS rate is smaller than for ASRO systems. H1c: ASRO products with a high sales variance have higher inventory levels than products with a low sales variance. H1d: For ASR2 products, the influence of sales variance on inventory level is smaller than for ASRO products.
Speed of Turnover A product's purchasing frequency can have an influence on OOS rate. In the study by Andersen Consulting (1996) the OOS rates were higher for fast-moving products. Products that sell very often per day have to be replenished more often than those with lower sales. Therefore, the danger of an OOS is far greater. On the other side, very slow moving goods that are only seldom sold have the danger of being overlooked when ordered manually. In the study by Stolzle and Placzek (2004) slow moving products had the highest OOS rate. The authors explain this by the fact that slow moving products receive less attention. This statement is confirmed in the study by Andersen Consulting (1996), where the authors, in addition, stress the risk for these products of remaining OOS for a long time. Therefore, for manual systems (ASRO) a quadratic curve is expected, meaning that products with very low and very high speed of turnover will have the most OOS incidents. As automatic systems do not have the danger of overlooking orders, only a small influence of the speed of turnover on the OOS rate should occur. The inventory level in a store is determined by logistics aspects but also by marketing aspects. Retailers want to avoid having too few products on the shelf. The reason for this is explained by one interviewed store manager, who said, "we create demand by having full shelves." Therefore, one can expect to see for slow moving SKUs more on stock than is actually necessary from a logistical point of view. In addition, for some slow moving units a single case pack size lasts for weeks, therefore increasing average inventory levels. This means that the lower the turnover of a product, the higher will be the inventory range of coverage. The situation is similar for ASR2 systems. Their inventory level will also depend on case pack size and turning speed: the slow moving products will have higher inventory levels. Yet, there should be a difference for very slow moving products. While store managers tend to have a simple procedure for such items (e.g. order a case pack every third The wording "ASRx product" is a short form for "a product that is ordered through an ASR level x system."
4. Development of Models
73
week) (cf. Broekmeulen, van Donselaar et al. 2005), MYFOOD's ASR2 systems check the inventory position daily, therefore optimizing the ordering and thus reducing inventory. Overall, It is hypothesised: H2a: ASRO products have higher OOS rates for very fast moving and very slow moving articles than for the other products. H2b: For ASR2 products, the influence of speed of turnover on the OOS rate is smaller than for ASRO products. H2c: ASRO products with a low speed of turnover have higher inventory levels than fast moving products. H2d: For ASR2 products, the influence of speed of turnover on inventory levels is smaller than for ASRO products.
Price The effect of price on OOS and Inventory level is ambiguous. DeHoratius and Raman (2002) found that inexpensive goods were more likely to have inaccurate records than expensive ones, as expensive SKUs receive more attention. According to this fact, one would expect more expensive goods to be less often OOS, for both manual and automatic systems. Normally, store managers would try to reduce the amount of expensive goods on stock, as they are bounded capital. Yet, in the data set of this thesis the most expensive goods are only worth 17 euro, 90% of all goods are cheaper than six euro. This price range is very small compared to the range in DeHoratius and Raman's (2002) work, where the most expensive goods had a value of 2,800 euro. Yet, as Broekmeulen, van Donselaar et al. (2004b) show, the inventory costs are secondary for typical grocery retailers, as they only represent 10% of logistics costs. Much more important are the handling costs in the store, which are five times as high. Therefore, it is reasonable to argue that for the observed grocery retailer MYFOOD another relationship between price and inventory performance will appear. Assuming that the more expensive goods have higher margins, the store manager will tend to have a higher inventory level on the shelves to avoid any OOS and consequently customer dissatisfaction. Yet, it is not clear if this strategy is successful, as many sources from the just-in-time and lean management literature have stressed the negative impact of high stock levels on availability (Schonberger 1982; Krafcik 1988; Fisher 2004). The higher inventory levels may even worsen the availability rate of these products. For automatic systems, on the other hand, the product price should not play such an Important role, as the automatic system reorders the product despite Its price. Based on this research It is expected that:
74
4. Development of Models
H3a: ASRO products with high prices have more OOSs than lower priced goods. H3b: ForASR2 products, the influence of price on OOS is smaller than for ASRO products. H3c: ASRO products with higher prices have higher inventory levels than low-priced goods. H3d: For ASR2 products, the influence of price on inventory level is smaller than for ASRO products.
Case Pack Size (CU/TU) Trading units (TUs) are the cases used to transport the products to the stores and contain several small consumer units (CUs)-the units that are finally bought by shoppers. For a typical grocery retailer, the size of the case packs is the same for all stores, as it is often determined by the supplier without taking into account the store's individual demand. The consequence is high inventory levels. Broekmeulen, van Donselaar et al. (2005) found in their research that 96% of the items in their sample had a larger order case pack size than the stores needed in a reordering period. Therefore, stores typically order a single case pack unit or wait a longer time to place the order. The higher the CU/TU ratio, the stronger this effect and the more inventory will be on stock. On the other hand, the smaller the CU/TU ratio is, the better a store's replenishment system can work. It is not only manual systems that are influenced by the CU/TU ratio. Faick (2005) has shown in theory that ASR3 systems will only manage to reduce inventory compared to ASR2 systems if the CU/TU ratio Is small.®^ Therefore, it Is hypothesised that ASR2 systems will also be influenced by the CU/TU ratio. Big case packs force employees to place orders more seldom. Therefore, the danger of forgetting to replenish the item is greater. As automatic systems do not forget any order, one can assume that the OOS rate will be higher for manually ordered products compared to ASR2 ordered products. One effect that might cover the influence of the CU/TU ratio on OOS rate is shelf space. Depending on the shelf space allocation, a certain case pack may or may not
* It is worth noting that the resulting effect of the CU/TU ratio on performance depends on the store. Small stores are more affected by the CU/TU ratio than bigger stores. For example, for some non-food products of MYFOOD (e.g. shoe laces) the smallest case pack lasts in small stores for several months, while in a bigger store one case pack is sold within days. Therefore, instead of looking at the absolute case pack value, all analyses are made looking at the relative case pack size (case pack size divided by average daily sales), analogous to Falck's procedure.
4. Development of Models
7^
be beneficial. Wegener (2002) hypothesises that the OOS rate will strongly be influenced by whether it is possible to place the whole case pack directly on the shelf or whether part of its contents has to be stored in the backroom. The replenishment from the backroom not only creates high handling costs, it is also one of the major reasons for OOS incidents (Gruen, Corsten et al. 2002). Taking these studies into consideration, it is hypothesised: H4a: ASRO products with a small CU/TU ratio have smaller OOS rates than products with a bigger CU/TU ratio. H4b: ForASR2 products, the influence of CU/TU on the OOS rate is smaller than for ASRO products. H4c: Products with a small CU/TU ratio have lower inventory levels than products with a bigger CU/TU ratio. H4d: Products with a small CU/TU ratio have lower inventory levels than products with a bigger CU/TU ratio.
Product Size In the study by Andersen Consulting (1996) different findings on the effect of product size on OOS are encountered. One could argue that bulky products need more space on the shelf; therefore, retailers may opt to allocate them less shelf space than would be advisable from a logistical point of view. Consequently, small products will tend to have more facings to increase visibility to customers and therefore should be comparatively more on stock in the store. Wegener's (2002) data analysis of four product families from the beauty and household category showed that products with bigger package size were indeed more often OOS than products with small package sizes. The explanation for this effect given by an interviewed manufacturer was that especially for bulky products that are not visually pleasing (such as toilet paper), retailers tend to reduce the amount of facings on the shelves. Yet Wegener (2002) relativises this result: for other product categories the correlation effect of product size and OOS is likely to be smaller, as other product characteristics (such as speed of turnover) are probably the predominant influence. As regards inventory level, Broekmeulen, van Donselaar et al. (2004b) found in their analysis of two Dutch retailers that smaller products had more shelf space than required from a logistical point of view. Therefore, one can expect to see higher inventory levels for small products. For automatic systems, on the other hand, the product size should not play such an important role, as the automatic system reorders the product despite its size. This Is why it Is predicted that:
76
4. Development of Models H5a: ASRO products with a small size have lower OOS rates than big-size products. H5b: ForASR2 products, the influence of product size on the OOS rate is smaller than for ASRO products. H5c: ASRO products with a big size have lower inventory levels than small products. H5d: ForASR2 systems, the influence of product size on inventory level is smaller than for ASRO products.
Shelf Life Retailers are constantly aiming to offer products with longer remaining shelf life, as consumers demand fresh products (Petrak 2005). Therefore, for products with a very short shelf life (e.g. fresh pasta), store managers seek to reduce the amount of store inventory. It is clear that the smaller the average inventory of any SKU (compared relatively to its mean sales) the fresher the product will be due to a higher turnover. Therefore, it is reasonable to expect low inventory levels for short-shelf-life products. This finding is supported by physical audits such as that carried out by Broekmeulen, van Donselaar et al. (2004b). Small (safety) stocks reduce the danger of spoilage, but at the same time increase the danger of OOS. On the other hand, fresh products such as green grocery are, in terms of marketing, the most important goods for many grocery retailers. Therefore, skilled store employees will pay extra attention to these products. In the study by Stolzle and Placzek (2004), cooled products with a short shelf life had smaller OOS rates. For products with a longer shelf life the retailer can afford to have bigger quantities stored, as the danger of spoilage is less. Consequently, inventory levels will increase as well as the time between the ordering thus increasing the danger of OOS incidents. Overall, this leads to a quadratic effect of shelf life on OOS: products with very short and very long shelf life will experience more OOS situations. Yet, the effect of shelf life on OOS could be covered by other factors, such as sales variance, that have probably a stronger influence on the inventory level. The reasons mentioned for the influences of shelf life on inventory performance do not hold for automatic systems. Systems such as ASR2 do not pay more or less attention to products with a certain shelf life, consequently there should be no influencing effect visible. Overall, it can be expected that:
4. Development of Models
77
H6a: ASRO products with a very short shelf life have an increased OOS rate, as well as products with a very long shelf life. H6b: ForASR2 products, the influence of shelf life on the OOS rate is smaller than for ASRO products. H6c: ASRO products with a long shelf life have higher inventory levels than products with a short shelf life. H6d: For ASR2 products, the influence of shelf life on inventory levels is smaller than for ASRO products. The following table summarizes the product characteristics hypotheses H1-H6: Dependent Variables Independent Variables Sales variance Speed of turnover Price CU/TU Product size Shelf life
Out-of-Stock
\
Inventory Level
+
(HIa.ASRO)
+
0
(H1b,ASR2)
0
+ -+
(H2a, ASRO)
0
(H2b, ASR2)
0
+
(H3a, ASRO)
+
(H3c ASRO)
0
(H3b, ASR2)
0
(H3d. ASR2)
+
(H4a, ASRO)
+
(H4c. ASRO)
0
(H4b, ASR2)
+
(H4d, ASR2) (H5d, ASR2)
(H 1c. ASRO) (H1cl,ASR2) (H2c, ASRO) (H2d, ASR2)
+
(H5a, ASRO)
0
(H5b. ASR2)
0
+ -+
(H6a, ASRO)
+
(H6c. ASRO)
0
(H6b, ASR2)
0
(H6d, ASR2) 1
"+" Positive effect "-" Negative effect "+ - +" quadratic relationsiiip
(H5c ASRO)
"0" Smaller effect of ASR2 system compared to ASRO
Table 11: Overview of hypotheses concerning product characteristics
4.3.3. Hypothesis Development: Store Characteristics Several descriptive OOS studies have mentioned that there is a strong variance in the OOS rates between countries and retailers studied (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b). The same sources also indicate that the OOS rate differs from store to store, even though the stores' contexts (i.e. ASR system, distribution network, assortment, marketing activities, etc.) are fairly similar. This latter phenomenon will be analysed in the data of MYFOOD. In the following, hypotheses concerning the influence of store characteristics on the OOS rate are
78
4. Development of Models
derived, as the understanding of these influences is important when assessing the effect of ASR systems.
Different Performance Between Stores As stated before, one of the common results found in previous availability studies is that the OOS rate differs significantly between stores (e.g. Gruen, Corsten et al. 2002; Roland Berger 2003b; Stolzle and Placzek 2004). A transfer thought would be that the same relationship holds true for inventory performance (i.e. inventory levels and inventory variance). As MYFOOD managers are at liberty to change the parameters of their replenishment system they have an important influence on the inventory level. The hypotheses are therefore: H7a: The mean OOS level of a store differs from store to store. H7b: The mean Inventory performance of a store differs from store to store.
Shrinkage and OOS The pivotal question is how some stores manage to have lower OOS rates other stores facing the same logistical context and using the same replenishment systems. A possible answer could be that store managers decide willingly to reduce OOSs by having more shrinkage. Yet, the interviewed practitioners from MYFOOD stressed the importance of the capabilities of the store managers to the performance of the store. The literature has often stressed the Importance of good management for performance (e.g. Fahmi 1998). In MYFOOD's interviewees' opinion, stores with experienced mangers are able to reduce shrinkage and OOSs at the same time. Consequently, the hypothesis is: H8: There are some excellently-managed stores that are able to reduce OOSs and shrinkage at the same time.
SKU Density The more complex a store is, the more difficult it will be to manage. Raman, DeHoratius et ai. (2001a) report how difficult it is for employees ordering manually to keep track of a product's availability. Yet it would be misleading to compare the absolute number of products between stores. Bigger stores will tend to have higher sales and a broader product range, but at the same time there will be more store staff and more space on the shelves for the product. Therefore one cannot conclude that bigger stores will have higher OOS rates. A way to compare different-sized stores may be looking at the SKU density. The assumption is that the amount of
4. Development of Models
79
products per square metre is a good way of quantifying the complexity of a particular store. This variable has also been used by DeHoratius and Raman (2002). The authors were able to prove a positive correlation between the inaccuracy of records in a store and the SKU density. The authors argue that in crowded stores it is more difficult to track data inaccuracy, therefore increasing the danger of OOS incidents. Besides, the more products a store has per m^, the less space there will be on the shelves for each SKU, thus increasing the chances of an OOS. The hypothesis reflecting this considerations is: H9: Stores with high SKU density will have more OOSs than low-density stores.
Work Intensity The economic theory of work states that there is a quadratic relationship of work effort and efficiency (cf. Fairris 2004). Transferred to this context this means that the number of employees in the store should also have a quadratic influence on the quality of store operations. When there are very few employees in a store, every single one has to work more; exhaustion and stress can set in, thus reducing the efficiency of work. The replenishment of shelves and the accuracy of data will suffer. Gruen, Corsten et al. (2002) found overworked staff to be one of the reasons for OOSs. On the other hand, if there are too many employees, boredom can set in (Fairris 2004), motivation can sink (Connell 2001), the responsibility for tasks can become unclear, and store managers find it more difficult to coordinate tasks (Anon. 2004). Wegener (2002) states that the marginal rate of productivity sinks, therefore she expects stores with an average personnel intensity to have the lowest OOS rates. To be comparable, the number of staff should be seen relative to the size of the store. The bigger a store the more the staff necessary to manage the shelves accurately. Overall, the hypothesis can be formulated as: H10: Stores with too many or too few employees per m^ sales area will have more OOSs.
Store Managers' Experience For retailers that have a very decentralized decision system, the single store can be regarded as a little independent company. Therefore, the influence of the store manager as an entrepreneur on the performance of the store will be significant, as stated in classical entrepreneur theory (cf. Schumpeter 1935; Fahrni 1998). Stores with good leadership will tend to have lower OOS rates. The more experience the store manager has, the better the performance of the store will be as store operations have a radical impact on the OOS rate (Gruen, Corsten et al. 2002). In
80
4. Development of Models
the case of MYFOOD, store operations are still within the responsibility of store managers; they have a strong influence on replenishment parameters. A rather crude way to asses the experience of a store manager is by looking at the number of years he or she has worked in that particular store. The longer the store manager has been working in a particular store, the better the performance should be. Already Raman, DeHoratius et al. (2001a) confirm that store managers who have been working for a long time on a particular store had a smaller rate of misplaced SKUs (i.e. items that were in the backroom but not on the shelf). Another advantage of experienced managers is that they have gained knowledge about the demand behaviour of their customers. For instance, experienced managers know to quantify the influence of a soccer game in the city on beverage consumption. However, an effect of diminishing returns will most probably appear. This means that for managers that have been on site for a long time, additional years in the store only result in small improvements. Based on these statements it is hypothesised: H11: Store managers that have been working in the same store for a long time have lower OOS rates than other store managers.
Backroom Authors such as Krafcik (1988) and Schonberger (1982) from the just-in-time and lean management philosophy argue that inventory hides problems in manufacturing. Transferred to this thesis, this means that having too much stock can be counterproductive to the availability rate. First, in backrooms that are crammed with goods, it will be very difficult to achieve a high visibility. And second, high inventory Impedes process improvement, as errors of suboptimal replenishment systems are obscured by high inventories. Raman, DeHoratius et al. (2001a) report on a case in which a certain shop suffered a substantial deterioration in performance after the capacity of the backroom was increased. Fisher (2004, p. 14) therefore demands from retailers "to get rid of the backroom." If this is really recommendable can be tested by the following hypothesis: H12: Stores with large backrooms have higher OOS rates than stores with small backrooms.
Customer Satisfaction Finally, a question asked by many retailers is, whether it is worth putting effort into improving the OOS rate at all. Although a high availability is crucial for the satisfaction of consumers (Sterns, Unger et al. 1981; Angerer 2004) one could argue that by investing too much time in increasing availability (by having, for example.
4. Development of Models
81^
store employees replenishing the shelves more frequently) the direct service to the customer could suffer, as e.g. less check-out lanes are open. In addition accurate scanning of goods takes longer, and is therefore not welcomed by customers. Therefore, it is interesting to analyse whether there is indeed a positive correlation between customer satisfaction and OOS rate: H13: Stores with lower OOS rates have more satisfied customers than stores with high OOS rates. The following table summarizes the store characteristics hypotheses H7-H13: 1 OOS rate differs from store to store Inventory performance differs from store to store 1 Reduction of OOSs and shrinkage is at the same time possible
(H7a) (H7b) (H8)
1 The more SKUs per m^ the more OOSs
(H9)
1 Too many and too few staff increases OOSs
(H10)
1 More experienced store managers have fewer OOSs
(H11)
1 Bigger backrooms lead to more OOSs
(H12)
1 Stores with low OOS rates have higher customer satisfaction
(HI 3)
Table 12: Overview of hypotheses concerning store characteristics
4.3.4. Hypothesis Development: ASR Characteristics The last group of hypotheses test one of the main questions of this thesis (Q2): can retailers achieve a better replenishment performance by using higher ASR systems? From the theoretical research of ASR-related topics®^ it Is known that ASR systems should be able to reduce OOSs (cf. Achabal, Mclntyre et al. 2000), decrease inventory level (cf. Ellinger, Taylor et al. 1999), have lower inventory variance and show robust results compared to manual systems (cf. Gloss, Roath et al. 1998). Therefore, these performance measures have been chosen to test the capabilities of higher ASR systems.
OOS Performance and ASR Level Fisher, Raman et al. (2000) point out that many retailers still use very basic methods for their ordering. According to the authors, rules of thumb and gut feelings are often used when determining the quantities to be ordered. They see a great opportunity in mixing "art and science" by using the vast possibilities of new technologies and data See research on ARPs in section 3.2.
82
4. Development of Models
availability to improve retailers' replenishment systems. Such technologies are the ASR systems. It would be a major relief for retailers to have more reliable replenishment system, as 72% of all OOSs are caused by faulty store ordering and replenishment practices (Gruen, Corsten et al. 2002). The reports from practitioners (e.g. Beringe 2002) and the literature (e.g. Mau 2000) of the reduction of OOSs by ASR systems are very courageous, yet there are no tests available in academia that prove these statements. Therefore, this thesis aims to close this gap by showing that such systems do indeed have the potential of reducing OOSs caused by replenishment faults. As data for ASR2 systems is available, it is hypothesised: H14: If the ASR level Is chosen and parameterised correctly for a certain product category, ASR2 products have less OOS Incidents than ASRO products.
Inventory Performance of ASR Systems While for the OOS performance the situation is clear-the fewer OOS incidents in a store the betternt is slightly more difficult to assess inventory performance. There are various definitions of what a good performing replenishment system should optimize regarding the stock level over time. Two possible perspectives on how to measure inventory performance used in this thesis are: • Cost side: to reduce inventory costs, there should be as low stock in the stores as possible, therefore, the aim is to minimize the absolute inventory mean. Q Marketing side: there should be little variance between the stock levels each day, as the customer expects full shelves at any time. Consequently, the ordered quantity should be equal to the demand. The result of such a strategy is a constant inventory level. Effective replenishment systems should minimize inventory variance.®^ Depending on the characteristics of the retailer, higher ASR systems should improve inventory performance, as long as the adequate level of ASR is not reached. The MYFOOD data includes data from three different automatic systems, which can be compared to ASRO systems. While ASR2 and ASR3 products should benefit from the automation, an ASR2* automation is rather crude. ASR2* systems use the same reorder parameter for many products. As products might exhibit different sales patterns, the replenishment will probably be suboptimal. Overall, it is hypothesised:
^^ Whether the shelves will be constantly filled or not, will also strongly depend on store operations, as employees have to refill the shelves from the backroom. However, a replenishment system that ensures that at any time the same quantity of goods is in the store is the basis for a good refilling of the shelves.
4. Development of Models
83
H15a: ASR3 products will have a better inventory performance than ASRO products.^^ H15b: ASR2 products will have a better inventory performance than ASRO products. H15c: ASR2* systems will not perform better than manual systems. The Figures 14-16 show an overview of the hypotheses.
Figure 14: Overview of hypotheses, product characteristics '
' The underlying assumption is that the ASR level was chosen and parameterised correctly for the product category that is examined in the sample. It is worth noting that these three figures are not meant to be a model, even if they have some visual similarities to structural equation models. These figures are an illustration of the three different groups of hypotheses that will be tested in chapter 5. On the left side, there are the independent variables, on the right side the dependent ones. The tests of the hypotheses are carried out independently of each other. Furthermore, they are based partly on different data bases.
84
Figure 15: Overview of hypottieses, store characteristics
Figure 16: Overview of hypotheses, ASR characteristics
4. Development of Models
85
5. Quantitative Analysis
5. Quantitative Analysis To be able to test the hypotheses formulated in the previous chapter, two different data sets of the company MYFOOD, a major central European grocery retailer, are analysed.^^ One part of the data stems from a physical OOS audit project that the KLOG conducted together with this retailer in summer and autumn 2004. In addition, historical sales and POS data was extracted from the retailer's ERP system to complement the findings. The quantitative analysis made on this dataset consists of three parts, analogous to the three groups of hypotheses. First, correlations between OOS rate and product characteristics are explored. Second, the influence of store characteristics on the OOS rate is examined. Finally, overall replenishment performance is tested. Table 13 gives an overview of the datasets that are used for the hypothesis testing and which will be presented in this chapter. Hypothesis
Data Set
H1-H6
Dataset 1
Question Answered
1
Which product characteristics influence the OOS rate and inventory level? Is there a difference between manual and automatic systems?
H7-H13
Datasetl
H14-H15
Dataset 1
Which store characteristics influence the OOS rate? Is there a difference between manual and automatic systems?
and 2
Do ASR2 and ASR3 systems perform better than ASRO systems?
Table 13: Overview of the utilization of the two datasets for hypothesis testing
5.1. Sample and Methodology 5.1.1. Datasetl: Testing of Out-of-Stock Hypotheses Researchers who want to empirically test hypotheses concerning OOS incidents require first of all a reliable data sample. To obtain the availability rate of products on the shelves is, however, a difficult task, as today's IT systems only store the quantities in the store, without making a difference between the backroom and the sales area. To overcome this problem, an on-shelf availability project was launched in summer 2004 together with the European retailer MYFOOD. In the context of this project, 100 products were chosen across seven different categories and their on-shelf availability was audited manually. The choice of the products ensured that
For more information about the retailer MYFOOD, refer to chapter 6.
86
5. Quantitative Analysis
tiiere was at least one article per each of the seven most important categories of the retailer.®® The top-selling products were selected in each category (e.g. bananas in the category green grocery), as their availability has a major impact on the retailer's financial results and customer satisfaction. In the sample are also products that are very important for customers as they are difficult to substitute (e.g. yeast in the category bakery products). In addition, care was taken to have all replenishment channels represented in the sample. This means that products were chosen that are distributed through regional and national DCs as well as products that are supplied directly from the manufacturers. Finally, products in promotion were taken to check for this influence as well. The final product sampling was determined by logistics and marketing directors of MYFOOD together with KLOG project members. The next step was to select the stores to be audited. Ten stores in a certain region of MYFOOD's home market were chosen taking care to represent three parameters that, according to the practitioners in the project, have most influence on current store performance: store size (large vs. small), location (rural vs. urban location) and historical performance (stores that had in the last years above and below average performance). The judgement of the stores' historical performance was carried out by marketing and logistics directors of MYFOOD, who took Into account aspects such as financial performance, sales growth, happiness of customers, etc. Five students were sent during a period of two weeks in autumn 2004 to the stores to count the OOS incidents. Their task was to check twice a day the availability of the products, once in the morning and once in the evening. Therefore, this project resulted in an OOS data base of 100[products] * 12[days] * 2[measures per day] * 10[stores] = 24,000 data points. Furthermore, for 15 selected products, inventory levels were manually counted to check the accuracy of inventory records. At the end of each day, the students handed a list with ail OOS incidents to the store managers. The managers' task was to check the underlying reasons for the missing products from a provided checklist.^° Store managers chose from this list the reasons for the OOS incidents for each SKU. The statements made by the store managers were sent to headquarters and were checked for plausibility by logistics and marketing
The seven most important categories identified by MYFOOD managers are: packaged food (e.g. coffee), green grocery (e.g. tomatoes), non-food (e.g. toothpaste), dairy products (e.g. Parmesan cheese), meat and fish (e.g. minced meat), bakery products (e.g. white bread), frozen (e.g. French fries). The products in each of these categories are all distributed and ordered through the same logistics system. ^° This checklist was analogous to the 13-items list ECR Europe has published (see Roland Berger 2003b).The ECR-list had to be enhanced by one OOS root cause: "Problems with the replacing of an SKU." This means
5. Quantitative Analysis
87
representatives. The employees in the stores, as well as the store managers, were Informed that students would be in the stores during two weeks making a physical audit. But they were not informed what exactly was being measured. In personal face-to-face meetings, store managers were informed that the aim of the project was to measure the performance of the logistics system, not to check the performance of store staff. The managers were requested not to let the daily business be in any way influenced by the physical audit. This information strategy lead to the desired result, as the OOS rate for the first week, compared to the OOS rate in the second week, differed only by 0.1 percentage points.^^ There were several cases where products were OOS for several days in a row, although the store manager could have intervened at any time. The results of other similar OOS audits undertaken before by other researchers differ radically from these results. Stolzle and Placzek (2004) had In their study a reduction of the OOS rate of 26% between the first, anonymous week and the second week, where the employees were informed. To calculate the weekly OOS rate of each product, the daily measures per product are converted into a percentage value, the same method as used by Gruen, Corsten et al. (2002). For instance, if an article was found to be absent from the shelves three times in the first week, the OOS percentage is 25%.^^ For the final analysis, some products had to be discarded from the 100 original products available in Datasetl. These were products that were during any of the two weeks on promotion or out-listed. The reason for this is that promotional articles are not ordered regularly through the systems, instead the central office sets a fixed starting amount for each store (push-replenishment). As this procedure is executed manually, it would be not possible to compare the results of the ordering systems with each other, as the promotional articles change ASR system during the promotion. One article was OOS at ail times in the store, as It was out-listed and the new product was not available for order. Overall, the final sample consists of 84 products in 10 stores, as can be seen in Figure 17. The unit of analysis is in most of the following each SKU in each store. Therefore, the sample population consists of n=840 units.
that for some discharged SKUs the new replacing product was not yet available for ordering, therefore resulting in stock-outs. This statement is only valid for non-promotional articles. The OOS rates of products in promotion differ radically from one week to the other. ^^ 3 divided by 12, as during six days two daily measures were conducted.
88
5. Quantitative Analysis
Figure 17: Distribution of the 84 products in Datasetl
During tiie OOS project additional information on each SKU was acquired: product price, product size, shelf life and the CU/TU ratio. Another source of information was MYFOOD's ERP system, which provided for each product the delivered quantity and the daily sales. This data was used to calculate inventory performance (inventory level and its variance), the mean sales variance and the speed of turnover of each product.^^ As the hypotheses H7-13 relate to the influence of store characteristics on the performance, some additional store related variables were collected as well. These are the yearly shrinkage rate (in percent of the turnover), the SKU density (number of SKUs per m^ sales area), the personnel density (number of employees per 100 m^ sales area), the years the store manager has been working in the particular store, and the size of the backroom in proportion to the sales area. The products in the sample are classified according to the replenishment process. The methods used are manually (ASRO: 31 products) and automatically (ASR2: 53 products). The latter group has to be split into two subgroups. For some ASR2 products, the store staff can only change the ordering parameters based on a discrete index scale from 1 to 10. This means that store managers can only decide if they want low replenishment parameters and stock levels ("1") or very high ones ("10") without specifying the exact quantity. Besides, the store manager can only do this on a group level. This means, for example, that store mangers have to change the replenishment parameters of all batteries, not only of one special battery type. As one will see, this procedure changes completely the performance of the products.
^^ In the appendix, the method for the calculation of inventory levels from the ERP-data is presented in detail.
89
5. Quantitative Analysis
Consequently, the term ASR2* level is applied to make a differentiation to the usual ASR2 systems discussed in this dissertation. Table 14 summarizes the available store, product and ASR variables in use. Variable
Variable Name
Operationallzation
Product Variables 1 Sales variance
Coefficient of variance of the daily sales
Sales_coefvar
1
1 Speed of turnover (=sales per day)
Mean sales of an SKU per day: low, medium and high category
Speed_of_ turnover
1 1
Price in local currency (for all stores the same)
Price
1 CUn'U (case pack size)
Relative CU/TU ratio: number of consumer units per trading unit divided by mean sales
CU_TU_relative
1 Product size
Size of the product: small products (up to the volume of a closed fist) medium size products (up to the size of a one litre milk pack) and big products (for all stores the same)
Package_size
1 Shelf life
Shelf life in days
Shelfjife
|
1 Shelf life^
Shelf life squared
Shelfjife_squared
1
1 Product sales
Daily sales per SKU and store
Product_sales
1
Dally delivered quantity to the stores per SKU
Product_deliveries
Price
Product deliveries
1
Store Variables 1 store Identification
Number (1-10) to identify the stores
Store_ID
|
Size category of the store (as defined by the retailer): small, medium, large
Store_size
1
1 store location
Rural or urban location
Store_location
1
1 Shrinkage rate
Yearly shrinkage in % of turnover
Shrinkage
1
SKU density
Amount of SKUs in a store divided by sales area
SKU_density
Personnel density
Number of employees in a store divided by sales area
Personnel_density
store manager experience
Number of years the store manager has been working in the store
Store_manager
Ratio backroom size to sales room
Backroom_size
ASRO: manual ASR2, ASR3: automatic ASR2*: automatic, but crude parameterised
ASRJevel
store size
1 Backroom size
ASR Variable 1 Type of ASR system
1
1
5. Quantitative Analysis
90
Performance Variables 1 Inventory level
Inventory range of coverage^^
lnv_range_of_coverage
1
1
Mean of the stock level divided by average daily sales 1 Inventory variance
Inventory coefficient of variance
lnventory_coefvar
1 oos
All out-of-stocks
OOS
|
1 Order-related OOS
Only OOS incidents are considered, where the
OOS_only_order
1
underlying reason is an ordering fault/^ Weekly mean Table 14: Overview of variables used in the analysis
The descriptive statistics of the product characteristics by ASR group can be seen in Table 15. Although the manual group has much higher selling quantities, the sales coefficient of variance Is very similar for all three groups. The average CU/TU ratio is higher for the ASR2 group than for the ASRO group, meaning that the latter products have more flexibility when ordering. A clear difference in shelf life exists between the groups; the manual products tend to have the shortest shelf life.
Instead of using absolute inventory levels, in the analysis the inventory range of coverage is computed. The inventory range of coverage is the absolute inventory level divided by the mean sales. This relative measure allows the comparison of products with different sales behaviour. As an example: if a product has a mean daily inventory level of 100, is this a lot or not? It depends; if the store sells 50 units per day, than the inventory is rather low as the amount is sufficient to meet demand for two days (100 units divided by 50 units per day equals 2 days of inventory). If the product is sold on average four times a day, than the store has a substantial amount of inventory. The inventory range of coverage is in this case 25 days (100/4). ^ The possible OOS causes that are directly related to replenishing behaviour are: order was created too late, the last order quantity was not sufficient, forecasts were not accurate, during the creation of the order an en^or occurred, fault with transmission to the DC, other order-related faults. In the case of MYFOOD, 60% of all OOS incidents are order-related. In the following, the wording "OOS (order-related)" is used to point out that only these kind of OOS incidents are meant. ^ Some of these variables, like price, are originally metric. To make it possible to use them in ANOVAs, they are banded before with the help of 33.33% quantiles. The result is three equal size groups that are called "low," "med" (medium) and "high." Whenever a confusion of the metric and the ordinal variable is possible, the suffix "(banded)" is used.
91
5. Quantitative Analysis
1 ASR I Group ASRO
ASR2
Maximum
Mean
Sales per day [units]
482
.25
299.75
41.16
Sales coefficient of variance
481
.04
2.83
.53
Price (local currency)
Std. 1 Deviation | 57.82 1 .26 4.71 1
490
.0
27.5
3.94
Product size (category)
490
1
3
1.84
.74 1
CU/TU relative [days]
482
.03
40.00
1.61
3.10 1
Shelf life [days]
490
1
700
48.51
127.58 1
Sales per day [units]
295
.50
161.92
17.72
23.82 1
294
.00
2.50
.57
.27
310
.5
23.0
3.41
4.18
Sales coefficient of variance Price (local currency)
ASR2*
Minimum
N
Product size (category)
310
1
3
1.77
.79 1
CU/TU relative [days]
295
.01
20.00
2.83
3.52 1
Shelf life [days]
310
60
700
460.00
259.34 1
39
1.00
26.67
6.82
6.40 1
39
.28
1.12
.53
40
2.1
4.9
3.09
1.09 1
Sales per day [units] Sales coefficient of variance Price (local currency) Product size (category)
.18
40
1
1
1.00
.000
CU/TU relative [days]
39
.94
17.50
4.27
3.90 1
Shelf life [days]
40
700
700
700.00
.000 1
•
Table 15: Product characteristics ofDatasetl by replenishment system
In the two week period there was in average a 3.0% rate of order-related OOS. A first evidence of the performance differences between ASR systems can be obtained by regarding the OOS rate separated by replenishment group, as depicted on Table 16. The products ordered manually had an OOS rate of 4.7% and a standard deviation of 11.7%. This contrasts to the much lower values of the ASR2 and ASR2* groups.
92
5. Quantitative Analysis
N
Minimum
l\/laximum
l\/lean
Std . Deviation
840
0.0%
91.7%
3.0%
9.4%
1
|ASRO
310
0.0%
91.7%
4.7%
|ASR2
490
0.0%
29.2%
0.6%
11.7% 3.1%
1 1
|ASR2*
40
0.0%
20.8%
1.3%
4.0%
1OOS (order-related) 1 By ordering system
Table 16: OOS rates (order-relatecl)of the sample
This finding is consistent in almost all stores. As depicted on Figure 18 in all stores but in store 10, the ASR2 systems perform better than the ASRO systems. Store 10 has overall very low OOS values, therefore, the absolute OOS rate difference between the ASR systems is minimal. Another finding depicted in Figure 18 is that the good performance of the ASR2 systems is not repeated with in the case of ASR2* products. Theses products have in all cases higher OOS rates than the ASR2 ordered products; for two stores, the ASR2* products perfonn even worse than the manual ordered products.
Figure 18: Comparison of the replenishment systems by store
As with the OOS performance, a first indication on the inventory performance differences depending on the ASR system can be obtained by regarding the stock levels by replenishment group, as depicted on Table 17. The products ordered manually have larger inventory range of coverage and standard variation than the ASR2 and ASR2* groups. This difference is the more striking as the manual products ^^ The stores 1-3 are the smallest stores in the sample. Their assortment do not include ASR2* ordered products, therefore there is no value for these products in the figure.
5. Quantitative Analysis
93^
are mostly from fresh categories. These products should also have by far smaller inventory levels compared to the canned goods to keep the products fresh. ASR Group ASRO ASR2 ASR2*
Inventory range of coverage Inventory range of coverage Inventory range of coverage
N
Minimum
{Maximum
Mean
Std. Deviation
442
.00
45.01
10.89
9.07
287
.00
50.88
9.46
9.99
36
.33
45.09
10.73
8.44
Table 17: Inventory range of coverage ofDatasetl
5.1.2. Dataset2: Pretest/Posttest Analysis As valuable as Datasetl is, there is a small deficit in its analysis, namely that a true pretest/posttest comparison is not possible. When assessing the performance of the ASR systems, one is comparing different product categories with each other. As seen before, the mean characteristics of the products (e.g. shelf life) are slightly different between the ASR groups. One could argue that these differences are responsible for the findings seen before, namely that higher ASR levels tend to perform better. To be able to make final conclusions, it is necessary to have a true pretest/posttest methodology as described by Sheeber, Sorensen et al. (1996) to underline the findings. This means comparing the performance of the same products In the same stores before and after the introduction of a new ordering system and having a control group where the replenishment system is not changed. This method has the highest control possibility for biasing influences. The goal of this comparison is to find out whether there is a significant benefit in implementing automatic systems or not. Therefore, a second dataset was extracted from the EPR system. As no OOS data is available from the ERP system, the dependent variables in these tests are inventory level and variance.
For the pretest/posttest, data from the same 10 stores as in the OOS-audit were chosen. A period of 180 working days was defined, 15 weeks before and 15 weeks after the implementation of the new ASR system. This comparison was only possible for categories that experienced a change in the replenishment system in the last three years, namely dairy, beauty/household and non-food products. Only products that were included in the OOS study were chosen for this study, to profit from the comprehensive product information already collected in this first study. In order to increase the validity of the comparison it is important to reduce the effects of
5. Quantitative Analysis
94
non-controlled variables. One of these variables is seasonality. Therefore, when choosing the periods to compare, the same weeks In the year were taken to minimize the effect of holidays and seasonality. OthenA^ise, comparing the stock levels in summer with stock levels in winter could bias the result. For example, dairy products were automated in November 2004. Consequently, the weeks chosen for the test were the first 15 weeks of 2004 and 2005, respectively. Both data sets contain thus the weeks after Christmas and the Easter holidays. Another aspect taken into account is that there is normally a time period after the introduction of new ASR systems with strong irregularities in performance, as staff get used to the system. ASR systems need some time until performance stabilises. Therefore, the time period was chosen in such a way that there was at least a two months gap between ASR introduction and the performance measurement. Table 18 summarizes the periods chosen.
Category
Number of Products
Data Period "Before"
1 Dairy products
5
Week 1-15,2004
1 ControM
6
Week 1-15, 2004
1
Beauty and household 1 Non-food
12
Week 30-44,2003
4
Week 30-44,2003
1
14
Week 30-44,2003
Control2
Change of ASR System November 2004 (ASR0^ASR3) None (ASRO) May 2004 (ASR0^ASR2) May 2004 (ASR0^ASR2*) None (ASRO)
Data Period "After" Week 1-15, 2005 Week 1-15,2005 Week 30-44, 2004 Week 30-44, 2004 Week 30-44, 2004
Table 18: Dataset for the pretest/posttest
For each product the data from the inventory systems was extracted in analogy to Datasetl supplying in this way per day and store the delivered quantities and the sales. This procedure results in 41 [products] * 10[stores] * 180[days] * 2[measures] = 147,600 data points. In addition, information about whether and when a product was in promotion was provided. There was no logistical change for any of the groups during the period under examination; an unbiased comparison of the replenishment systems was thus possible.
Comparison 1: Dairy Products (ASR0->ASR3) In November 2004, a new ASR3 system was introduced for dairy products. Based on the sales of the preceding four weeks, the new system creates a forecast for the following days and dynamically calculates and places the new order. From the 100 products in Datasetl, the dairy products where chosen for further analysis as they
5. Quantitative Analysis
95^
experienced a change from ASRO to ASR3. Only products without promotional activity were included in the analysis, resulting in a sample of five products. The choice of non-promotional SKUs Is of most importance, as during promotions the inventory levels sometimes increased by a factor of 20. Having these products in the analysis would distort the results enormously. Besides, promotions are still handled manually. The products studied were yoghurts, cheese and eggs. As a control group (ControU), six other randomly selected products were taken: packaged (e.g. convenience food, frozen goods) and non-packaged food (vegetables, fruit). The control group products were ordered In the same way during the entire period, namely manually. As the data set consists of measures during 180 days in ten stores, this results in 9,000 data points for the ASR3 and 10,800 data points for the ASRO ControU group. These figures can be considered large enough to make significant comparisons. Descriptive statistics from these two categories can be seen in Table 19. ^78
ASRO ControU 1 Sales coefficient of variance Price Pacl(agesize [category] Shelf life [days] Speed of turnover [units sold per day]
ASRO ControU
ASR3 Dairy
ASRO Control!
Std. Deviation
Mean
Maximum
Minimum ASR3 Dairy
ASR3 Dairy
ASRO ControU
ASR3 Dairy
ASRO Control!
120
100
.27
.24
2.35
1.02
.82
.54
.38
120
100
1.00
.60
5.60
3.87
3.28
2.21
1.84
120
100
1.00
1.00
3.00
2.00
2.33
1.20
.75
120
100
2.00
10.00
350.00
12.00
63.67
11.60
128.62
120
100
1.14
1.80
317.44
349.60
41.55
53.80
66.38
ASR3 Dairy
.15
1.37 1 .40
.80 1 80.78
Table 19: Descriptive statistics of the dairy products (ASR3) and ControU group (ASRO)
When the dairy products data is split by stores the result is that on average, before the introduction of the ASR3 system, in four out of ten stores the dairy products had a bigger inventory range of coverage than the control group. After the change, this was still the case in only one store.
Comparison 2: Beauty and Household Products (ASRO -> ASR2) Typical products in this category are beauty and hygiene articles such as handkerchiefs, toothpaste, nappies and toilet paper, as well as household articles such as washing powder. A change in the replenishment system for these 12
The unit of analysis is again the SKUs per store and time period. The 90 days periods for each product are condensed into one mean value. Therefore, the number of n=120 for the ASRO group results from the computation: 6 products * 10 stores * 2 time periods (before/after) =120.
96
5. Quantitative Analysis
products was made in May 2004, changing from manual ordering (ASRO) to a minimun-maximun system (ASR2), in which a fixed order-up-to quantity exists for every product/store combination. These replenishment parameters are normally stable during the course of the year (no seasonality is taken into account) but can be changed by store managers at any time without involving the retail central office. The control group (Control2) consists of 14 randomly chosen ASRO products including fruit and vegetables, convenience food, meat and fruit juice. Analogous to the analysis of the ASR3 group in the last section, promotional activities are a major problem when analysing inventory performance. In contrast to dairy products, promotions are in this category much more frequent. Virtually all products were on promotion for at least a week during the period. Therefore, it was not possible, as before, to reduce the analysis sample to products without any promotional activity at all. Instead, the weeks with promotional activities were ignored when calculating stock levels and variances.^^ It was observed that the effect of a promotion lasts sometimes a couple of days longer than the promotion itself. Stores that had ordered manually too many SKUs still had above-average inventory some days after the end of the promotion. Therefore, the week following the promotion was taken out as well. The same procedure was applied to the control group, thus eliminating the promotional effect for these products as well. With this procedure, approximately 15% of the daily data is lost, but the gain in accuracy far outweighs this loss. The same method of dealing with promotions is also applied to the ASR2* group in the next section. As depicted in Table 20, the product groups have exactly the same sales variance. ASR2 products tend to be larger and have a lower rate of turnover than ASRO products.
This is the same method as used by the ordering system of MYFOOD to compute forecasts for ASR3 articles: dates with promotional activities are ignored.
97
5. Quantitative Analysis
N ASRO Control2 1 Sales coefficient of variance
ASR2 B&H
ASR2 B&H
Maximum ASRO ASR2 B&H Control2
Mean ASRO ASR2 Control2 B&H
Std. Deviation | ASRO ASR2 Control2 B&H
.16
.00
1.69
2.21
.63
.63
.26
.32
243
1
2
3.00
3.00
1.93
2.46
.46
.50
243
1
1
2.00
2.00
1.22
1.23
.41
.42
266
243
1
2
2.00
2.00
1.50
2.00
.50
.00
266
243
0.82
0.76
339.63
139.40
30.61
13.29
56.97
253
243
266 266
Shelf life [category] 1 Speed of turnover 1 [units sold per day]
1 Package size [category] Price
ASRO Control2
17.99 |
B&H: Beauty and household products Table 20: Descriptive statistics of the beauty and household group (ASR2) and Control2 (ASRO)
Comparison 3: Non-Food Products (ASRO -^ ASR2*) With just four articles, this is the smallest category studied. The products examined were batteries, baking foil, glue and stickers. The replenishment system for this category was automated at the same time (May 2004) as that for the beauty and household articles category. After May 2004, the non-food products have been ordered through an ASR2* system. As before, the main difference between the nonfood products and the control group (Control2) is the very low rate of turnover of the ASR2* products (cf. Table 21). Minimum AbKO Control2
Sales coefficient of variance
ASR2* Non1^^^ food
74
Maximum
ASRO AbKO ^°"^-°'^
ASR2* ^^^f ^^^_ ^c^-
.16
.36
Mean
Std. Deviation
ASRO
ASR2*
^3^^
ASR2-
^gRO
^^^'