|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 THREE ESSAYS ON TARGETING, CUSTOMIZATION, AND THE MARKETINGOPERATIONS MANAGEMENT INTERFACE By XIAOQING JING A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORID A IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010
2 2010 Xiaoqing Jing
3 ACKNOWLEDGMENTS I would like to thank my committee mem bers and other faculty members in the Department of Marketi ng at the University of Florida. Specia l thanks are due to Professor Michael Lewis and to Professor Bart on Weitz for their invaluable guidance on the preparation of this dissertat ion. I thank all my friends who have supported me on this special journey and, most importantly, my mother, who ga ve me the strength and courage to overcome the obstacles I encountered along the way.
4 TABLE OF CONTENTS page ACKNOWLEDG MENTS .................................................................................................. 3 TABLE OF CONTENTS .................................................................................................. 4 LIST OF TABLES............................................................................................................ 6 LIST OF FIGURES .......................................................................................................... 8 ABSTRACT..................................................................................................................... 9 CHA PTER 1 INTRODUCTION.................................................................................................... 10 2 STOCKOUTS IN ONLI NE RETA ILING ................................................................... 14 2.1 Introduction .................................................................................................... 14 2.2 Empirical Contex t........................................................................................... 20 2.3 Empirical Analysis .......................................................................................... 22 2.3.1 Explanatory Variables .......................................................................... 23 2.3.2 Estimation Results ............................................................................... 26 2.3.3 Validat ion ............................................................................................. 32 2.3.4 Robustness Tests ................................................................................ 34 2.4 Simulation Ex periments .................................................................................. 35 2.5 Discuss ion ...................................................................................................... 40 3 REWARDING SOME BUT PUNISHING OT HERS? AN EMPIRICAL STUDY OF FREQUENT SHOPPI NG DISC OUNTS .................................................................. 55 3.1 Introduction .................................................................................................... 55 3.2 Background .................................................................................................... 58 3.3 Conceptual Fr amework .................................................................................. 61 3.4 Analyse s ......................................................................................................... 71 3.5 Discuss ion ...................................................................................................... 79 3.6 Conclu sion ..................................................................................................... 81 4 AN EMPIRICAL STUDY OF THE IMPACT OF RESTRICTING RETURN POLICY ................................................................................................................... 93 4.1 Introduction .................................................................................................... 93 4.2 Empirical Contex t........................................................................................... 96 4.3 Model ............................................................................................................. 98 4.3.1 Purchase Incidence M odel .................................................................. 98
5 4.3.2 Return Ra te Model .............................................................................. 99 4.3.3 Return Dura tion M odel ...................................................................... 100 4.4 Empirical Analysis ........................................................................................ 102 4.4.1 Estimation Results ............................................................................. 104 4.4.2 Validat ion ........................................................................................... 108 4.5 Conclu sion ................................................................................................... 109 5 CONCLUSION...................................................................................................... 115 LIST OF RE FERENCES ............................................................................................. 118 BIOGRAPHICAL SKETCH .......................................................................................... 127
6 LIST OF TABLES Table page 2-1 Descriptive statistics ........................................................................................... 47 2-2 Correla tions ........................................................................................................ 47 2-3 Key variabl e definit ions ....................................................................................... 47 2-4 Shippi ng fees ...................................................................................................... 48 2-5 Fit st atistic s......................................................................................................... 48 2-6 Estimati on results ............................................................................................... 49 2-7 Segment descrip tions ......................................................................................... 51 2-8 Holdout samp le vali dation .................................................................................. 51 2-9 Dynamic validation ............................................................................................. 52 2-10 Repeat pur chase model ..................................................................................... 52 2-11 Simulati on results ............................................................................................... 53 3-1 Customer descrip tive stat istics ........................................................................... 84 3-2 Promotion descrip tive stat istics .......................................................................... 84 3-3 Hypothesized effects of card-re s tricted promoti ons on segments...................... 84 3-4 Variable definitions ............................................................................................. 85 3-5 Store traffic and basket size estimati on results................................................... 86 3-6 Descriptive measures of chosen products categories ........................................ 87 3-7 Brand choice es timation re sults .......................................................................... 89 3-8a Category incidence result s number of promot ions ........................................... 90 3-8b Category incidence result s depth of promot ions .............................................. 91 3-9 Hypothesized relationship s and empirica l find ings ............................................. 92 4-1 Descriptive statistics ......................................................................................... 111 4-2 Descriptive statistics pre vs. post policy change............................................ 111
7 4-3 Variable definitions ........................................................................................... 111 4-4 Fit st atistic s....................................................................................................... 112 4-5 Estimati on results ............................................................................................. 112 4-6 Holdout samp le vali dation ................................................................................ 114
8 LIST OF FIGURES Figure page 2-1 Year 1 profit ve rsus stock out rate ....................................................................... 542-2 Customer equity versus stockout rate................................................................ 543-1 Concept ual map................................................................................................. 92
9 Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy THREE ESSAYS ON TARGETING, CUSTOMIZATION, AND THE MARKETINGOPERATIONS MANAGEMENT INTERFACE By Xiaoqing Jing August 2010 Chair: Barton Weitz Major: Marketing This dissertation focuses on the opportunities and challenges faced by retailers in coordinating Marketing and Oper ations Management activities. Two issues of special interest are (1) the impact of coordinating marketing and operation strategies, e.g., the impact of customized operationa l and marketing activities on customer behavior, and (2) the actions firms can take to coor dinate marketing and operations management, focusing especially on the opportunity to cu stomize interface decisions based on the vast amount of customer inform ation now available to firms. The thesis consists of three empirical st udies, all of which collectively investigate issues related to the marketing/operati ons interface and the costs and benefits of customized marketing. Chapter 2 examines how product st ockouts influence consumer purchase behavior and evaluates the potent ial value of customized inventory management; Chapter 3 studies how consumers react to customized promotions based on loyal programs; and Chapter 4 studies how restricting return policies affects consumers purchase and return behavior, whic h influences both the retailers revenue and operations cost.
10 CHAPTER 1 INTRODUCTION The interface between marketing and operations management is becoming increasingly important as advances are made in supply chain management (the supply side) and customer rel ationship management (the demand side). Retailers, and other businesses, can realize significant benefits from coordinating these activities (Ho and Tang 2004, Agrawal and Smith 2009). Interfac e research has generated rising attention in both the marketing and operat ions management literature over the past few years. For example, the Operations Management literature has begun to incorporate important marketing variables such as pricing and advertising into traditional operations management problems such as inventory, a ssortment, and location models (e.g. Smith 2009 a, b, Anupindi, Gupta, and Venkataramanan 2009), and ma ny operations research methods have been adopted to model and solve marketing issues such as direct mailing decisions (Bitran and Mondschein 1996) On the marketing side, an emerging literature studies the impact of operat ions management decisions on consumer behavior, for example, how consumers react to product stockouts (Anderson, Fitzsimons, and Simester 2006) and to assort ment reductions (e.g., Boatwright and Nunes 2001, Borle et al. 2005). One important advance in marketing pr actice and literature is customer relationship management, which has generated a lot of attention because of the availability of a vast amount of individual-level data from consumers as well as firms capability to customize their marketing effort s towards each individual customer at a low cost. Existing literature has ex amined how the traditional market ing mix, such as price, loyalty programs and communication, can in fluence consumers in different ways and
11 identified the potential benefit of more customized mark eting programs (e.g., Lewis 2004, 2005, Anderson and Simester 2004, Simest er, Sun, and Tsitsiklis 2006). More theoretical work has begun to examine the im pact of such customization and targeting efforts in a competitive mark et and/or when customers react strategically (Villas-Boas 1999, Iyer, Soberman, and Villas-Boas 2005, Pazgal and Soberman 2008). The availability of customer-level data and a firms ability to collect more individuallevel data also offer new opportunities to coordinate marketing and operations management decisions. Traditionally, operations managers often make decisions based on aggregate-level demand information such as sales, and customer heterogeneity (i.e., the individual-level or segment-level differences among consumer s) and/or the longterm impact of these decisions on cons umer behavior are often neglected. The availability of a vast amount of customer data and database marketing techniques, together with more integrated information systems and decision support systems, allows retailers to integrate consumer insights in to the decision-making process of people outside marketing such as operations managers. Such integration may not only improve the performance of operations managers, but also offers new tools that can help marketing managers improve the e fficiency of customer management. This dissertation focuses on two main issues related to the above discussion: First, the impact of coordinat ing marketing and operation stra tegies, e.g., the impact of customized operational and marketing activiti es on customer behavior, and second, the actions firms can take to coordinate marketing and operation management, focusing especially on the opportunity to customize interface decisi ons based on the vast amount of customer information now available to firm s. The first issue is of importance because
12 all the effort of customization and tar geting must be based on a good understanding of how consumers view these activities and wh ether and how they respond strategically. The second issue is interesting because, as st ated above, most of the effort in the area of customization and targeting has focus ed on traditional marketing mixes and little attention has been paid to decisions that are made by people outside marketing such as operations managers (e.g., inv entory management) or jointly made by marketing and operations managers (e.g., produ ct return management). This dissertation consists of three empi rical studies, all of which collectively investigate issues related to the market ing-operations management interface and the costs and benefits of customized marketing. Chapter 2 exami nes the value of incorporating the yield managem ent concept in managing inv entory and order fulfillment for online retailers. More specifically, usi ng the transaction histor y data from an online grocery retailer, I model how consumers purchase behavior is influenced by their stockout experience and use these consumer insights to illustrate the potential value of customized inventory management. Chapter 3 studies how consumers react to one type of customization effortcustomized prom otions based on loyalty programs. More specifically, this chapter conducts an em pirical study to examine how customized promotion activities (loyalty-card-holder-onl y discounts) influence consumers in-store purchase behavior at different levels, includ ing store expenditures, category incidence, and brand choice. Chapter 4 uses a dataset fr om a multi-channel reta iler that restricted its return policy during the data collection pe riods to study another important interface decisionmanaging product returns. The lenien cy of the return policy often influences both the firms marketing performance (e.g., sa les) and operations cost (e.g., number of
13 returns and return duration). A joint model of consumer purchase incidence, return rate, and return duration is developed to examine and evaluate the impact of restricting return policy, which has been a growing practice in recent years.
14 CHAPTER 2 STOCKOUTS IN ONLINE RETAILING Demand uncertainty and supply rigidity often create inventory shortages. Inventory shortages can have adverse c onsequences on both shortand long-term customer behaviors. Using data from an onli ne grocer, this chapter has the following three main objectives. First, we empirically investigate the true nature of shortage costs by examin ing how stockouts and fulfillment ra tes impact customers purchase behavior in both the shortand long-term. Second, we study how consumers reactions to inventory shortages differ acro ss customer segments. We use these insights to illustrate how the seller may benefit from prioritizing fulfillment policies based on customer traits. Third, we study how the impact of inventor y shortages varies across product categories. Based on these insights we discuss how the se ller should allocate limited resources in improving the order fulfillment across these categories. 2.1 Introduction Ensuring pr oduct availability is a basic ma rketing function that impacts both shortterm revenues and long-term customer loya lty (Amato 2009; Grocery Headquarters 2006). Meeting demand with appropriate su pply, however, may be challenging and expensive when sellers face demand unc ertainty and/or supply rigidity ( Forbes Dec 14th, 2006). In fact, product shortages due to inaccurate demand forecasts are a fairly common problem. In the grocery industry, 8% of products in supermarkets in North America are out-of-stock at any particular time and the rate is 15% for products on promotion (Grocery Manufac turers of America 2002). While the immediate consequence of invent ory shortages may be significant lost sales and revenue (Andersen Consulting 1996), the impact of stockouts cannot be fully
15 evaluated without underst anding how inventory shortages influence long-term customer behavior. This basic issue has been considered in the stochastic in ventory literature, where models often determine inventory le vels by balancing expected revenues with shortage and overstocking costs (Zipkin 20 00, Porteus 2002). Shortage costs are intended to capture the impact of stock outs on consumer purchase behavior in subsequent periods (Hall and Porteus 2000; G aur and Park 2007). Unfortunately, there is a limited literature that atte mpts to quantify shortage costs by empirically investigating how stockouts affect long-term consumer purchase behavior (Anderson et al. 2006). However, technology trends, such as the proliferation of dat a warehouses, and the development of customer metrics, such as customer equity (Blattberg and Deighton 1996) and customer lifetime value, now make it increasingly possible to evaluate the long-term consequences of stockouts. Online retailing is a particularly inte resting context for studying inventory management, stockouts and customer management. The use of a virtual interface between the retailer and the customer, and the abi lity of online retailers to identify individual customers, create opportunities for innovat ive approaches to inventory management. First, the separat ion between consumers and firms means that customers view images of products rather than physical products. This is salient because the firm controls access to inventory. Second, online retailers can often identify customers prior to transactions being completed. This affo rds opportunities for customizing marketing efforts based on individual customer characte ristics. Third, t he online environment facilitates efforts to longitudinally track cust omers. This enables firms to better track the
16 long-term ramifications of stockouts. The purpose of our research is to investigate how fulfillment policies can impact the economic value of a heterogeneous customer base. In practice, the most common tactic fo r preventing stockouts caused by demand uncertainty is to increase inventory levels This approach, however, increases inventory holding costs and the likelihood that the firm will need to use clearance sales (Smith 2009). Inventory management is also incr easingly viewed in terms of consumer behavior. Anupindi et al. (2009) and Smith ( 2009) consider retailers assortment and stocking decisions when consumers have heterogeneous preferenc es and willingness to substitute. The yield management literature also offers interesting insights into how a seller may improve the prof it from a fixed amount of supply when facing demand uncertainty and consumer heterogeneity (Talluri and Van Ryzin 2004). Yield Management has been defined by American Airli nes (1987) as a set of tools that maximize passenger revenue by selling the right seats to the right customers at the right time. The emphasis on customizing ma rketing and inventory availability based on customer traits makes yield management systems interesting from a marketing perspective (Shugan and Desiraju 1999). Howe ver, yield management techniques overwhelmingly (Talluri and Van Ryzin 2004) focus on short-term profit maximization. A key yield management question is how inventory rationing may impact the long-term economic value of different customer segments. This discussion suggests several resear ch questions. First, how do consumers react to product stockouts in both the shortand long-run? Second, are there important segment level differences in response to stockouts? Third, do stockouts in different types of categories have different impacts on consumers? Fourth, how can insights
17 related to the preceding questions help sellers manage inventory in a manner that increases profits? In the current paper we address these questi ons via an empirical investigation into how stockouts impact the long-run economic value of a firms customer assets. Our analysis is conducted using data on individu al customer transactions from an online grocer. Like many retailers (Forbes 1999; Black 2001) this firm has experienced difficulties with order fulfil lment. The data provides an opportunity to assess how imperfect fulfillment impacts key CRM metrics related to customer retention and profitability. Consumer response is analyzed using a joint model of incidence and amount that accounts for the inter dependence between purchase incidence and purchase amount. We also use a latent cl ass approach (Kamakura and Russell 1989) to account for unobserved heterogeneity. The estimation results suggest that stock outs impact customer relationships in a complex manner for the focal retailer. Stockouts can have both positive and negative effects on customer retention. In the s hort-run we find that orders that are not completely filled have a tendency to increase future buying1 while in the long-run cumulative stockouts have a detrimental effect on customer retention. We also find the effects of stockout experiences vary across both customer segments and product categories. These findings sugge st that the firm under study c an increase profitability by strategically managing fulfillment. We use the estimation results to conduct a series of simulati on experiments that assess the effects of improvements in inventor y fill rates. We find that stockout rates 1 It is important to note that the firm does not employ a policy of automatic backordering.
18 have a nonlinear impact on the retailers custom er equity. For example, we project that eliminating all stockouts w ould increase single year contribution by 12.5% and long-term customer equity by 56.2%. However, decreas ing stockouts by just 50% would achieve much of the potential benefit as first year cont ribution is projected to increase by 8.8% and long-term customer equity by 37.1%. We also conduct simulation experiments t hat examine the benefits of prioritizing fulfillment based on customer characteristics. These results suggest that significant benefits can be achieved with relatively small overall reductions in stockouts. We find that prioritizing inventory for new custom ers, long-time loyal customers and baby product purchasers can yield substantial benefits while requiri ng only moderate improvements. For example, by decreasing the stockout rate experienced by these households to 5%, the retailer can achieve 73.5% of the potentia l improvements in customer equity. Notably this improvement would require only a 47% drop in the total rate of stockouts.2 We also evaluate the effects of decreasing stockouts in different types of categories. We find that lo w-penetration but high usage products should be treated with highest priority when managing stockouts. While these results are idiosyncratic to our retailer they do suggest t hat customization may be valuable to other retailers. Our research offers several contributions to the inventory and customer management literatures. First, the issue of stockouts has begun to receive increased interest in the marketing literature as re searchers have begun to examine the shortand 2 It should be noted that these findings are involve si gnificant assumptions about the cost side of prioritizing inventory. The prioritization of inv entory may have cost consequences that are beyond the scope of our research.
19 long-run consequences of stockouts on consum ers. Fitzsimons (2000) experimentally studies how item unavailability impacts sati sfaction and Anipundi et al. (1998) develops a procedure for estimating item level dema nd when stockouts occur. Anderson et al. (2006) examine the shortand long-term costs of stockouts using a field experiment. Anderson et al. examine the impact of a si ngle out-of-stock sit uation and alternative retailer communication strategies on long-te rm customer value. Differences between the Anderson et al.s study and the current research include the type of stockouts under study and the modeling approach utilized. The mo st significant difference is our focus on segment and individual level effects of stockouts. The emphasis on long-run customer res ponse to stockouts makes our work relevant to the CRM lit erature. The CRM literature includes research that has examined how customer asset value (Gupta and Lehm an 2005) is impacted by marketing mix elements such as pricing (Lewis 2005; A nderson and Simester 200 4), loyalty programs (Lewis 2004) and communications (Simeste r et al. 2006). The current research contributes by showing how product availability impacts customer retention. A notable finding is that stockouts have a particularly large impact when they occur early in the customer lifecycle. This is salient because it reveals the importance of a consumers early experiences on customers long-te rm economic value. Our approach also highlights the interaction between stockouts and customer characteristics and shows the value of considering marketing and operations data simultaneously. Our paper also contributes to the yiel d management literat ure. We offer an innovative application that show s the value of inventory cont rols in a non-traditional category. While yield management has been applied in a variety of travel industries our
20 work shows how segment level inventory cont rols can be of value in the retail sector. The results also illustrate the long-term economic consequences of yield management. These results add to a growing literature t hat examines behavioral response to yield management (Kimes 1994; Wangenheim and Ba yon 2007) and considers the link between CRM and yield management (Noone et al. 2003). The remainder of the paper is organiz ed as follows. In the next section we introduce the data and present descriptive analyses that illustrate several of the complexities involved in studying long-term consumer response to stockouts. Section 2.3 describes our modeling approach and Sect ion 2.4 discusses the estimation results. Section 2.5 reports findings from simulation experiments that investigate the potential benefits of segment level inventory rati oning. We then conclude with managerial implications, implementation issues and avenues for future research. 2.2 Empirical Context The data for the study is fr om an online retailer selli ng nonperishable grocery and drugstore items.3 The data contains records of all customer transactions through the firms first 14 months of operations. The sample for our empirical estimation was randomly drawn from the dataset. It incl udes 2283 customers who made at least two purchases.4 The average order size is about $58 and contains more than twenty items. Table 2-1 provides selected descriptive statistics. 3 The firm operated from a single warehouse and distribut ed nationwide. This structure made it difficult for the firm to sell perishables such as fresh meats and produce. 4 We use each customers first two observations to infer demographics and initialize transaction history measures.
21 For the subsequent discussion we define a stockout as any instance when a customer does not receive all items that were ordered.5 The stockout rate is fairly high, as about 25.4% of all orders ar e imperfectly filled. For the orders with stock outs, the fill rate in terms of dollars of merchandise delivered as a percentage of dollars ordered is 90.3%. For the sample, on average 3.59 purchases were made, $211 was spent and 0.9 stockouts were experienced. We also are interested in the effects of stockouts in specific types of product categories. For this analysis we categorize products based on usage and penetration rates. Specifically we use median splits to define categories as high or low usage and high or lo w penetration. We also break out two categories, baby and pet products, for detailed analysis. Table 2-2 presents correlations for several measures of interest. In this table the Orders corresponds to the number of or ders placed by an individual, Cumulative Revenue is the customers total spending, Stockouts is the number of orders that were imperfectly filled over the customers tenure, Stock out Rate is the number of stockouts divided by the num ber of orders, and Stock 1st is a binary variable that indicates the customer experienced a sto ckout on her first order. PET and BABY are binary variables that indicate purchases in the pet or baby categories. The correlations are instructive in several ways. First they illu strate a potential difficulty in assessing the relationship between lifetime value and stockouts. The positive correlation between cumulative spending and total stockouts suggests that service failures increase retention. This positive relationship occurs because customers with a high preference 5 Individual level stockout experiences are observable because product availability information was not revealed on the product webpage. Consumers were not aware of inventory short ages at the time of the ordering and only become aware when the order was received. When a stockout occurred the firm simply removed the missed item from the custom ers bill. The firm did not make any additional adjustments to compensate consumers for the inconvenience.
22 for the retailer tend to order more often and therefore experie nce more stockouts. From an analysis standpoint, the positive correlati on means that simple regression analyses of the relationship between CLV and stockouts will yield incorrect results. Furthermore, the correlations between stockout rate and th e cumulative buying measures are small (0.055 & -0.022). The lack of correlation may occur for a variety of reasons and highlights the need for more detailed analyses. 2.3 Empirical Analysis For the empirical analysis we use a modeling approac h developed by Zhang and Krishnamurthi (2004) to capture the consumer s joint decision of purchase incidence and purchase amount. First, a logit specificati on is used to model purchase incidence. The second component models the continuous nature of purchase amounts. The model also includes a structure that ca ptures the dependence between the two dependent variables. The full development is given in Zhang and Krishnamurthi (2004). We also use a latent class method (Kamakura and Russell 1989) to account for unobserved heterogeneity. The log-likelihood function of this model is written as: 01 11 loglog 2 loglogPr(0)Pr(1,) where 1 Pr(0), 1 1 Pr(1,)1 11 1 1it it i it it it it it it it it it itT NM II mmitmitit im tt mit X ZyZy X itit XX Z ZyLL I Iy I e ee e Iy ee e e logit ity (1-1) wherem is the proportion of latent typemcustomers in the population; it I is a binary variable which equals 1 if customerimakes a purchase in week t and 0 otherwise;
23 it X is a vector of independent va riables that influence the purchase incidence decision; it y is the observed purchas e amount for customeriin week t ; it Z is a vector of independent variables which may affe ct the purchase amount decision. 2.3.1 Explanatory Variables We use four categories of covariates in the model: Pricing and Transaction F ees, Household Demographics, H ousehold Transaction History Measures and Stockout Measures. Table 2-3 provi des variable definitions. Price and transaction fees. The scope of the retailers product mix creates a challenge for including pricing effects in the model. The retailer sells over 14,000 distinct products that are classified into several hundred categories. To account for pricing effects, we construct an individual-level discount index. TOTDISC is the average of the discounts available on the t op 50 selling items in each category weighted by the customers category activity. This measur e is large when discounts are abundant and low when prices are high. In addition to product prices, customers al so pay for shipping and handling. During the data collection period, the retailer used four sh ipping fee structures.6 Table 2-4 describes the different shipping fee stru ctures. For example, Schedule 1 charged $4.99 to ship an order of less than $50, $6.97 to ship an order of between $50 and $75, and $0 to ship orders that exce ed $75. Schedule 1 is unique be cause it includes an order size incentive since the largest order si ze category is assessed the lowest fee. Schedules 2 and 3 are increasing fee schedules that differ in terms of the magnitude of 6 The seller used the first shipping structure from week 1 to week 12, the second structure from week 13 to week 15, the third from week 16 to 32 and from w eek 37 to 55; and the fourth from week 33 to week 36.
24 the fees. Schedule 3 charges the lowest fees while Schedule 2 charges the highest. Schedule 4 is a free shipping promotion. We define three variables (SH50W, SHSUR1, SHSUR2) to describe the shipping fees. SH 50W is the shipping fee for orders lower than $50, SHSUR1 is the incr emental fee for orders high er than $50 but less than $75, SHSUR2 is the incremental fee for orders higher than $75. For example, for schedule 1 SH50W is $4.99, SHSUR1 is $1.98 and SHSUR2 is -$6.97. Household demographics. In addition to marketing va riables we also include customer information in the model. The first ty pe of customer specific data we consider is information inferred from the contents of customer baskets. Two variables, BABY and PET, are defined based on purchases. If a cust omer purchases baby products such as diapers, we infer that the family has an infan t. Similarly, a purchase of pet food indicates the presence of a pet. The Pet and Baby vari ables are used as examples of how basket contents can be used to infer important demogr aphic characteristics. For instance, baby product purchases reveal family size and com position while pet products are of special importance for online grocery retailers becau se they tend to be low-cost but bulky. Thirty three percent of the households are classifi ed as pet owners and nineteen percent have babies. Household transaction histories. Customer transaction data is also useful for explaining customer de cisions. Transaction history data is particularly important since it allows for the identification of different s egments of customers such as new customers or long-term loyal customers. For the purchase incidence model, we include the following transaction history measures: orderi ng rate (FREQ), time since last order (RECENCY), RECENCY squared (RECSQ) and t he log of the customers last purchase
25 amount (LMT). These variables are sim ilar to the RFM (recency, frequency, and monetary value) measures used in direct marketing (Hughes 2000). We also include the cumulative number of order s the customer has made (CORDERS). In the purchase amount model, we use the same set of variables with one exception. We use the log average purchase amount (AMT) instead of the log last pur chase amount (LMT). Stockout measures The focal variables for the analysis describe each customers experiences with stockouts. We define two main stockout measures. CSTOCK is the cumulative number of st ockouts the customer has experienced 7 and PSTOCK is a binary variable which indicates if the customer experienced a stockout on the previous order. PSTOCK captures w hat happened on the last order while CSTOCK is designed to capture the long-term ra mifications of inventory problems. In order to study the impac t of stockouts across product categories we define the categoryand product-level stockout m easures. Specifically, following Fader and Loddish (1990), we divide the categories into four types, high-penetration and highusage products (Staples), high-penetra tion and low-usage products (Variety Enhancers), low-penetration and high-usage products (Niches), and low-penetration and low-usage products (Fill-ins). This fr amework is common in the category management literature (Dhar et al. 2001, Ho ch 2002) and is useful because the four categories play different roles in driving store visits and spending. The category level stockout measures are defined similarly to the aggregate level stockout measures. For instance, CSTOCK_STAPLE is the number of orders in whic h an item is missed in a 7 A weekly decay parameter of 0.98 is used to construct the cumulative variables (CORDERS and CSTOCK). We estimated models with a decay parameter from 0.90 to 1. The results are robust to the decay parameter.
26 high-penetration, high-usage category and PSTOCK_STAPLE is a binary variable equal to 1 if the customers previous purchase in the high-penetration and high-usage category is not perfectly filled. We also separate out the stock outs for baby products and pet products. For the purchas e amount model, we use the f ill-rate of the customers previous order (PORDSHIPRAT) instead of PSTOC K. The fill-rate is defined as the ratio between the shipped and the or dered dollar amounts. A fill-rate of less than one indicates that the customer experienced a stockout. We also include interactions between stock out measures and transaction histories. These interactions are intended to assess whet her the impact of stockouts varies based on customer experience. For example, we ar e interested in how stockouts impact new customers. We define EARLYCUST as a binary variable which divides the customers lifecycle into two stages. EARLYCUST is set equal to 1 if the customer has made no more than 2 orders and the time since customer acquisition is less than 6 weeks. If these conditions do not hold the variable is se t to 0. Interactions between the stockout measures and the EARLYCUST indicator c apture how newer customers react to service failures compared to co mparatively exper ienced customers.8 2.3.2 Estimation Results Estimation results are provided in Tabl es 2-5 and 2-6. Table 5 prov ides the number of parameters, log-likelihoods, and Bayesian in formation criteria (BIC) fit measures for models that vary the number of unobserved types within the population. Fit statistics are also provided for model s that do not include interactions between 8 In our model, EARLYCUST enters as an interaction since we treat CORDERS as a continuous measure of the lifecycle. We also estimated a model with EARLYCUST as a main effect (see web Appendix D).
27 transaction measures and stockouts, and fo r models without category level stockout measures or interactions. The two-segment model with interactions pr ovides the best fit in terms of BIC and yields intuitive parameter estimates. Tabl e 2-6 provides the parameter estimates and the standard errors for the tw o-segment model. There ar e some elements that are similar across the two segments. For instance, the transaction history variables tend to operate similarly across the tw o segments. However, we find significant difference between how the two segments respond to mark eting instruments and to stockouts. The estimation results indicate that the population is comprised of two segments of customers. Table 2-7 provides descriptive statistics for the segments as determined by posterior probabilities. Segment 1 customers have higher or der rates and expenditures. As a result they tend to experience more st ockouts. Table 2-7 also includes predicted retention rates for each segment. The retention rate is calculated using a method proposed by Fader, Hardie, and Lee (2005). For both segments the retention rate is significantly lower when customer s experience high stockout rates.9 The impact of shipping fees differs for the two segments. For the first segment, which represents 80.3% of the total population, there is ev idence that higher fees for smaller orders reduce incidenc e. Shipping fees play a la rger role for members of Segment 2. The incremental c harge for medium size orders significantly decreases the rate of ordering. The shipping fees also significantly impact the second segments expenditures. The base shippi ng fee increases the average order size but the quantity surcharges (nonlinear pricing el ements) result in signific antly smaller orders. These 9 Overall 34.2% of the population experienced higher t han average stockout rates. The rates were 34.4 % for segment 1 and 33.8% for segment 2.
28 findings suggest that Segment 2 tends to be mo re responsive to shipping fee structures. In contrast, we find that TOTDISC has a significant positive effect on purchase incidence only for the first segment. This is an interesting pattern of findings since the variables involve two different elements of pricing. The TOTDISC measure involves prices over a large number of items while the shipping fees are single prices. The results suggest that members of the two segments emphas ize different elements of pricing information in their decision making. The estimation results indicate that sto ckouts impact consumer behavior in several ways. We begin the discussion of the stock out effects by considering Segment 1 and then describe how Segment 2 diff ers. From Table 6, we see that cumulative stockouts (CSTOCK) have a negative impact on purchase incidence for Segment 1. In contrast, the negative sign of PSTOCK suggests that a low fill-rate from the previous order increases the likelihood that a cu stomer makes another purchase. A possible interpretation of the results related to previous stockouts and order incidence is that customers are re-ordering to obtain produc ts that were missed in the preceding order. To evaluate th is possibility we investigat ed the relationship between category purchases and previous categor y level stockouts. We conducted the investigation at the category level because of the large num ber of individual products that would need to be evaluated if the analysis were conducted at the SKU level. For the analysis we estimated simple regression models that used ca tegory level quantities as the dependent measure and previous sto ckout and previous quantity shipped as explanatory variables. Overwhelmingly, these regressions yielded positive and significant estimates for the previous quantity shipped and more importantly for the
29 previous category stockout. For example, t he analysis of the pet category produced the following equation: Pet Units = .717 +.60*Previous Pet Units Shipped + 1.775 Previous Pet Product Stockout. The t statis tic for the previous stockout in this equation coefficient is 4.81.10 Overall, across categories 85% of the previous stockout coefficients were positive and significant at the .05 level. The likelihood that a custom er re-orders to replace missed items does vary across the customer lifecycle. The estimate for the interaction between PSTOCK and EARLYCUST is negative, suggesting that new cu stomers have a lower tendency to reorder missing items. The interaction between cumulative stockouts and EARLYCUST is also negative, which means the negative impact of service failures is more severe for new customers. Both results emphasize th e importance of avoiding stockouts for consumers in the early stages of the customer lifecycle. Evaluation of the stockout va riables at the category level is also instructive. The negative long-term effect mainly comes from the high-usage and the low-penetration (niche) products. This may be due to the fact that niche products are of particular importance to customers that purchase them. The odds ratio is 0.85 (exp(-0.165)) for CSTOCK_HULP. Baby products also have a significant negative long-term effect. In terms of comparisons with other parameters, the results indi cate that a 1 unit increase in CSTOCK_HULP or CSTOCK_BABY would have the same impact as a 1.23 or .91 increase in the base shipping fee (SH50W). The implication is that a stockout in these categories has about the same impact as a $1 in crease in shipping fees. In other words, 10 The firms management of stockouts merits discussi on. The data used in the analysis is provided by a start-up firm operating in a new category. The firms practices did evolve over time. For example, following about two years of operation the firm did begi n to remove stocked-out SKUs from the website.
30 to repair the relationship following a stockout the firm would need to decrease shipping fees by $1. The positive short-term effect comes fr om both the niche products and low-usage and high-penetration products (variety enhancers). The odds of a purchase increases multiplicatively by 1.27 when a customer ex periences a stockout in a niche category and by 1.19 when a customer experiences a stockout of a variety enhancer. This is interesting, as it suggests that variety enhancers have mostly a net positive impact on store visits for this segment of custom ers. High-usage and high-penetration products (staples) as well as low-usage and low-pen etration products (fill-ins) do not have significant influence on purchase incidence. In terms of purchase amount, we find t he impact of stockouts varies across categories. For example, stockouts in Niche categories lead to larger subsequent orders. However, the effect is relatively small as the purchase amount increases by 3 percent when CSTOCK_HULP increases by 1. Our conjecture is that these products are of particular importance to consumers that purchase them. As such, previous stockouts may lead to future stockpiling strat egies. In contrast, the long-term effect of stockouts in the baby, pet and low-usage and high-penetration categories rates is reduced purchase amounts. The effects r anged from -3% to -3.5%. We also find that the previous fill rate (PORDSHIPRAT) has a negative impact on expenditures. This is an interesting finding as it suggests that customers tend to increase expenditures when more items are missed from the last order. This is consistent with the idea that stockouts have the immediate effect of causing customers to backorder missing items. The expenditure m odel includes an interaction between the
31 average order amount and the previous fill rate The coefficient for this interaction is positive and significant. This means that t he relationship between previous fill rate and expenditures is moderated by average basket size. The estimation results for the second s egment of customers are different in several respects. The coefficients for the previ ous stockout terms are all insignificant for both the incidence and expenditure models. The implication is that there is less linkage between subsequent purchases for this segmen t. For the cumulative stockout terms there are some similarities between the s egments. Similar to Segment 1, stockouts involving baby products have a significant ne gative long-term effect on incidence (odds ratio of 0.53) and a negative long-term im pact on purchase amount (purchase amounts decrease by about 17% with each additional stockout). There are also crucial differences between the segments. In contra st to Segment 1, stockouts in the high penetration categories (staples and variety enhancers) cause significant long-term harm. We also observe inte resting effects in the expe nditure model. Members of Segment 2 increase expenditures in response to cumulative stockouts in the high usage product categories. The purchase amounts increas e by 9% for each stockout in a high usage-high penetration category and 14% for a high usage-low penetration category stockout. Overall, the two segments have fairly different responses to stockouts. The first segment is generally more tolerant. They te nd to re-order to obtain missing items but repeated stockouts do cause attrition. In contra st, we do not find any positive short-term effects of stockouts for Segment 2. The larger negative parameter s associated with the
32 cumulative stockout terms also indicate t hat members of Segment 2 are more likely to defect based on cumulative stockouts. While the above differences may suggest the seller should favor the second segment when prioritizing inventory there are additional result s that call this notion into question. For instance, the transaction history parameters suggest that members of Segment 1 increase their purchase incidence and amount as the num ber of total orders increases. In other words, customers in th is segment become more valuable as they make additional purchases. This type of feedback effect is im portant as these customers are more likely to become prof itable if the retailer can successfully encourage repeat buying during the early portion of the relationship. 2.3.3 Validation To validate our model we used a random sa mple of 500 customers that were not included in the estimation sample. For t he purchase incidence validation we computed the Mean Absolute Deviation (MAD) and the Hit Rate for each observation in the holdout sample. For the purchas e amount decis ion, we calculate the MAD by taking the average absolute value of the difference between the actual purchase amount and the predicted purchase amount from our model. We calculated the above measures by first obtaining the segment-specif ic predicted purchase incidence probability and purchase amount, and then computing the predicted val ues for each household by averaging the segment-specific quantities weighted by the househol ds posterior segment probabilities. The results are shown in Table 2-8.11 For purchase incidence, the MAD is 11 We also conduct the holdout sample test for a two-segment model without any stockout measures (Appendix F) The model with stockout measures performs better in the holdout sample test.
33 0.135, and the overall hit rate is around 85%. For the purchase amount, the MAD is about $20. Furthermore, given that the primary goal of the analysis is to investigate the relationship between stockout occurrences and long-term customer behavior it is also useful to validate the dynamic implicatio ns of the model. To conduct the dynamic validation we use the estimation results to si mulate customer purchasing decisions over an extended term. These simulations are based on customer demographics, marketing tactics and the transaction history meas ures used in the demand model. The transaction history measures and associated in teractions provide a dynamic structure to the simulations. Specifically, time since previous purchase, cumulative orders, cumulative stockouts and average order size are updated after each simulated decision. We use the customers first transaction to predict whether the customer orders in specific categories and use conditional catego rical stockout rates to simulate whether the customer experiences a stoc kout within each category. Table 2-9 reports summary measures fo r the estimation and holdout samples and for simulations that use the demographics, initial c onditions and the number of weeks of observations for each individual. The result s for mean orders per customer, average order size and stockout incidence are similar. Minor differences occur because of small differences between the firms marketing policies and the simulated marketing policies. The most useful comparison between the observed data and the simulated results are the retention measures. For this analysis we look at the likelihood that a customer is retained at the end of the data collection period after experiencing a high stockout rate (a stockout rate higher than the average rate of 25.4%) or a low stockout rate. We
34 calculate the retention rate using the method in Fader, Hardie, and Lee (2005). In the actual data the retention rate for customers experiencing a low stockout rate is 55.6%. In the simulation the retention rate is 56.0%. In the case of high stockout rates, the retention rate in the data is 51.5% while in the simulation the rate is 51.7%. 2.3.4 Robustness Tests We also tested the robustness of our resu lts to various modeling assumptions (see Appendix C12). We examined alternat ive definitions of early customers, different initialization periods, eliminat ion of the log transformation for the monetary measures (order amount, average order size, and the last purchase amount) and adding EARLYCUST as a main effect. We also com pared the results of a hazard model to our incidence model results. The estimation resu lts are generally robust except for some minor changes. For example, in the purchase incidence model, the immediate effect of stockouts for baby products becomes insignifi cant when the init ialization period is extended to the first four observations. An additional issue is whether the use of a sample limit ed to repeat buyers creates biased results. Overall, the percentage of cu stomers that make only a single purchase is 63% of the customer base. We estimat ed a binary logit model to analyze whether stockouts on a first purchase influence subsequent purchasing. We use similar covariates as in the main model but some variables such as the RFM are not included. Table 2-10 reports the results. The first model uses only an aggregate level stockout measure, STOCK1ST. After controlling for demographics, marketing variables and purchase amount we find that a stockout on the first purchase does not significantly 12 All appendixes are available online or upon request.
35 affect whether a customer will become a re peat purchaser. These results suggest that the focus on repeat custom ers in the main model does not introduce bias. 2.4 Simulation Experiments In this secti on we consider how inventor y policies affect the value of customer assets. Specifically we investigate the benefits of the following rules for prioritizing inventory. 1. Broad reductions in stockout rates. 2. Inventory prioritized based on transaction history measures? 3. Inventory prioritized based on basket contents? 4. Inventory prioritiz ed based on customer type? 5. Stockout rates adjusted acro ss different product categories? For the studies the choice model results ar e used to simulate customer behavior for a period of two years. In each weekly period we simulate customer decisions in terms of purchase incidence and purchase quantity.13 The first row of Table 2-11 reports the results of a simulation that uses the historical stockout rate. This simulation prov ides a baseline set of results. We then run additional simulations that quantify the value of alternative inventory policies. For each simulation we report the followi ng results. The Revenue column reports the 1 year revenue total for the sample, the Contribution column reports the contribution of the sample after adjustments for product and shipping costs.14 The Mean Orders column reports the average number of orders per household and t he Mean Stockouts column gives the average number of stockouts per household. The fi nal two columns report the 13 The simulation is run for a population of 6164 custom ers (both 1 time and repeat customers) in two steps. The Repeat Buying Model is used to simula te whether the customer will become a repeat customer. For the repeat customers we use the main model to simulate their activity for two years. 14 A complexity in assessing the overall impact on profitability is that it is difficult to assess the eventual costs of the proposals. The costs of reducing stockout rates in specific categories require cost data that was not available.
36 retention rate at the end of the first year and an estimate of customer equity for the sample. The Retention Rate is the proporti on of repeat customers who are still active at the end of the first year. To estimate customer equity, the second year contributions are converted to a measure of long-term value using a discount rate of 20%.15 The customer equity estimate is t he sum of year 1 contribution and th is continuing value. The next several simulations examine the effect of across the board decreases in stockout incidence to 20%, 13%, 5% and 0%. The relative benefits of incrementally reducing stockout rate are graphically illu strated in Figures 2-1 and 2-2. Figure 2-1 shows the relationship between 1 year returns and the overall stockout rate while Figure 2 shows the relationship between long-term customer equity and the stockout rate. From Figure 1, we see that 1-year returns decrease non-linearly with the stockout rate. The benefit from reducing stockouts is greates t when the stockout rate is reduced from 20% to 13%. This is important because it suggests that minor reductions in stockouts may lead to substantial benefits and that there may be a point where reducing shortages will not yield sufficient returns to justify the needed invest ments. For example, a 50% reduction in stockouts (i.e., decreasing the stockout rate to 13%) improves 1 year contribution by 8.8%. This r epresents 70.2% of the gain t he retailer would achieve by eliminating all stockout s. However, reducing the stockout rate from 5% to 0% yields less than a 1% increase in contribut ion. The nonlinear pattern also exists when we look at the retention rate. For example, the retention rate increases by 4.8% when the stockout 15 The online grocery industry was new and rapidly evol ving at this time. A 20% discount rate is used for the sake of conservatism. Customer equity estimates calculated using higher discount rates are provided in Appendix F. The following conclusions are not sensitive to the discount rate.
37 rate decreases to 13%. This is 73.9% of pot ential improvement that is achievable by eliminating all stockouts. Figure 2-2 shows that stockouts have a larg er impact on customer equity than on 1 year contribution. Given that stockouts caus e attrition, this highlights the importance of inventory management to CRM. The elimination of all stockouts results in an increase of 56% in customer equity compared to an incr ease of 12.5% in 1 year contribution. A nonlinear relationship with stockouts also ex ists for customer equity. Customer equity increases by 37% when the firm decreases the stockout rate to a moderate level of 13% while a reduction in the stockout rate from 5% to 0% yields only a 3.1% improvement in customer equity. The next set of simulations study t he impact of rationing inventory based on transaction history data. The first simula tion evaluates the impact of favoring new customers. In this scenario, customers t hat are classified as Early Customers experience stockouts about 5% of the time. All other custom ers continue to experience stockouts at historical rates. This emphasis on the early portion of the customer lifecycle produces large near-term benefits. The improvem ent in single year contribution for this scenario is 8.4%. This repres ents 71.4% of the potential incr ease in one year return that is achievable if the total stockout rate were reduced to 5%. This improvement is striking because it is achieved with only a 17% reduc tion in stockouts. The second scenario examines the value of improv ing fulfillment rates for loyal customers. For this scenario we reduce the stockout rate to 5% for cust omers that have made at least 6 purchases. This scenario yields a 4% improvement and involves a 23% reduction in stockouts.
38 The 1 year contribution results sugges t that new customers should receive preferential treatment. Howeve r, when customer equity is used as the criteria the importance of maintaining relationships wi th loyal customers becomes clear. Reducing stockout rates for loyal customers to 5% generates a 23.6% improvement in customer equity compared with a 15.7% in crease for early customers. This suggests that the firm can realize 42% of the possi ble improvement in customer equity while only needing to reduce total stockouts by 23%. The third set of simulations analyzes the benefits of using basket contents to segment customers. Specifica lly, we evaluate scenarios that reduce stockout rates for customers that order baby or pet products. When we decr ease the stockout rate of households who purchase baby products to 5%, the 1 year profitability improves by 6.9% while the total number of stockouts needs to decrease by 20.5%. When we decrease the stockout rate of households wh o purchase pet products to 5%, the 1 year profitability improves by just 4.7% while the total number of stockouts needs to decrease by 41%. These projections suggest that households that order baby products should have a higher priority. While these pr oduct category results are unique to the retailer under study, they do suggest that ce rtain categories may more significantly impact consumers. For example, baby product s may be viewed as more important than other products since they may impact parents ability to care for children. The fourth set of simulations examines the benefits of basi ng fulfillment on unobserved heterogeneity. These simulations suggest that leveraging unobserved consumer heterogeneity in inventory rati oning decisions is less effective than customization based on transaction histories or basket contents. For example, when the
39 stockout rate experienced by the first segment is decreased to 5% the 1 year contribution increases by 8.7% While this is a decent impr ovement the required drop in total stockouts is far higher than in the ot her scenarios (58.5%). When we decrease the stockout rate of the second s egment to 5%, the 1 year cont ribution increases only 3.2% while the total number of stockouts would need to fall by 23.7%. These results suggest that the seller should not always prioritize based on sensitivity to service quality. The preceding simulations show that diffe rences in customer response mean that a substantial amount of the potential benefits from eliminating stockouts can be garnered by increasing fulfillment rates for cert ain types or segments of customers. In the final simulation studies we use the fi ndings from the precedi ng experiments to assess the effectiveness of using a targeted in ventory policy that uses a combination of customer traits. Specifically, we examine t he benefits of prioritizing inventory to early customers, loyal customers and baby product purchasers. Decreasing the stockout rate experienced by this subset of households to 5% increases the retailers 1 year profit by 9.7% and customer equity by 41.3%. This is 77.4% of the possibl e improvement in 1 year profit and 73.5% of the potential improvement in customer equity.16 These improvements require a 47% decrease in the total number of stockouts. We also examine the comparative import ance of stockouts in different product categories. For each of the six categories, we lower the stockout rate by half while assuming the historical stockout level for the other categories. Among the four main categories, Niche products offer the seller the highest return from improved fulfillment. A 16 The improvement in customer retention rate has a si milar pattern. If the retailer eliminates all stockouts, the retention rate increases by 6.5% compared to t he baseline case. If the retaile r decreases the stockout rate for only this subset of households to 5%, the retention rate increases by 5.5%.
40 4.8% decrease in stockouts increases 1-year profit by 1.3% and customer equity by 5.6%. Reducing stockouts in high penetration categories also yields benefits. In the case of high-usage products, the seller can in crease 1-year profit s by 3% by cutting stockouts by 19%. In the case of low-us age products, a 12.6% dec rease in the number of stockouts will increase profit by 2%. In co ntrast, improving the fulfillment in the lowusage and low-penetration categories has a minimal impact on shortand long-term profits. For the baby and pet ca tegories, the simulations s how that baby products have the highest return from improving order fu lfillment among all si x categories examined while the pet category has minimal impact. It is important to note that the count erfactual simulations involve several assumptions. First, changing order fulfillm ent policies may create reactions from the consumers that have not been captured by the existing model. Second, our analyses mostly illustrate the impact of different polic ies on the benefit side. On the cost side, we use the total number of stockouts as an approximate measure of the inventory cost/investments associated with different polic ies. It should also be noted that different customer segments may vary in the product categories they purchase. Therefore, in order to decrease the number of stockouts for different cust omer segments, the retailer may need to make different investments. In practice, companies would need more careful evaluations on the cost side to further evaluate alte rnative policies. 2.5 Discussion Our research describes an empirical investig ation of the effects of stockouts in an online retailing context. The research yields s everal managerially interesting results. First, the research suggests that stockouts may have both positive short-run effects and negative long-run effects. Second, the research provides va luable data on the long-term
41 costs of fulfillment failures. Third, the statistical results and the simulations highlight the potential benefits of using cust omer level information to differentially manage fulfillment across heterogeneous customer segments. It should be emphasized that our results are retailer specific and that we have a lim ited ability to generalize our findings. The idiosyncratic nature of our findi ngs therefore also highlights areas where future research may be useful. The analysis suggests a complex role of st ockouts. In the shor t-run we find that lower stockouts are associated with additi onal ordering and that consumers tend to increase buying in categories where a stoc kout occurred. Howe ver, cumulative stockouts have a negative effect on long-term retention. Inventory optimization models often include shortage penalties. Our work provi des insight into the true cost of these shortage penalties. For example, the customer equity calculations suggest that the average CLV of customers is about $150 when the stockout rate is about 25%. In contrast, when all stockouts are eliminat ed average CLV increases by 56%. However, this estimate only considers revenue and product costs but does not include the incremental costs needed to eliminate all stock outs. Additional research related to the assessment of marginal inventory hol ding costs would be beneficial. The simulation experiments show that there is potentially great value in prioritizing inventory based on customer tr aits. We find that an emphasis on newer customers can have substantial medium-term effects while basing inventory decisions on customers basket contents and placing an emphasis on lo yal customers can have a significant impact on long-term customer equity. We al so find that an inventory policy that discriminates in favor of newer, long-term loyal and baby product buying customers can
42 achieve significant benefits while only requiring a moderat e reduction in stockouts. Again, as our findings our specific to the retailer, theoretical research focused on developing more general findings related to product and customer types would be valuable. Our research highlights that invent ory management and cu stomer management are interrelated and that integration of t he two perspectives offers opportunities to improve both processes. The research provides insights that are relevant to CRM and Yield Management. In terms of CRM practice the empirical results and the simulations illustrate how stockouts can impact a firms customer assets and highlight the benefits of customization (K ahn et al. 2008). The research also has implications for yield management practice. The benefits of customization shown in the simulation an alyses suggest that s egment level inventory rationing can provide significant financial benefit s. Segment level inventory rationing is a key element of yield management as these systems increase revenue by reserving inventory for higher paying segments. Howeve r, while the results suggest that segment level inventory rationing can be beneficial, there are signific ant implementation challenges that the retailer would need to overcome to achieve these benefits. Yield management systems require dem and forecasting systems that predict segment level demand. Accurate predictions of future demand across segments are needed to make decisions as to whether inventory should be sold now or saved for a higher valued customer that may or may not arrive in the future. Our proposals regarding customizati on are derived from yield management principles but would be challenging and w ould require some innovations. The basic
43 requirement would be demand forecasting at the SKU level. When inventory of an ordered SKU is low, the firm will wish to know the probability that the inventory would be ordered by a more valuable segment prior to the time when the item would be restocked. While the large number of SKUs in retailing might suggest that the forecasting problems are significant, the vast majority of the firms stockouts are from a relatively small percentage of products as 1,800 of the 14,000 SKUs are responsible for about 80% of stockouts and 2,800 SKUs cause 90% of all stockouts. In comparison, a major airline may control inventor y on more than 3000 daily flights. The proposal is best illustrated with a numerical example. At the simplest level the firm might have two classes of customers that are defined based on sensitivity of lifetime value to stockouts. As an example, we will assume that the Sensitive segment has a lifetime value of $300 if no stockouts occur and $200 if the customer experiences a stockout. For the Insensitive segment we assume the future value is $250 in the absence of a stockout and $225 if a stockout occurs. If an In sensitive customer orders the final unit of a product, the fi rm will face a decision of whet her or not to fill the order. A simple decision rule for this problem w ould mimic the expected ma rginal seat revenue criteria (Belaboba 1987) commonly used in yield management. The expected marginal loss of customer equity from not filling the order woul d be $250 $225 or $25. The expected loss of cu stomer equity from filling the order would be the potential equity lost from a Sensitive customer requesting the item prior to the restocking of the product. This loss would be the probability of a Sensitive ordering the product multiplied by the marginal impact on CLV for this segment. Gi ven these assumptions, the expected loss would be equal to Pr(Sensitive Customer Order) ($300 $200). The retailer would
44 therefore fill the order if the probability of a Sensitive customer orde ring the item is less than 25%. The identification of a long-term negative e ffect of stockouts has implications for traditional yield management systems. In the ai rline and other travel industries inventory rationing decisions are based on expected marg inal revenues from a unit of inventory. These systems typically allocate inventor y to segments based on the expected revenue produced by a unit of inventory. For exam ple, the goal of yield management in the airline industry is often to maximize t he value of revenue produced by each flight (Belobaba 1989). Our analysis suggests that this is an incomplete solution since inventory rationing also impacts long-term meas ures such as customer equity (Rust et al. 2004; Blattberg and Deighton 1996). Our analyses provide insights and techniques that retailers can use to improve inventory management efficiency. For instance, the retailer under study experienced fairly high stockout rates. Our analyses sug gest ways in which the retailer can increase customer equity without much additional inventory investment. Alternatively, for retailers who carry large inventories in an effort to minimize stockout rate s, similar analysis and simulations may be conducted to help identify potential ways to significantly decrease inventory investment without sacrificing customer equity. Different retailers, depending on the current state of inventory managemen t, resources, and constraints, may benefit from our approach. In terms of feasibility, assuming that the retailer possesses a customer database and database marketing expertis e, the only additional item needed for implementation is individual-level stockout information. This information can easily be collected in online
45 environments. For example, instead of putting t he stockout information directly on the product webpage, retailers might delay when sto ckout information is revealed, e.g., by asking customers to put the product in the shopping cart to find out whether the product is in stockout, or asking customers to click the product picture to find out whether it is in stock. These methods may allow a retailer to collect stockout data without excessive costs to customers. In terms of implem enting segment level in ventory management, retailers can also customize product availabi lity information and give better customers priority in knowing where and when a product is available. This can reduce stockouts for preferred customers while not making rationing explicit. The preceding implementation issues highlight several limitations to our research. For instance, one limitation is that the analyses are conduct ed at the category or supercategory level. It would be useful to conduc t additional analysis at more refined levels and even with individual products. These type s of analyses would be particularly useful because they would allow manager s to transform stockout rates directly into inventory levels. In addition while our results are clearly retailer specific they do highlight avenues for additional research. For example, we found that stockouts of baby products result in significant damage to the value of customer a ssets. Additional research that focuses on what makes a product category more or less problematic would be useful from both an academic and practitioner standpoint. An additional issue is that the type of st ockout captured in our data is merely one form of the phenomena. A more common form of a stockout occurs when a product is simply unavailable. However, even within this class of stockout we could differentiate between instances where the product was explicitly listed as unavailable versus
46 instances where the product simply doesnt appear on the shelves or website. These types of stockouts present difficulties fo r field experiments and empirically focused analyses because retailers usually do not observe when individual customers wish to purchase an unavailable product. The types of stockouts that we examine may represent a more severe type of service failure than occasions when a product is unavailable. This distinction between types of stockouts highlights the complexity in investigating the long-term effects of st ockouts and inventory rationing. The preceding discussion highlights that there are a range of methods for controlling inventory and revealing shortage information to consumers. The retailers approach to fulfillment, where customers only l earn about a stockout at the time of delivery, may be viewed as a controversia l approach as it replaces an out-of-stock event with a fulfillment failure. The retailers approach does contain several potential advantages. First, if retailers remove icons associated with out of stock products, consumer demand is censored. This can comp licate future demand forecasting. The second issue is that explicit acknowledgement t hat an item is out of stock at the time of ordering may lead to shopping cart abandonment and perhaps long-term attrition. It is not clear whether the attrition caused by product unavailability is greater or less than the attrition caused by the type of stockout that we evaluated. Research that compares the attrition rates from different types of stockouts would therefore be useful.
47 Table 2-1. Descriptive statistics Mean Standard Deviation Order Amount $57.727 $44.422 Stockout Rate 25.4% 0.435 Fill Rate 90.3% 0.102 Cumulative Orders 3.587 4.398 Cumulative Spending 211.302 315.023 Cumulative Stockouts 0.911 1.324 Stock 1st 0.350 0.477 Baby 0.191 0.394 Pet 0.333 0.471 Table 2-2. Correlations Orders Stockouts Stockout Rate Stock 1st Pet Baby Cumulative Spending .827 .706 .055 .078 .157 .179 Orders .713 -.022 .031 .156 .104 Stockouts .459 .079 .180 .087 Stockout Rate .081 .076 .027 Stock 1st .091 -.011 Pet -.034 Table 2-3. Key variable definitions Variable Names Definitions BABY 1 if a household includes a baby, 0 otherwi se. Inferred from initialization data. PET 1 if a household includes a pet, 0 otherwise. Inferred from initialization data. TOTDISC Average category discount weighted by individual customers category activity. SH50W The shipping fee for orders lower than or equal to $50. SHSUR1 The additional shipping penalty for orders higher than $50 but lower than $75 SHSUR2 The additional shipping penalty for orders higher than $75. RECENCY Time (in weeks) since the last order. RECSQ The square term of RECENCY. FREQ Weekly ordering rate (total orders divided by weeks as a customer). LMT The log last purchase amount. AMT The log average purchase amount. CORDERS The cumulative number of transaction s the customer has made with the seller. CSTOCK The cumulative number of stockout s the customer has experienced during the PSTOCK 1 if the customer experienced a stocko ut for the previous order, and 0 otherwise. PORDSHIPThe fillrate of the customers previous order. Computed as the ratio between the dollar of shipped dollar amount and the ordered dollar amount. EARLYCUS1 if the customer has made no more than 2 orders and the time since customer acquisition is less than 6 weeks.
48 Table 2-4. Shipping fees Order Cate g or y Small orde r Medium Lar g e ( $0 to $50 ) ( $50 to $75 ) ( $75 p lus ) Structure 1 4.99 6.97 0 Structure 2 4.99 6.97 8.95 Structure 3 2.99 4.99 4.99 Structure 4 0 0 0 Table 2-5. Fit statistics Model ParametersLL BIC Single Segment With Interactions 48 -29961.6 60469.72 Without Interactions 45 -29989.3 60490.84 Aggregate Stockout Without Interactions 30 -30099.7 60540.98 Two Segments With Interactions 97 -29557.2 60218.73 Without Interactions 91 -29614.4 60264.84 Aggregate Stockout Without Interactions 61 -29846.8 60388.11 Three Segments With Interactions 146 -29502.3 60666.77 Without Interactions 137 -29511.1 60581.9 Aggregate Stockout Without Interactions 92 -29634.3 60316.05
49 Table 2-6. Estimation results Segment 1 Segment 2 CoefficientSE Coefficient SE Purchase Incidence Parameters INTERCEPT -2.888*** 0.158 -1.965*** 0.375 BABY 0.367*** 0.055 -0.01 0.124 PET 0.178*** 0.042 0.266*** 0.083 TOTDISC 0.054* 0.031 -0.012 0.063 SH50W -0.134** 0.065 0.059 0.136 SHSUR1 0.117 0.104 -0.362* 0.214 SHSUR2 -0.007 0.022 0.04 0.049 RECENCY/10 0.345* 0.207 0.496*** 0.185 RECSQ/100 -0.665*** 0.174 -0.132*** 0.037 FREQ 1.639*** 0.125 0.959** 0.412 LMT 0.009 0.029 0.044 0.05 CORDERS/10 1.566*** 0.097 0.004 0.398 CSTOCK_STAPLE 0.01 0.036 -0.29** 0.138 CSTOCK_NICHE -0.165*** 0.054 0.122 0.218 CSTOCK_VARIETY -0.031 0.04 -0.264* 0.154 CSTOCK_FILLIN -0.037 0.051 -0.318 0.222 CSTOCK_PET 0.066 0. 054 -0.481 0.363 CSTOCK_BABY -0.122* 0.067 -0.633** 0.302 PSTOCK_STAPLE 0.031 0.074 0.232 0.165 PSTOCK_NICHE 0.24** 0.115 -0.29 0.266 PSTOCK_VARIETY 0.176** 0.081 0.093 0.176 PSTOCK_FILLIN 0.006 0.103 0.212 0.25 PSTOCK_PET -0.153 0.132 0.178 0.366 PSTOCK_BABY 0.328** 0.149 0.131 0.395 CSTOCKEARLYCUST -0.417** 0.199 0.118 0.514 PSTOCKEARLYCUST -0.187* 0.106 0.218 0.212
50 Table 2-6 Continued Segment 1 Segment 2 CoefficientSE Coefficient SE Purchase Amount Parameters INTERCEPT 2.396*** 0.468 2.241*** 0.817 BABY 0.035** 0.017 0.069 0.05 PET -0.005 0.014 0.056 0.037 TOTDISC -0.02** 0.01 -0.017 0.027 SH50W -0.024 0.019 0.11** 0.051 SHSUR1 0.036 0.031 -0.155* 0.084 SHSUR2 -0.003 0.006 -0.041** 0.018 RECENCY/10 0.87*** 0.075 0.327*** 0.067 RECSQ/100 -0.446*** 0.061 -0.059*** 0.015 FREQ 0.201*** 0.032 -0.022 0.135 AMT 0.333*** 0.12 0.414* 0.232 CORDERS/10 0.05** 0.021 -0.168 0.135 CSTOCK_STAPLE -0.002 0.009 0.089** 0.045 CSTOCK_NICHE 0.03 ** 0.012 0.131* 0.072 CSTOCK_VARIETY -0.03*** 0.01 -0.041 0.049 CSTOCK_FILLIN 0.004 0.013 -0.061 0.065 CSTOCK_PET -0.035** 0.014 -0.021 0.099 CSTOCK_BABY -0.035** 0.016 -0.191* 0.11 PORDSHIPRAT -2.3*** 0.475 -0.692 0.756 PORDSHIPRATAMT 0.552*** 0.122 0.067 0.238 Scale ( ) 4.344*** 0.111 3.146*** 0.147 Correlation ( ) -2.711*** 0.253 -2.992** 1.142 Segment size (w) 1.406*** 0.28 ***0.01 p ; **0.05 p ; *0.1 p Note: We estimated the segment size par ameter using a logit formulation such that1exp()/1exp() ww.
51 Table 2-7. Segment descriptions Mean Standard Deviation Overall Total Orders 3.587 4.398 Average Order Amount 55.585 37.494 Total Stockouts 0.911 1.324 Retention Rate 55.6% Retention Rate of High-Stock out Rate customers 51.5% Retention Rate of Low-Stock out Rate Customers 57.8% Segment 1 Total Orders 3.724 5.020 Average Order Amount 56.845 39.029 Total Stockouts 0.97 1.473 Retention Rate 56.3% Retention Rate of High-Stock out Rate Customers 51.4% Retention Rate of Low-Stock out Rate Customers 58.9% Segment 2 Total Orders 3.281 2.494 Average Order Amount 52.789 33.695 Total Stockouts 0.779 0.895 Retention Rate 54.2% Retention Rate of High-Stock out Rate Customers 51.7% Retention Rate of Low-Stock out Rate Customers 55.4% Table 2-8. Holdout sample validation Actual Predicted Predicted (2-segment model w/o stockout measures) Mean Orders 3.458 3.552 5.019 Mean Amount $ 56.723 $ 54.831 $63.91 MAD (Purchase Incidence) 0.135 0.161 Hit rate (Overall) 85.0% 79.8% Hit rate (Purchase) 19.1% 18.6% Hit rate (Nonpurchase) 91.1% 86.4% MAD (Purchase Amount) $ 20.051 $ 23.992
52 Table 2-9. Dynamic validation Estimation Sample Holdout Sample Actual Simulation Actual Simulation Mean Orders 3.587 3.576 3.458 3.568 Mean Amount $ 55.585 $ 54.06 $ 56.723 $ 54.163 Mean Stockouts 0.911 0.908 0.894 0.903 Retention Rate 55.6% 56.0% 55.4% 56.0% Retention Rate of HighStockout Rate Customers 51.5% 51.7% 51.2% 50.6% Retention Rate of LowStockout Rate Customers 57.8% 58.2% 57.9% 58.8% Table 2-10. Repeat purchase model Aggregat e Stockout Categorical Stockout Coefficie nt SE Coefficient SE INTERCEPT -1.39*** 0.188 INTERCEPT -1.401*** 0.188 BABY 0.77*** 0.093 BABY 0.775*** 0.095 PET 0.925*** 0.064 PET 0.948*** 0.066 TOTDISC 0.196*** 0.049 TOTDISC 0.192*** 0.051 SH50w -0.074* 0.042 SH50w -0.073* 0.042 SHSUR2 0.006 0.012 SHSUR2 0.007 0.012 AMT1ST 0.254*** 0.038 AMT1ST 0.251*** 0.038 STOCK1ST -0.037 0.06 STOCK1ST_HUHP -0.14* 0.078 STOCK1ST_HULP -0.057 0.155 STOCK1ST_LUHP 0.198** 0.091 STOCK1ST_LULP -0.107 0.133 STOCK1ST_PET -0.264 0.187 STOCK1ST_BABY -0.046 0.19 LL -3873.9 -3869.4 Observations 6164
53 Table 2-11. Simulation results Scenario 1 Year Revenue 1 Year Contribution Mean Orders Mean Stockouts Total Stockouts (Categorical) Retention Rate Customer Equity Baseline $466,733 $75,659 3.861 0.980 2325 54.8% $341,938 Lower Stockouts 20% $477,251 $77,477 3.930 0.812 1852 56.2% $376,285 13% $506,286 $82,322 4.118 0.524 1218 59.6% $468,518 5% $519,501 $84,577 4.197 0.203 468 60.7% $517,799 0% $521,635 $85,146 4.245 0.000 0 61.3% $533,972 Lifecycle Targeting Early Customers 5% $506,056 $82, 025 4.106 0.808 1931 60.0% $396,284 Loyal Customers 5% $481,799 $78, 711 3.960 0.780 1794 55.7% $422,828 Basket Contents Pet Products 5% $488,230 $79,243 3.955 0.662 1384 58.4% $410,223 Baby Products 5% $496,363 $80, 880 4.076 0.823 1848 56.2% $434,367 Latent Segments Segment 1 5% $506,262 $82,226 4.117 0.411 965 59.3% $476,451 Segment 2 5% $479,679 $78,054 3. 935 0.778 1774 56.1% $386,132 Combination Early, Loyal & Baby 5% $510,123 $83,007 4.151 0.538 1233 60.3% $483,113 Early, Loyal & Baby 10% $503,040 $81,147 4.072 0.651 1474 58.6% $445,215 Lower Stockouts across Categories HUHP $480,419 $77,935 3.960 1.018 1893 56.5% $381,074 HULP $473,411 $76,639 3.914 1.019 2214 55.9% $361,014 LUHP $475,793 $77,177 3.915 1.013 2032 55.9% $370,437 LULP $466,931 $75,857 3.862 0.974 2063 54.8% $344,663 PET $468,395 $75,939 3.881 1.013 2195 55.0% $351,973 BABY $471,043 $76,222 3.913 1.020 2292 56.1% $353,674
54 $75,000 $77,500 $80,000 $82,500 $85,000 0%5%10%15%20%25% Stockout Rate1 Year Profit Figure 2-1. Year 1 profit versus stockout rate $336,000 $386,000 $436,000 $486,000 $536,000 $586,000 0%5%10%15%20%25% Stockout RateCustomer Equity Figure 2-2. Customer equity versus stockout rate
55 CHAPTER 3 REWARDING SOME BUT PUNISHING OT HERS? AN EMPIRICAL STUDY OF FREQUENT SHOPPING DISCOUNTS Frequent buyer and loyalty programs are increasingly common marketing instruments. These programs provide a means to customize marketing based on customer characteristics. An underappreciat ed consequence of these programs is that while some customers benefit ot hers experience a form of pric e discrimination. In this Chapter, I conduct an empirical study to examine the response of members and nonmembers to a grocery retailers frequent buyer program. 3.1 Introduction The promise of Customer Relatio nship Management (CRM) is that firms will be able to understand individual customers and develop customized marketing programs that maximize the long-term value of each customer asset. However, while customization of marketing policies based on i ndividual customer traits is intuitively appealing, the act of custom ization can also be viewed as discrimination across customers. For instance, the growing us e of frequent shopper and loyalty programs allows firms to increasingly customize the marketing envir onment for preferred segments of customers. For example, it has become commonplace for grocery retailers to limit many promotions to frequent shopper ca rd holders. This type of customization or discrimination may also have adverse c onsequences on consumers that are not members of the loyalty program In this paper we empirically investigate the impact of loyalty program-restricted sales promotions on both frequent shopper program participants and non-participants. Frequent buyer cards and loyalty progr ams have become common marketing instruments. Research by the Food Marketing Institute indicates that about half of food
56 retailers offer loyalty programs (FMI 2005). Furthermore, loyalty programs have been widely accepted by consumers as survey data reveals that 86% of adults have at least one grocery store loyalty card and 76% report that they use their cards each time they shop (Boston University 2004). These progr ams often operate by providing discounts only to customers that have the loyalty card. In return for these discounts, participants implicitly allow the retailer to track their individual purchases. Loyalty programs provide retailers with capabilities that may yield benefits. First, they provide a means to track the retaile rs most valuable customers. It has been estimated that while only 12% to 15% of cu stomers are loyal to a single retailer, these loyal customers can generate 55% to 70% of total sales (FMI 2006). In addition to tracking high value customers, loyalty progr ams can be used to restrict benefits to a firms most valuable customer assets. Fina lly, loyalty programs also provide detailed longitudinal data that can be used to re fine assortment plann ing, merchandising and other marketing tactics. Despite these pot ential benefits there remains a debate as to the advisability and success of these programs. In contrast to the FMI study noted above, an earlier FMI study indi cated that 61% of food reta ilers had or planned to have a frequent shopper program (The Commercial Appeal, May 17, 1998). The downward trend or limited spread of these programs suggests that these programs may be problematic in some respects. For instance, the evidence regarding the profitability of these programs is ambiguous as the academic literature has reported mixed findings (Dowling and Uncles 1997; Dreze and Hoch 1998; Lal and Bell 2003; Lewis 2004). However, it should be noted that the academic literature studying consumer response to loyalty programs (Liu 2007; Kivetz 2002; Nunes and Dreze 2006) has only
57 focused on how loyalty programs change the behavior of participants. A gap in the literature concerns the impact of the pr ograms on non-participants. To best implement these programs, merchants need to underst and the impact on both consumers that garner benefits and consumers that do not. For in stance, while a majority of sales may come from program members, a sizable portion of sales is likely to come from nonparticipants. In our data, we find that 30% of trips are made by non cardholders and that these trips generate 20% of t he firms total revenues. For non -members promotions that are restricted to members may be viewed as di scriminatory or unfair. There is a growing literature focused on the topic of price fairness. This liter ature suggests that unfairness perceptions are likely to occur when sellers customize prices across similar customers (Haws and Bearden 2006). Howeve r, there is a dearth of studies that examine how unfairness impacts actual purchasing behaviors. The primary purpose of our paper is to investigate how a frequent buyer program impacts the behavior of non-cardholders relative to program members. Our investigation is empirical in nature as we study res ponse to promotions by members and nonmembers of a food retailers loyalty card pr ogram. This empirical context affords the opportunity assess program response along a variety of dimensions. We begin with analyses related to overall customer behavior. Specifically we investigate how the prevalence of loyalty card program influenc es store traffic and average basket size for card holders and non-card holders. We then decompose the high-level results through analyses at the brand and category level. Our research yields multiple contributions to the literature. First, while program members account for the majo rity of a retailers rev enues, nonmembers often account
58 for a significant portion of revenues. Giv en the small margins involved in grocery retailing this nonmember revenue may be a critic al driver of profitability. Understanding the response of these consumer s is therefore a critical m anagerial topic. The research also provides real world empirical data on response to price discrimination. While previous laboratory based studies provi de useful theory regar ding psychological response to price discrimination (Haws and Bearden 2006; Bolton et al. 2003), the analysis of actual behavior in the marketplac e is a vital undertaking and the field results complement the lab based findings. For exam ple, the multi-categor y nature of modern retailing means that price discrimination c an have complex and multidimensional effects that are not easily studied in labor atory settings. The remainder of the paper is organized as follows: In the next section we review literature relevant to our research objectives. Specifically we focus on literature related to price discrimination, price fairness and customer relationship management. We then describe the data and present descriptive statistics. The analysis is then conducted in stages. We begin with analyses of store traffic and basket size effects. We then conduct category and brand choice level analyses that illuminate the drivers of the store-level effects. We then conclude the paper with a discussion of managerial implications, study limitations and avenues for future research. 3.2 Background In this secti on we consider several str eams of literature that are relevant to understanding consumer response to price di scrimination. We begin with a review the literature focused on price fairness. Within th is literature we focus both on the factors that drive perceptions of unfairness and on the psychological responses to price discrimination. We then propose a conceptual framework that relates price
59 discrimination to customer purchasing in a retailing context involving multiple categories. In many marketing contexts it is possible for firms to increase profits by employing non-uniform pricing. Non-uniform pricing, or price discrimi nation, is roughly defined as instances where firms sell identical goods or services at different prices. While the application of price di scrimination can increase firm profit s, the practice of charging different customers different prices may be controversial. The decision to employ price discrimination techniques ther efore involves both benefits in terms of profits but can also have negative consequences depending on how consumers re act to unfairness. The marketing literature has begun to pay a significant amount of attention to the topic of price fairness (Xia et al. 2004). T he foundation for much price fairness research has been the principle of dual entitlem ent proposed by Kahneman et al. (1986). Kahneman et al. (1986) use a wi de range of scenarios to identif y the relative perceived fairness of different practices. This research finds that it is a cceptable for firms to increase prices when costs increas e, it is unfair for firms to exploit market power, and it is reasonable for firms to reap additional profits when costs decrease. The general principle is that consumers feel entitled to a reference transaction and that firms are entitled to some reference profit. The i dea of equitable transactions is also a key element in Thalers (1985) work on transaction utility. Thalers work highlights the role of contextual factors in transaction evaluation. Collectively, the key result from this research is that fairness judgments are based on comparis ons of an offered price to some reference price.
60 Other researchers have expounded on the issue of reference transactions. Campbell (1999) examines how inferred motive and firm reputation can moderate reaction to deviations from the reference tr ansaction. Bolton et al. (2003) investigate how factors (competitor prices, vendor co sts) beyond the current transaction can influence reference points. Bolton et al. also study dynamic elements such as previous prices paid by a consumer. However, relati vely little attention has been paid to the impact of differential pricing across consum ers (Feinberg et al. 2002). Recent work by Haws and Bearden (2006) focuses on how pr ice discrimination across customers impacts fairness perceptions. Haws and Be arden find that price differences across consumers results in very str ong perceptions of unfairness. Xia, Monroe and Cox (2004) pr ovide a review of the existing literature and a conceptual framework for price fairness percept ions. The review is largely conducted in the context of the principl es of distributed justice (Homans 1961) and equity theory (Adams 1965). These theories also emphasiz e the role of comparisons across individuals. Xia et al. conclude that unfai rness perceptions are influenced by several factors. First, consumers are more likely to view prices as unfair when transaction similarity is high. Second, providing r easons for price differences can mitigate unfairness perceptions. Xia et al. also note the importance of consumers previous experience and general knowledge in assessing fairness. The literature also suggests that perceptions of unfairn ess will have psychological effects and will impact behavior. In terms of psychological effects, unfair prices have been found to reduce satisfaction (Campbell 1999; Darke and Dahl 2003), create negative emotions (Xia et al. 2004) and in crease price consciousness (Sinha and Batra
61 1999). In terms of behavioral intentions, res earchers have found that unfairness leads to reduced purchase intentions (Oliver and Swan 1989) and a tendency to leave relationships (Urbany et al. 1989). 3.3 Conceptual Framework We now shift to considering how unfair price environments may influence consumer decision making. We use Figure 1 as a description of a representative consumers shopping process. Our empiri cal context is a grocery store where customers typically assemble baskets of items across multiple categories. This figure is stylized to fit the type of sequential decision making that occurs in retail environments where consumers purchase collections of products rather than single items. The goal of the figure is to replicate the structure and dynamics of consumer decision making during a grocer y shopping trip. Specifically, the figure is concerned with how consumers mental states and subsequ ent behaviors towards a retailer and various brands are influenced by the presence of rest ricted promotions. The shopping process in Figure 1 proceeds as follows. We assume customer status in the program is determined outside of t he current shopping trip.17 Customers either possess the rewards card (Card Customers) or do not (No Card or Non-Members). Customers decisions regarding whether to shop at t he retailer are influenced by consumers expectations regarding the promot ions available each week. At the onset of shopping, customers ar e assumed to begin in some unobserved mental state. This state is likely multidimensional and may vary systematically between 17 An interesting element of this mechanism is that all customers have the ability to become program members. This may be salient as consumers have c ontrol over whether they will have access to the restricted promotions.
62 members and non-members. Th e framework assumes that consumers construct their shopping baskets through a repeated two stage purchasing process. The first stage involves the choice of a category to cons ider. Following category choice the consumer then evaluates the category envir onment in terms of the type of promotions in place, the breadth or number of promotions and the dept h of the discounts av ailable. In our context, promotion type indicates whether the pr omotion is restricted to card holders or whether the promotional price is available to all customer s. The second stage of the process involves consumer choice in terms of selection of items. For this choice, consumers decide whether or not to buy in the category and what brand, if any, to select. While the literature and anecdotal data support the notion that price discrimination across customers will often be viewed as unfai r, a limitation of existing research is that the focus has tended to be on single products. In practice, targeted promotions are often delivered in reta il environments that possess a degree of complexity that cant be recreated in the labor atory. Many retailers offer a vast array of products and customers typically purchase many items on each visit. This is salient as instances of price discriminat ion occur at the level of in dividual brands while consumers are typically shopping across multiple catego ries. It is possible that category level unfairness can impact consumer decision ma king beyond the specific category. If the retailers pricing practices ar e viewed as unfair (or generous) then consumers mental states may evolve ov er the course of a shopping trip. In our framework we include a feedback loop that considers the po ssibility that consumers mental states are based on exposure to diff ering price environments. Specifically, the
63 consumers mental state changes as cat egories are evaluated and these changes may impact consumers decisions in subsequent categories. The category evaluation process repeats with consumers evaluating additional categories until the customer decides to end the process and complete the transaction. This process, while a simplified versi on of reality, is intended to provide a framework for considering how reward program based promotions can influence consumers both at the categor y level and in terms of overall purchasing activity. An important point is that the construction of the overall basket includes dynamic aspects and that there may be cross category effects. We next review ex isting theory to speculate about how card-restricted promoti ons will impact the shopping decisions of program members and non-members. Store traffic. The first stage of the conceptual fr amework involves a store choice decision. Due to a lack of competitive info rmation we view this as a binary incidence decision. Previous empirical work on store traffic has emphasized response to promotions (Bell and Lattin 1998). For inst ance, Walters and MacKenzie (1988) and Lam et al. (2001) investigate the role of weekly promoti onal activity. Following this literature on promotions and st ore traffic we expect that member traffic will increase when there are more card-restricted promotions. This prediction is not explicitly related to our structural model of decision making as it is based on an assumption that cardmembers use the retailers weekly advertising materials in their store choice decisions. For the non-card holders we hypothesize the reverse as previous research suggests that customers will seek to av oid the negative consequences of price unfairness. Oliver and Swan (1989) report that unfairness per ceptions result in lower
64 purchase intentions and Xia et al. (2004) co njecture that consumers will react in ways that protect their financial interests such as leaving the relationship. Furthermore, Urbany et al. (1989) found that the greater the degree of unfairness the higher the propensity for consumers to leave a relation ship. This last point about the degree of unfairness highlights an important empiri cal issue. In our empirical analyses we measure price discrimination in two ways. We use the number of card-restricted promotions as an indication of the breadth of unfairness and the dept h of card-restricted promotions as a measure of the degree of unfairness. While we expect the following predictions to hold for both discrimination breadth and depth, it is not clear what the relative effects of discount depth and breadth consumers will be on fairness perceptions. It is an empirical question w hether more widespread restricted promotions are more or less unfair than deeper restri cted promotions. Based on the preceding discussion we propose the following hypotheses. In terms of notation, hypothesis H1_M is our prediction for members or card-holders and H1_N is our prediction regarding the behavior of non-cardholders. H1_M: There will be a positive relationshi p between card-only promotions and store traffic / purchase incidence by LP members. H1_N: There will be a negative relationship between card-only promotions and store traffic / purchase incidence by non-members. While these hypotheses are straight-forwa rd, there are a number of factors that may moderate the impact of price discrimination. Xia et al (2004) stress the importance of social norms and past experience. An in teresting element of price discrimination practice is that some practices such as student discounts are viewed as acceptable while others such as gender based discounts (Stevens 1996) can be controversial. For example, yield management techniques employ ed in the airline and travel industries
65 involve significant variation in prices ac ross consumers. However, these techniques have largely become accepted by consumers (Kimes 1994). The prevalence of frequent buyer card programs may ther efore moderate unfairness perc eptions if these programs are simply viewed as part of standard retailing practice. Finally, if non-card holders tend to be less involved consumers we might also expect that they would pay less attention to the retailers advertising. If non-members tend not to ev aluate the retailers weekly circular we would expect that the impact of restricted promotions on store traffic would be mitigated. Basket formation. In this subsection we consi der category level decisions. We begin by considering how promotions may impact brand choice and category incidence at the level of a specific category being evaluated. We then consider dynamic and cross-category effects. Brand choice The typical card-restricted promotion involves discounts on a specific brand for the period of one week For program members the predictions regarding brand choice are straight-forward as card-restricted and non-restricted promotions each provide economic incentiv es for purchases of the promoted brands. The extensive promotions lit erature (see Neslin 2004) suggests that card members will respond positively to both restricted and non -restricted promotions. Furthermore, we expect that deeper discounts will increase brand choice. For non-members the existence of restricted promotions does not change the prices faced by the consumer. A simplistic argument is that sinc e there is no change in the price environment there s hould be no changes in behavior. However, the availability of a discount to a segment of customers may change the non-members reference price
66 for a brand. The degree to which the promotional price is viewed as the reference price for the brand is a function of the similarity of the transactions for members and nonmembers (Xia et al. 2004). Transaction similarity may be judged in terms of contextual factors such as terms of sale or time and also in terms of similarity between consumers. Social comparison theory (Wo od 1989) suggests that similar others will be the most salient comparison. When similarity is high unfairness perceptions are expected to be large. In the context of grocery shopping, the transactions and customer traits of loyalty program members and nonm embers are likely to be perceived as very similar. This suggests that card-restricted discounts may be viewed as highly unfair. The immediate consequences of unfairness percept ions are likely to be reduced satisfaction (Darke and Dahl 2003; Campbell 1999), lower purchase intentions (Oliver and Swan 1989) and negative emotions. For the non-ca rdholders we therefore expect that shoppers will avoid brands on card-restricted promotion. H2_M: Card-only promotions will have a positive effect on brand choice of program members. H2_N: Non-members will avoid brands on card only promotions. Category incidence. The discussion of brand choice suggests that card-holders will be attracted to brands on card-only pr omotion and that non card-holders will avoid these brands. At the level of category incidence the predicti ons are less straightforward. For card-holders our initial conjecture is th at card-only promotions will have a positive impact on category incidence. However, this prediction is complicated by the multicategory nature of our retailer. While promotions increase the attractiveness of purchases in a category, an addi tional factor is that promot ions in other categories can
67 alter a categorys relative attractiveness. On balance our conjecture is that the effect on members will be positive. For non-members the impact of card-only promotions on category incidence is also difficult to predict. For nonmembers t he decision to avoid a promoted brand may be just one of several elements of the consumers category dec ision. The question is how the desire to avoid perceived penalties will trans late to buying behavior. A mild form of avoidance might entail simple switches away from promoted brands to other brands in the category. Alternatively, a more severe form of avoidance might involve switching to other categories. Bougie et al. (2003) f ound that dissatisfaction evokes thoughts about what buyers are missing and sug gests that buyers will attempt to devote attention to something else. In grocery shopping it is difficu lt to predict how this type of tactic may occur. Consumers may switch to non-prom oted brands where everyone pays the same price, switch to products in substitu te categories or reduce total buying. H3_M: Card-only promotions will incr ease category incidence by card-holders. H3_N: Card-only promotions will increase or decrease category incidence by non-card holders. Grocery shopping is an inherently dynamic process since customers typically make sequential selections across a variety of categories. Our conceptual framework therefore includes a feedback loop whereby consumers choose categories, make product selections and then r epeat until the transaction is completed. The second box in the framework is labeled Customer Emotional State. Our assumption is that this emotional state may impact category level deci sions and may evolve in response to the promotions encountered during the shopping occasion. Specifically, we consider how
68 the nature of a category may interact with price discrimi nation and how price unfairness may impact the number of ca tegories a customer shops. Expenditure levels. The relationship between prom otional activities and basket size is somewhat ambiguous. For the general ca se of promotions that are available to all customers, discounts can have both positive and negative effects on total expenditures. Promotions may increase basket si ze if they motivate customers to make incremental purchases while promotional discounts make reduce total expenditures simply because they reduce prices. Beyond th is basic economic issue, the use of cardrestricted promotions may also impact consumer moods in a manner that affects basket size. The framework involves a feedback process whereby by purchases and promotions in each category may impact the customers emot ional state. One implication of this feedback is that the number and depth of restricted promotions may influence the numbers of ca tegories that are shopped. This may occur because consumers moods may evolve during the s hopping trip as the consumer sequentially evaluates different categories. For card-holders, greater card-restricted prom otional activity may result in positive emotions. This may lead to larger purchases as positive feelings may increase purchase intentions (Brown et al. 1998). However, the degree to which restricted promotions actually improve mood is an open empirical question. For non-card holders the situation is perhaps more complex. Card -restricted promotions may create negative emotions that may influence consumer behavior. However, we do not have a clear prediction as to how consumers will res pond. Bad moods have been found to increase
69 impulsive consumption (Tice et al. 2001) but unfairness may also reduce purchase intentions (Xia et al. 2004). An additional consequence of price unfairness may be increasing price consciousness (Sinha and Batra 1999). Sinha and Batra find that when National brands prices are perceived as unfair that consum ers become more price conscious and more likely to select private labels. An analogous situation may exist if brands on cardrestricted promotion are viewed as unfair. This may result in the victims of unfairness exhibiting greater price sensitivity. T hese conflicting arguments preclude the development of clear hypot heses regarding the impact of restricted promotions on expenditure levels H4_M: Greater card promotion breadth will result in larger or smaller expenditures by card holders. H4_N: Greater card promotion breadth will result in larger or smaller expenditures by non-card holders. The preceding predictions relate to the breadth of card-restricted promotions. For discount depth our prediction is similar fo r card-holders. Agai n the prediction is ambiguous because discounts may simultaneously increase purchasing while reducing revenue. For nonmembers we predict that promotion depth wil l negatively impact purchasing. Our conjecture is that as discounts become large customers will view the discrimination as unaccept able (Urbany et al. 1989). H5_M: Greater card promotion depth will resu lt in larger or smaller expenditures by card holders. H5_N: Greater card promotion depth will result in smaller expenditures by noncard holders. Category type When price discrimination occurs at a retailer that sells many brands and categories the consequences of price unfairness ma y have multifaceted
70 effects. While unfairness is expected to l ead customers to avoid discrimination, in a retailer selling many categories and brands it is not clear how these avoidance or coping mechanisms will manifest. It has been not ed that emotions can often motivate consumers to engage in behaviors that atte mpt to alter their mood (Andrade 2005; Bagozzi et al. 1999). One possible consequence of negative consumer emotions is that the consumer may seek out products that im prove mood. This is relevant because the individual products (and categories) in a reta ilers assortment may vary in terms of whether they are primarily hedon ic or utilitarian in nature. The degree to which a product offers hedonic or utilitarian benefits has been shown to alter consumer decision making (Dhar and Wertenbroch 2000; Batra and Ahtola 1990; Mano and Oliver 1993). Products that offer hedonic benefits like ice cream pr ovide more emotional value to consumers (Sloot et al. 2005) while the buying process for utilitarian products may be more rational in nature. As noted, consumers have been shown to reac t to unfairness via different types of coping mechanisms. Given that hedonic products provide attributes like fun, excitement and pleasure (Batra and Ahtola 1991; Dhar and Wertenbroch 2000) purchase of these products may be used as a means of coping. Gi ven that different products can provide different levels of emotional benefits cons umers may make choice s in an effort to regulate their mood (Andrade 2005). This conj ecture raises a question about whether the adverse consequences of price discrimi nation will vary across product types. We propose that the consequences of price discrimination will be mitigated for more hedonic products. H6_M: The positive consequences of pr omotions on members will not increase for more hedonic products.
71 H6_N: The negative impact of card-restri cted promotions on non-cardholders will be less for more hedonic products. Summary. In general the predictions for cardholders are straight forward. Cardrestricted promotions are predi cted to increase store traffic, brand choice and category incidence. For overall ex penditures we lack a clear prediction because additional purchases motivated by the promotions must be traded off against the discounts offered. For the non-card holders, while exis ting theory offers the means to speculate about likely behavior, we have several questions that are best informed by data. In particular, we are interested in how noncard holders coping behaviors may impact purchasing behavior. These coping behaviors can possibly have both negative and positive consequences for the retailer. We summa rize our hypotheses in Table 3-3. 3.4 Analyses In this secti on we describe key aspects of the firms business environment, provide descriptive statistics related to customer behavior and the firms marketing and present our analyses. The firm is a retailer that operates a small chain of grocery stores. As is common in the grocery industry the firm opera tes a frequent buyer program that allows the firm to offer promotions that are restricted to frequent shopper program cardholders. The program is open to all customers, but customers do need to sign-up for the program to obtain a card. The sign-up proc edure requires basic data such as name, phone number and address. The cust omer level data contains complete records of all items purchased. The data thereby provides opportunities to study the relationship between frequent shopper discounts and severa l different measures of consumer activity such as store traffic, brand choice, category incidence and trip level expenditures.
72 Table 3-1 presents descriptive statistics related to the behaviors of cardholders and non-cardholders. On average the firm rece ived 1435 daily visits. Of these 68.4 % were by members and 31.6% were by noncardholders. In terms of spending, the average expenditure by program member s is $25.56 compared to $14.28 for nonmembers. Overall, 80% of revenues are derived from cardholders. Table 3-2 describes the promotional practices of the firm during the data co llection period. The firm offers three types of prom otions. The first promotion is available only to cardholders (CARD). These promotions tend to offer the greatest savings and are advertised in the weekly circular. The second promotion is w eekly specials (SPEC). These are advertised specials that are available to all customers. The third type of promotion is designated as BUYLOW deals. These are unadvertised, available to all and offer the smallest discounts. The complexity of the retail environment presents several analysis challenges. In particular, the large number of SKUs make s it necessary to us e summarized variables to reflect the marketing environment. Additionally, the fa ct that customers purchase baskets of products rather t han single items means that there is no single central dependent variable. Our analysis strategy is, t herefore, to begin with fairly aggregate level analyses that speak to the overall impact of fr equent buyer programs. We then conduct analyses that highlight the factors that drive the aggregate level results. Specifically we begin with analyses focus ed on store traffic and average basket size and then present analyses focused on brand choice and category incidence. Store level analyses. We begin by analyzing measures related to overall response. These high-level measures are critic al as they are the most salient to the
73 retailer and speak to the overall impact of card-restricted promotions. Given the complexity of the retail environment thes e analyses are conducted with aggregate level covariates. For the analysis we examine store traffic levels and daily average basket size simultaneously. We jointly model store traffic and two basket size metrics using a Seemingly Unrelated Regressi on (SUR) model to account fo r correlation between the measures. We use equations for store tr affic, average expenditures and average number of items purchased. The equations are given below: 123 4 5 6 789 1 0 11 12 13 14 15 1 it i t t t t itittt ttttti tSTMEMNUMCARDNUMSPECDISCARDDISSPEC MEMNUMCARDMEMDISCARDMONTUE WEDTHUFRISUNHOL (3-1) 123 4 5 6 789 1 0 11 12 13 14 15 16 17 2 it i t t t t ttitit ttttttti tBSMEMNUMCARDNUMSPECNUMLOWDISCARD D ISSPECDISLOWMEMNUMCARDMEMDISCARD MONTUEWEDTHUFRISUNHOL (3-2) 123 4 5 6 789 1 0 11 12 13 14 15 16 17 3 it i t t t t ttitit ttttttti tItemsMEMNUMCARDNUMSPECNUMLOWDISCARD DISSPECDISLOWMEMNUMCARDMEMDISCARD MONTUEWEDTHUFRISUNHOL (3-3) The variables in these expressions are defined as follows: itST is the store traffic from customer groupi(1,2 i )on day t Customer group1 i are card holders and customer group2 i are non-card holders (as captured by identity variable1iMEM if 1 i and 0 otherwise). it B S is the average basket size in dollars of customer groupi on day t Itemsit is the average number of items in baskets purchased by customer group i on day t For each type of promotion, we in clude two covariates. For example, tNUMCARD is the number of items in the store offered with card-only promotions on day t and tDISCARD is the average discount of these deals. The former captures the
74 breadth of price discrim ination and the latter captures t he depth of price discrimination. We account for differences in product popularity by weighting products based on penetration rate (i.e., the percentage of baske ts involved a purchase of the item). Our equation includes interactions that are of interest. The interactions between customer type (1iMEM ) and the card-restricted promotion measures (i.e., it M EMNUMCARD andit M EMDISCARD ) allow for measurement of the effects of price discrimination for the two types of cust omers. The specification also includes day of week (e.g., MON, TUE) and holiday indi cators (HOL). Tabl e 3-4 summarizes the variable definitions. The estimation results of the store-level SUR model are shown in Table 3-5. The results show that: (1) traffi c from card holders is generally higher than non-cardholders; (2) average basket size of cardholders is sign ificantly larger than non-cardholders; (3) the number of specials has a more significant effect on store traffi c than the magnitude of specials; (4) the number of advertised card-restricted promotions decreases the traffic of non-cardholders but the effect is not significant; (5) the number of advertised card-restricted promotions increases the traffi c of cardholders but the effect is about the same as advertised specials that apply to everyone; (6) for non cardholders the depth of card-restricted discounts negatively impacts basket size while the number of cardrestricted promotions has a positive im pact on basket size; (7) for cardholders advertised specials have a positive effect on basket size while unadvertised and cardrestricted promotions dont significantly in crease the basket size. The most intriguing result from the store level anal yses is that increasing num bers of card-only promotions lead to larger baskets for non -members. This result is c ounterintuitive and highlights the
75 need for more detailed analysis because t he positive relationship between cardrestricted promotions and nonmember basket size may be driven by multiple behaviors. Brand level analysis (market share analysis). For the second stage of the analysis we investigate how card-restrict ed promotions impact market share among program members and nonmembers. For ca rdholders we expect that restricted promotions will drive signific ant increases in market share. For nonmembers our expectation is that restricted promotions will result in lower shares. For nonmembers, this speaks directly to the consequences of price discrimination as the analysis deals with response to price differences across consumers rather than absolute prices. We conduct the analysis across a variety of categor ies to investigate how card-restricted promotions act across different product types. At the brand level, we estimate a MNL brand share models for multiple categories (c1,2, .,C). Since the retailer defines seve ral hundred categories we limit the categories under study to products with relatively high penetration rates and with well defined brands (i.e. we do not include fresh product categories such as meat and produce). We do include both food categories and non-food categories. Descriptive statistics for the categories under study are gi ven in Table 3-6. The market share model is specified below: 123 4 56 789 1 0 3 1exp( ) 1,2, 1,2, .,, c1,2, .cikct ccic kctckctckctci kct ci kctci kctckctckctikct ikct ikct c K iwct w A T MEMLCARDSPECLOWMEMLCARD MEMSPECMEMLOWDISCREGP AT SH ikK AT .,, t=1,2, .,T C (3-4) In terms of notation,ikctAT is the attraction of brandkfor customer groupiin category c in period t, andikctSH is the share of brandkfor customer groupiin category c
76 in period t. For covariates, we include three binary variables including whether there is a card-restricted discount offered for the brand (kctLCARD ), whether there is an advertised special offered for the brand (kctSPEC ), and whether there is a in-store special offered for the brand (kctLOW ). The specification also includ es the regular price of the brand (kct R EGP ), and if there is a discount, the monetary value of the discount (kctDISC ). Interactions are included to account for differ ences in how the three types of promotions influence the brand choices of cardholders and non-cardholders. The estimation results are summarized in Table 3-7. For non cardholders, market shares usually decrease when card-restrict ed promotions occur (the average of the coefficients of kctLCARD is -0.63). For card-holders rest ricted promotions usually result in significant increases in market s hare (the average of t he coefficients of ik c t M EMLCARD is 3.14). For non-cardholders the tendency to avoid price discrimination is not uniform across categories. To investigate the observed differences, we computed the correlations between the coefficient ofkcLCARD and category characteristics such as category size, conc entration level, and whether the category is hedonic in nature. A salient finding from th is analysis is that the coefficient of kcLCARD is positively correlated with hedonic level (0.349, p_value=0.009 ). This means noncardholders are less likely to switch away from the card-restricted promotions in hedonic product categories. Category level analyses. The preceding analysis suggests that restricted promotions impact market share among for members and nonmembers. For nonmembers, while the market share shifts vary across category types, the tendency is
77 for nonmembers to avoid the products on card -restricted promotion. Given the finding that nonmembers tend to increase basket si ze when card-restricted promotions are more prevalent, an obvious issue is that source of the in cremental demand. Two possibilities are that nonmem bers are switching to non-pr omoted brands in categories with card-restricted promotions or that the switching is to al ternative categories. This is an important issue given the prevalence of retailers that use category management techniques. At the category level, we estimate standard binary logit models for category purchase incidence for the same large group of categories (c1,2, .,C ). The definitions of the promotion covariates are similar to the st ore-level covariates with the only difference being that they are now meas ured at the category level. In addition to the category specific promotional variables we also include cross-category effects through measures of card-restricted discounts in adjacent product categories (e.g., categories in the same sub-department). T he cross category promotions are labeled _ctOTHNUMCARD for the number of promotions and _ctOTHDISCARD for the level of the discounts. We letitc I equal to 1 if a basket fr om a member of groupiincludes a purchase from categorycin week t and 0 otherwise. The utility derived from and the probability of making a purchase in category c is given by equation (3-5), and the probability that customer i purc hases in category c at time t is then given by equation (36). 123 4 5 6789 10 11 12 13 14__ _ictccic ctcctcct cc tcc tcc tcic t cictc ctci ct cc t cUMEMNUMCARDNUMSPECNUMLOW DISCARDDISSPECDISLOWMEMNUMCARD M EMDISCARDOTHNUMCARDMEMOTHNUMCARD OTHDISCARDMEMOTH 15 16 17 18 19 20 21 4_ ctctct ctctctctctictDISCARDMONTUE WEDTHUFRISUNHOL (3-5)
78 4 4Pr(1)Pr(0) 1,2, c1,2, .,, t=1,2, .,T 1ict ict ict ictU ictict Ue IUiC e (3-6) Estimation results are summarized in Tabl es 3-8a and 3-8b. Table 3-8a reports the parameter estimates that are related to the number of card -restricted promotions while Table 3-8b reports the results related to card-restricted promotion depth. For cardholders all three types of promotions have positive effects on category incidence. These effects are also significantly higher for than those for non cardholders. In addition, for most product categories positive cross-ca tegory effects do not exist for card-holders. For non-cardholders the estimation result s show that the breath of price discrimination does not always decreas e purchase incidence. For example,ctNUMCARD has a significant negative effect in only a small number of categories. For several categories with comparatively high hedonic levels we actually observe a positive impact on category incidence. In other words, card -restricted discounts increase the probability of non-card holders to purchase in these cat egories. This may be partially explained as a salience effect. The explanation is that card-restricted discounts increase noncardholders attention towards these categorie s. For most product categories, both the number and the magnitude of advertised specials and unadvertised specials have positive effects on purchase incidence by non cardholders. The effects are significantly higher for advertised specials than for unadvertised specials. Our analysis reveals limited cross-categor y effects. About half of the product categories yield positive cross-category effe cts. Most of the ca tegories with positive cross-category effects are hedonic categories. This positive cross-category effect may
79 be due to non-cardholders purchasing more hedonic products as a coping mechanism. It is important to note, however, that the si ze of the cross-categor y effects tend to be small compared with the main ef fects of different types of promotions offered within the category. The depth of card-restricted discoun ts has a significant negative impact on purchase incidence for the majority of utilitarian categories. 3.5 Discussion Our findings reveal a set of effects that should be of in terest to both practitioners and academics. For practitioners the findings hi ghlight an important downside to price discrimination. The increasing prevalence of loyalty programs and mechanism s for customizing marketing makes it increasingly possible for firms to provide preferential treatment to segments of consumers. Howeve r, while the use of targeted promotions has the potential to increase customer value (Khan, Lewis and Singh 2009) there is also a significant downside since unf airness is likely to result in customer dissatisfaction. Our primary goal in this research was to st udy the differences in how consumers that are the beneficiaries and victim s of price discrimination reac t. Table 3-9 summarizes the findings and our init ial predictions. The results for the cardholder were roughly as expected with one notable exception. The negative relationship between the number of card-only discounts and the number of items purchased by cardholders is an interesting result. This result is at odds with our hypothesized negative relationship. In addition to a basic economic argument that more discounts are likely to motivate items being purchased, previous research has also found that positive feelings increase purchase intentions (Brown et al. 1998). While much of our discussion has fo cused on whether pr ice discrimination creates negative emotions in consumers that do not receive discounts, the impact of
80 discounts on loyalty program members is also an open issue. Our results suggest that the loyalty program does not create positiv e emotions. While we might expect that providing discounts to member s would create positive emot ions, the existing survey evidence does suggest that frequent buyer pr ograms often fail to create attitudinal loyalty. For example, an IBM survey ( 2007) found that 73% of customers feel antagonistic towards or feel no loyalty to thei r supermarket. One of the drivers of these negative feeling are concerns about privacy. Survey data suggests that 52% of customers are concerned about the collection of their personal information (Boston University 2004). This may be a problem if consumers believe that the trade of discounts for personal data is skewed to the retailers favor. Our empirical results are consistent with a lack of positive emotions due to cardrestricted promotions. For exam ple, we find that basket size is not increased by the number of card-restricted promot ions. The lack of a basket size effect suggests that the program promotions are shifting demand within the store rat her than expanding demand. The negative relationship between items purchased and card-restricted discounts further suggests that these discoun ts do not create positive moods. It may be that this type of loyalty progr am is viewed as neutral by many consumers. This may be the case because the programs are commo n across many grocery retailers and because consumers may not feel that the benefit s are particularly valuable. For the non-members, or victims of price discrimination, the analyses yield multiple notable findings. In terms of store traffic we did not find a significant negative effect for the number of card-restricted promotions. This finding contradicts our hypothesized relationship as we expected t hat non-members would act to avoid price
81 discrimination. A possible ex planation is that non member s may not be differentiating between advertised card-only and ordinary advertised specials. This may be a reasonable conjecture as grocery shoppi ng has been found to be a relatively low involvement task where consumers have lim ited price knowledge ( Dickson and Sawyer 1990). The basket size results are also of great interest as we find a surprising positive relationship between the number of restricted promotions and the basket size metrics. Interestingly, our results sugges t that from the retailers perspective price discrimination has some positive effects on non-members as they tend to purchase larger baskets and also tend to pay higher prices sinc e they may avoid pr omoted items. At the brand level, card-restricted promotions are successful in driving purchase by card-holders but usually have a negativ e, rather than neut ral, effect on nonmembers. The implication is that manufacturers should be wary of participating in loyalty program restricted promotions. The combined findings that increasing numbers of card-restricted promotions lead to lar ger baskets and that nonmembers avoid cardpromoted brands is an interesting combinatio n of results. Earlier we suggested that nonmembers might engage in coping behaviors to avoid being discriminated against. It appears that they do avoid discrim inatory prices but that they switch to other products within categories rather than simply avoiding the category. The findings related to hedonic products also suggest that emoti on plays a key role in response to discrimination. 3.6 Conclusion As CRM systems become more a dv anced and better integrated with communications technologies, firms are increasingl y able to practice mi croor individual level targeting. However, instances of price customi zation are often inflammatory to
82 the point where they are discussed in the me dia. For example, the use of dynamic and customized pricing policies at companies rang ing from Amazon, Victorias Secret and Coca Cola have all been highlighted by t he media (Egan 2001; Ha milton 2001; Stevens 1996). Our research was conducted in the specif ic context of a grocery retailers loyalty program. One notable feat ure of this program is that customers choose whether or not to become a member of the pr ogram. Consumer reactions to this type of self-selection based discrimination may be different than ot her mechanisms, particularly mechanisms that remove control from consumers. Additi onal field research that studies consumer reactions to different types of price discrim ination would be valuabl e. For academic audiences the estimation results provide field data on the effects of price discrimination. These results both comp lement the existing li terature on price fairness and highlight opportunities for future re search. Specifically, it would be useful to further investigate several of the more notable findings in a laboratory setting. For example the findings related to hedonic ca tegories could be further explored in controlled environments. Laborator y studies that examine the emotional effects of the type of price discrimination we consider would be useful since we do not directly observe or measure customer emotions in the field data. In addition our empirical context occurs in an environment where cons umers have control over whether they are the victims of price discrimination. Laboratory studies t hat examine the consequences of consumer control in addition to reference transactions would be of value. It would also be useful to study how price discrimination impacts customer relationships. A key limitati on of our study derives from our inability to track noncardholders over time. This is critical in at least two respects. First we are unable to
83 understand the dynamics of non-card holders brand and category decisions. It would be interesting to evaluate how promotions change behavior based on past levels of brand loyalty. Our inability to track noncardholders also impacted our analysis of cardholders. For example, we chose to empl oy simple brand choice models that did not include measures of past loyalty. This was a necessary approach for the noncardholders and we used the same struct ure for the cardholders for sake of consistency. The lack of a means to tra ck the longitudinal behavior of non-cardholders also limits our ability to understand how pr ice discrimination impacts customer relationships. In our specific context, it would be interesti ng to track whether the price discrimination leads to attrit ion or if it leads non-mem bers to become card holders.
84 Table 3-1. Customer descriptive statistics Total Population Card Holders Non-Members Number of Daily Visits 1435 982 (68.4%) 453 Average Basket Size $ 22.47 $ 26.56 $ 14.28 Items 12.02 units 17,255.1 14.25 units 13,993.5 7.2 units 3,261.6 Revenue $ 32,246.34 $ 25,777.50 (79.94%) $ 6,468.84 Promotion Response Card Items 1.18 1.63 11.40% .21 2.96% Weekly Specials .51 .66 4.63% .18 2.44% Buy Low .46 .60 4.21% .16 2.22% Table 3-2. Promotion descriptive statistics All Products Number Range Mean ($) Discount Mean (%) Discount Card Promotions 123.9 77 to 161 $ 1.12 38.8% Weekly Specials 88.5 61 to 120 $ .83 24.0% Buy Lows 693.1 374 to 947 $ .50 22.5% Utilitarian Categories Card Promotions 68.5 43 to 114 $ 1.00 44.5% Weekly Specials 31.9 13 to 47 $ .88 33.6% Buy Lows 677.0 360 to 924 $ .50 22.5% Hedonic Categories Card Promotions 55.5 32 to 72 $ 1.25 32.0% Weekly Specials 56.5 37 to 73 $ .80 18.0% Buy Lows 15.9 9 to 26 $ .78 25.7% Table 3-3. Hypothesized effects of ca rd-restricted promotions on segments Card Holders Non-Members Store Traffic Positive Negative Average Expenditures ? Negative Average Number of Items Positive Negative Brand Choice Positive Negative Brand Choice (Hedonic) Positive ? Category Incidence Positive ?
85 Table 3-4. Variable definitions Variable Name Definition NUMCARD Weighted total number of card holder-only specials offered in the store (or in a category). OTH_NUMCAWeighted total number of card holder-on ly specials offered in all other categories sub-department. NUMSPEC Weighted total number of advertised specials offered in the store (or in a category). NUMLOW Weighted total number of unadvertised sp ecials offered in the store (or in a DISCARD Weighted average dollar value of card hol der-only discounts offered in the store (or in a category). OTH_DISCARWeighted average dollar value of card holder-only discounts in all other categories same sub-department. DISSPEC Weighted average dollar value of advertised discounts offered in the store (or in a category). DISLOW Weighted average dollar value of unadvertis ed discounts offered in the store (or in a cate g or y) LCARD 1 if a card-only special is offered for a brand and 0 otherwise. SPEC 1 if an advertised special is offered for a brand and 0 otherwise. LOW 1 if an unadvertised special is offered for a brand and 0 otherwise. REGP Regular price of a brand. DISC The dollar value of the discount if a br and is offered as a special of any type. MEM 1 if the dependent variable(s) is meas ured for a card-holder and 0 otherwise. HOL 1 if the day is during holidays. MON 1 if the day is a Monday. TUE 1 if the day is a Tuesday. WED 1 if the day is a Wednesday. THU 1 if the day is a Thursday. FRI 1 if the day is a Friday. SUN 1 if the day is a Sunday.
86 Table 3-5. Store traffic and basket size estimation results Variable Store Traffic Average Number of Items Purchased Average Basket Size ($) Coefficient SE Coefficient SE Coefficient SE INTERCEPT 396.338** 155.959 6.442*** 1.622 10.782*** 1.852 MEM 431.256*** 141.691 7.517*** 1.407 11.561*** 1.607 NUMCARD -0.589 0.747 0.026*** 0.007 0.049*** 0.008 MEMNUMCARD 1.476** 0.729 -0.022*** 0.007 -0.045*** 0.008 NUMSPEC 1.141*** 0.254 0.004** 0.002 0.011*** 0.002 NUMLOW 0.001** 0 0.001** 0 DISCARD -76.75 105.43 -1.832* 1.091 -3.847*** 1.246 MEMDISCARD 144.19 111.177 2.003* 1.115 5.749*** 1.273 DISSPEC 64.541 54.905 2.375*** 0.414 5.014*** 0.473 DISLOW -3.274** 1.496 -1.755 1.709 HOL 151.563*** 31.855 5.151*** 0.209 7.369*** 0.239 MON -260.189*** 20.236 -3.083*** 0.142 -6.448*** 0.163 TUE -265.822*** 20.236 -3.221*** 0.143 -6.249*** 0.163 WED -250.213*** 20.292 -2.896*** 0.142 -5.705*** 0.162 THU -242.278*** 20.349 -2.227*** 0.142 -4.269*** 0.162 FRI -94.823*** 20.352 -1.259*** 0.136 -1.94*** 0.155 SUN -234.165*** 20.351 0.847*** 0.142 0.772*** 0.162 R 0.717 0.856 0.835 Adjusted-R 0.710 0.852 0.830 WeightedR 0.721 ***0.01 p ; **0.05 p ; *0.1 p
87 Table 3-6. Descriptive measur es of chosen products categories Sales/Week Promotions/Week Number Mean($) Discount Category Quantity Dollar Penetrati on rate Card Promotion Weekly Special Buy Lows Card Promotion Weekly Special Buy Lows BAKED BEANS CAN 234.31 $251.59 2.35% 2.00 1.00 6.50 0.27 0.08 0.25 BOLOGNA 178.62 $537.74 2.93% 2.00 2.00 0.00 1.07 0.59 0.00 CEREAL 1654.04 $4,420.95 10.93% 2.00 1.50 2.45 1.31 1.14 0.85 COOKIES-BAKERY 537.38 $1,420.38 4. 98% 1.00 1.00 1.00 0.90 0.55 0.60 COOKIES-GROCERY 801.09 $1,782.53 6. 13% 4.50 2.50 29.68 1.24 0.82 0.46 CRACKERS SNACK 530.90 $1,270.27 4. 37% 1.00 1.00 14.81 1.12 1.01 0.43 CANDY NON SINGLES 512.43 $1,191.21 3.80% 1.00 2.00 5.88 3.00 0.40 0.46 CHIPS POTATO 709.02 $1,334.58 5.97% 9.00 5.45 12.63 1.44 1.03 0.63 CREAM SOUR 312.44 $401.22 3.04% 2.60 3.00 4.13 0.49 0.46 0.28 CANDIES UNWRAPPED 482.78 $1,055.63 4. 97% 1.60 1.00 0.00 0.44 0.50 0.00 DETERGENT DISH 357.63 $1,003.11 3. 09% 2.00 2.50 3.89 1.36 1.86 0.43 DETERGENT LIQUID LAUNDRY 164.62 $666.55 1.51% 2.88 2.67 3.55 2.01 2.28 1.04 DRINK CARBONATED SOFT 1946.67 $3,304.01 11.35% 17.67 13.13 7.67 1.05 0.55 0.63 ENGLISH MUFFINS 245.39 $399.23 2. 07% 2.00 0.00 6.00 1.20 0.00 0.92 FRUIT CANNED OR BOTTLED 954.50 $1,219.31 5.95% 4.00 3.00 14.00 1.01 0.35 0.34
88 Table 3-6 Continue Sales/Week Promotions/Week Number Mean($) Discount Category Quantity Dollar Penetrati on rate Card Promotion Weekly Special Buy Lows Card Promotion Weekly Special Buy Lows FACIAL TISSUE 838.40 $1,096.42 4. 94% 2.00 2.00 0.00 0.51 0.41 0.00 HOT DOG 236.00 $659.16 2.15% 3.73 2.00 0.00 0.99 0.60 0.00 ICE CREAM 592.79 $1,891.37 4.96% 9.60 19.00 0.00 1.69 1.71 0.00 CHEESE IMPORTED 211.75 $1,303.19 3. 13% 2.33 1.75 0.00 2.87 1.25 0.00 JUICE/DRINKS ASEPTIC 265.81 $553.87 1. 88% 2.60 0.00 9.57 1.63 0.00 0.52 JAMS/JELLIES 302.00 $699.91 2.57% 1.00 0.00 4.00 1.11 0.00 0.56 MIX CAKE 197.19 $268.42 1.64% 6.00 0.00 8.48 0.81 0.00 0.41 ORANGE JUICE FROZEN 320.96 $479.28 1.66% 3.00 0.00 3.22 1.06 0.00 0.32 PASTA 1216.22 $1,599.84 8.63% 4.43 5.00 17.26 0.64 0.77 0.31 PIZZA 12 IN 183.90 $775.92 1.53% 1.00 1.00 0.00 1.00 0.73 0.00 POTATOES FROZEN 282.24 $565.41 2.22% 3.00 3.00 6.94 1.30 0.79 0.63 PAPER TOWELS 909.52 $1,199.92 6.39% 2.50 1.50 1.67 0.31 0.50 0.34 SOUP CANNED 1754.53 $2,230.24 8. 74% 3.80 5.50 4.25 0.58 0.20 0.27 TOILET TISSUE 758.53 $1,565.03 6. 19% 2.00 1.80 2.00 0.31 0.67 0.41 VEGETABLES CANNED/JARRED 840.09 $973.21 5.84% 6.00 1.00 4.30 0.31 0.21 0.21 WAFFLES FROZEN 179.48 $374.66 1.52% 4.50 0.00 6.22 0.83 0.00 0.57 YOGURT 1494.47 $1,639.46 7.16% 20.74 6.00 13.98 0.37 0.31 0.46
89 Table 3-7. Brand choice estimation results Category LCARD SPE C LOW MEML CARD MEMS PEC MEM LOW REGP DISC FACIAL TISSUE -5.565 4.457n/a 13.199 1.489 1.152 -0.572 1.42 TOILET TISSUE -2.296 1.869 0.503 7.036 1.509 0.257 -0.142 -0.08 CREAM SOUR -1.622 0.649 -0.86 2.8 0.726 0.935 0.394 2.065 ENGLISH MUFFINS -1.542 n/a -0.90 2.687 n/a -0.073 1.489 1.842 VEGETABLES CANNED -1.164 1.654 n/a 3.905 0.472 1.109 -0.065 -0.51 BAKED BEANS CAN -1.049 2.246 0.292 4.89 1.078 0.152 -0.057 -0.13 PAPER TOWELS -1.018 1.113-0.04 4.298 2.184 0.597 -0.322 1.009 SOUP CANNED -0.963 0.522 -0.72 2.886 1.794 0.868 -0.35 2.152 MIX CAKE -0.95 n/a 0.409 2.843 n/a 0.143 -0.069 1.673 CRACKERS SNACK -0.771 1.108 0.223 3.058 -0.085 0.563 -0.01 0.58 ORANGE JUICE FROZEN -0.718 n/a 0.017 2.762 2.841 0.432 1.266 0.565 PIZZA 12 IN -0.675 0.297n/a 1.973 0.986 n/a -0.189 0.722 FRUIT CANNED -0.646 0.174 0.368 3.628 2.396 0.667 -0.31 0.808 BOLOGNA -0.619 0.617n/a 1.452 0.194 0.507 -0.107 0.783 POTATOES FROZEN -0.6 0.4360.08 3.611 2.691 0.244 -0.128 0.119 CEREAL -0.568 0.731 0.389 2.153 0.811 -0.014 0.127 1.304 YOGURT -0.449 0.946 0.205 2.751 1.216 0.2 -0.185 0.004 PASTA -0.395 0.472 0.326 2.615 0.717 0.394 -0.308 1.28 WAFFLES FROZEN -0.311 n/a 0.283 2.992 0.791 0.25 0.431 0.682 JAMS & JELLIES -0.285 n/a 1.121 2.28 0.514 0.165 -0.227 1.885 DRINK CARBONATED -0.28 0.136-0.05 1.476 1.023 0.636 -0.108 0.719 HOT DOG -0.27 -0.14 n/a 1.397 0.862 n/a -0.219 0.64 DETERGENT DISH -0.233 0.854-.004 2.874 1.762 0.514 -0.082 0.331 DETERGENT LIQUID LAUNDRY -0.196 -0.23 0.106 3.157 2.804 0.772 -0.015 0.272 JUICE/DRINKS ASEPTIC -0.12 n/a 0.15 2.443 2.262 0.413 -0.152 0.425 ICE CREAM -0.029 0.476n/a 2.239 0.906 0.573 -0.079 0.167 CHIPS POTATO 0.046 0.5150.02 2.926 1.754 0.71 -0.061 -0.25 COOKIES-BAKERY 0.087 0.757-0.38 2.17 0.831 0.31 -0.115 -0.11 COOKIES-GROCERY 0.181 0.93 0.179 2.205 0.557 0.658 0.035 0.178 CHEESE IMPORTED 0.724 0.413n/a 2.801 1.054 n/a -0.124 0.317 CANDIES UNWRAPPED 0.914 1.002n/a 1.364 0.467 n/a -0.286 0.149 CANDY NON SINGLES 1.236 -0.17 -0.06 1.649 0.736 0.811 -0.05 0.221 Average -0.629 0.6820.145 3. 141 1.167 0.358 -0.018 0.664
90 Table 3-8a. Category incidence re sults number of promotions Category MEM NCAR D MEM NCAR D NSPE C NLOW O_NCA RD MEM O_NCA RD FACIAL TISSUE 1.435 -0.194 0.466 0.382 -0.067 0.004 -0.006 ORANGE JUICE FROZEN 1.326 -0.078 0.289 0.948 -0.014 0.002 -0.002 TOILET TISSUE 1.088 -0.049 0.459 0.265 0.002 0.002 -0.001 PAPER TOWELS 1.556 -0.038 0.217 0.375 0.028 0.002 -0.002 BOLOGNA 1.075 -0.035 0.059 0.030 0.051 0.011 -0.015 SOUP CANNED 2.477 -0.024 0.088 0.158 -0.010 0.002 -0.003 BAKED BEANS CAN 2.142 -0.024 0.091 0.329 0.051 0.008 -0.008 CREAM SOUR 0.897 -0.012 0.162 0.223 0.029 0.001 -0.002 DRINK CARBONATED SOFT 0.883 0.006 0.006 0.019 -0.003 0.008 -0.008 VEGETABLES CANNED/JARRED 1.295 0.013 -0.014 0.040 0.001 0.004 -0.005 CEREAL COLD READY TO EAT 1.089 0.020 0.041 0.182 -0.005 -0.002 0.001 HOT DOG 0.517 0.023 0.001 -0.064 n/a 0.019 -0.020 WAFFLES FROZEN 0.461 0.023 0.024 0.051 -0.004 0.017 -0.017 DETERGENT LIQUID LAUNDRY 0.428 0.024 0.062 0.255 0.010 0.003 -0.002 YOGURT 1.006 0.027 -0.012 0.008 0.006 0.024 -0.026 PIZZA 12 IN 0.415 0.030 0.137 0.051 n/a 0.005 -0.004 PASTA 1.086 0.033 0.118 0.034 0.007 0.007 -0.008 DETERGENT DISH 1.573 0.037 0.132 0.082 0.013 0.004 -0.005 FRUIT CANNED OR BOTTLED 1.752 0.038 0.164 0.048 0. 019 0.002 -0.004 MIX CAKE 1.561 0.041 0.142 0.058 0.003 0.003 -0.001 JUICE/DRINKS ASEPTIC 0.718 0.042 -0.019 0.013 0.012 0.004 -0.005 POTATOES FROZEN 0.764 0.044 0.171 0.139 0.002 -0.003 0.003 COOKIES-GROCERY 0.668 0.050 0.073 0.019 0.005 0.012 -0.012 CHIPS POTATO 1.221 0.051 0.043 0.025 0.003 0.012 -0.013 CHEESE IMPORTED 0.299 0.055 0.060 -0.010 n/a 0.007 -0.008 ENGLISH MUFFINS 0.786 0.055 0.054 n/a 0.075 0.011 -0.012 ICE CREAM 0.547 0.059 0.051 0.032 0.002 0.024 -0.028 CANDIES UNWRAPPED 0.242 0.067 0.112 0.097 n/a 0.005 -0.004 CANDY NON SINGLES 0.113 0.080 0.127 -0.041 0.005 0.006 -0.009 COOKIES-BAKERY 0.043 0.140 0.249 0.037 -0.022 0.021 -0.037 CRACKERS SNACK 0.120 0.218 0.225 0.016 0.008 0.009 -0.008 JAMS/JELLIES/PRESERVES 0.712 0.293 0.473 0.124 0.009 0.005 -0.006 Average 0.939 0.032 0.133 0.126 0.008 0.007 -0.009
91 Table 3-8b. Category incidence results depth of promotions Category DISCA RD MEMDISC ARD DISSP EC DISL OW OTH_DISC ARD MEMOTH_DI SCARD FACIAL TISSUE n/a n/a 0.601 1.094 -0.246 0.264 ORANGE JUICE FROZEN -0.248 0.288 n/a 0.589 0.151 -0.244 TOILET TISSUE -0.209 0.122 -0.116 0.011 -0.586 0.550 PAPER TOWELS -1.309 1.290 -0.173 -0.609 -0.501 0.552 BOLOGNA -0.152 0.897 0.560 -0.167 0.177 -0.210 SOUP CANNED -0.313 0.998 0.672 0.377 -0.236 0.257 BAKED BEANS CAN n/a 2.812 1.055 -0.083 -0.765 0.861 CREAM SOUR 0.543 -0.097 1.069 0.475 -0.175 0.233 DRINK CARBONATED SOFT -0.034 0.041 -0.229 0.081 -0.082 0.044 VEGETABLES CANNED/JARRED -0.530 1.968 0.034 0.070 0.207 -0.269 CEREAL COLD READY TO EAT -0.104 0.202 -0.162 0.036 -0.336 0.102 HOT DOG 0.093 0.264 0.238 n/a -0.344 0.125 WAFFLES FROZEN -0.732 1.171 0.579 0.353 -0.621 0.486 DETERGENT LIQUID LAUNDRY -0.102 0.448 -0.056 0.133 -0.910 0.781 YOGURT -0.817 2.164 1.257 -0.071 0.017 -0.067 PIZZA 12 IN n/a -0.469 0.220 n/a 0.080 -0.094 PASTA 0.113 -0.061 -0.489 -0.481 0.243 -0.275 DETERGENT DISH -0.078 0.383 0.132 0.033 -0.353 0.235 FRUIT CANNED OR BOTTLED n/a 0.332 0.250 0.081 -0.373 0.519 MIX CAKE 0.470 -0.163 n/a 0.334 0.305 -0.477 JUICE/DRINKS ASEPTIC -0.164 0.369 0.618 0.109 0.125 -0.193 POTATOES FROZEN n/a 0.464 0.294 0.208 -0.248 0.298 COOKIES-GROCERY -0.088 -0.002 0.023 0.231 -0.105 0.109 CHIPS POTATO 0.161 -0.028 0.063 -0.252 0.119 -0.089 CHEESE IMPORTED 0.034 -0.027 0.102 n/a 0.077 -0.111 ENGLISH MUFFINS n/a 1.720 n/a 0.275 0.252 0.135 ICE CREAM -0.133 0.381 -0.125 0.140 -0.151 0.112 CANDIES UNWRAPPED 0.176 -0.094 -0.114 n/a 0.032 -0.012 CANDY NON SINGLES n/a 0.019 0.167 -0.123 -0.222 0.289 COOKIES-BAKERY -0.113 0.165 0.005 0.084 -0.240 0.397 CRACKERS SNACK -0.208 0.212 0.110 0.393 0.141 -0.200 JAMS/JELLIES/PRESER VES n/a 0.258 0.361 0.077 0.022 -0.096 Average -0.156 0.517 0.239 0.121 -0.142 0.125
92 Table 3-9. Hypothesized relationships and empirical findings Card Holders NonMembers Hypotheses Empirical Findings HypothesesEmpirical Findings Store Traffic Positive Positive (S upported) Negative No Relationship Average Expenditures ? Positive for Discount Depth No Effect for Discount Breadth Negative Positive for Discount Breadth, Negative for Discount Depth Average Number of Items Positive No Effects for depth or breadth of discounts Negative Positive for Discount Breadth Negative for Discount Depth Brand Choice Positive Positive (Supported) Negative Negative (Supported) Brand Choice (Hedonic) Positive Positive (Supported) ? Negative effects of discounts are reduced for hedonic products Category Incidence Positive Positive (Supported) ? Figure 3-1. Conceptual map Customer Status Card/No Card Customer Emotional State Category Activity Type Depth Breadth Category Choice Category Outcome Incidence Brand Consumer Choice Customer Total
93 CHAPTER 4 AN EMPIRICAL STUDY OF THE IMPAC T OF REST RICTING RETURN POLICY This chapter studies another important interface decision, managing product returns. I conduct an empirical study to ex amine the impact of return policies with different levels of leniency on consumers purchase and return behavior. 4.1 Introduction Many retailers offer lenient return polic ies, under which customers are allowed to return the product with no (or minimal) hassl e cost. A lenient return policy is an important element of the reta ilers effort to build good customer relationships and maintain a competitive advantage based on good service ( Business 2.0, Mar. 15, 2007). However, a lenient return policy may encourage product returns and significantly increase operation costs ( Sloan Management Review Oct. 1, 2006). Based on a survey conducted in 2010 by the National Retail Federation (NRF) and The Retail Equation, the return rate of the NRF retail industry in 2009 was over 8%, which represents an increase of 10% from 2007 ( 2009 Customer Returns in the Retail Industry Report ). The significant operation costs associated with product returns may explain the prevailing observation that many retailers have tightened return policies in recent years by, for example, shortening time limits, asking for the original receipt, or even only offering partial refunds ( The Wall Street Journal, May 08, 2008, Consumer World Dec. 15, 2008). While the existing literature on m anaging product returns has proposed and examined a variety of benefits for sellers who allow product returns (e.g., Moorthy and Srinivasan 1995, Che 1996, Davis et al. 1995) very limited attention has been paid to the issue of how to design a return policy, and especially how to choose the leniency
94 level (Davis et al. 1998, Wood 2001, Hess et al. 1996). Davis et al. (1998) developed an analytical model to solve for the optimal leve l of hassle cost the seller should impose on consumers when returning a product and used cross-industry data to support the predictions of the theoretic al model. Wood (2001) conduct ed several experiments to examine how the leniency level of the return policy influences consumers purchase and return decisions. Hess et al. (1996) proposed that the retailer may charge nonrefundable fees to prevent consumers from exploiting return policies. In order to choose the optimal level of return policy leni ency, retailers must have a good understanding of customers purchase and return behavior and, more importantly, how such behavior is affected by different re turn policies. Some initial studies have investigated customers purchase and/or re turn behavior using the detailed transaction and return information that is generally avail able in retailers customer databases (Hess and Mayhew 1991, Anderson et al. 2009, Pe tersen and Kumar 2009). All the existing work, however, examines consumers behavior under a single return policy. We conduct an empirical study using the tran saction and return data from a multichannel retailer who restricted its return policy during the data collection period by shortening the return time limit from 90 to 30 days. We develop a joint model of customers purchase incidence, return rate and return duration de cisions to examine how these decisions are influenced by the po licy change. Interestingly, we find that consumers respond to the restricted return po licy by returning faster but not returning less. Furthermore, shortening the return limit also leads to a significantly lower purchase rate.
95 Our paper contributes to the product return literature. The theoretical literature on product returns has mostly focused on the benefit of allowing cust omers to return a purchased product (Moorthy and Srinivasan 1995, Che 1996, Davis et al. 1995, 1998). For example, the ability to re turn a purchased product may be offered to the customer by the seller to signal high product quality (M oorthy and Srinivasan 1995), or to work as a risk-sharing mechanism between seller an d buyer (Che 1996), or to increase the customers willingness to pay upfront when there is incomplete information about product quality (Davis et al. 1995). The increa sing empirical literature on product returns focuses on modeling whether and when a cu stomer will make a return (Hess and Mayhew 1991), evaluating the degree to which customers value the option to return a product by comparing their purchase and return behavior both with and without the return opportunity (Anderson et al. 2009), and examining potential antecedents and consequences of returning a product (P etersen and Kumar 2009). On the issue of return policy leniency, Woods (2001) experiment al work identified a pattern similar to ours, i.e., that the less lenient return poli cy leads to a lower purchase rate, but does not significantly decrease the retu rn rate. Davis et al.s (1998) analytical model captures how the return hassle cost influences retu rn and purchase rates and looks at how the optimal hassle cost is influenced by factors such as cross-selling opportunities, salvage cost of a returned product, and how fast the product benefit is consumed. Our paper makes an initial effort to examine and com pare different return policies using real transaction data from a retailer. This paper also adds to the literature on the interface issues between marketing and operations management (Ho and Tang 2004). Confli cts between marketing and
96 operations management are longexisting challenges faced by retailers. Using detailed transaction data generally available to reta ilers to analyze the impact of interface decisions on consumer behavior and to im prove decision making based on consumer insights offers a potential way to coordi nate marketing and operati ons decisions. This opportunity has been illustrated by the prior empirical literature that has studied interface issues such as inventory m anagement and stockouts (Anderson et al. 2006, Jing and Lewis 2010). Product return is an impor tant interface issue and design of the return policy influences both market ing and operations management outcomes (Anderson et al. 2009). We model different as pects of customer decisions related to both marketing returns (purchases) and operati on costs (number of returns and return duration). Our paper also contributes to the literat ure on service marketing (Rust and Chung 2006). Our findings illustrate that return policy, as an impo rtant measure of the service offered by retailers, will influence consumers purchase decisions (Zeithaml et al. 1996). Over the past few years, some firms have dec reased their service quality for all or some consumers, or have increased the hassle cost for customers to obtain good service, in order to decrease service cost ( Business Week Oct. 23, 2000). Our results emphasize that, when the consequences of decreased service quality on consumer behavior especially on consumers strate gic reactions are taken into consideration, this type of practice may have a negative impact on customer relationships and on firm profitability. 4.2 Empirical Context The data for the study is der ived from a multi-channel reta iler selling footwear and other accessories. The dataset contains re cords (both purchases and returns) of all customer transactions and returns over a 62month period, i.e., mo re than 27 million
97 transactions and 3 million returns made by ov er 11 million customers. The sample for our empirical estimation is randomly drawn fr om the dataset. It includes 2805 customers who made at least one purchase and one return during the data collection periods and, in total, represents 13520 transactions and 3802 returns. Table 4-1 provides some basic descriptive statistics related to transaction results and customer histories. On average, each transaction is valued at $61.68, and the product cost averages $35.43. Customers make one purchase every 5 months and return at least one item from a transaction for an average of 28.1% of transactions. The average return duration is 9.18 days. Cumula tively, the average customer makes about 5 purchases and 1.4 returns in the 62-month period. Among all the orders, 6.7% are online transactions and the av erage customer resides 6.82 m iles away from the closest offline store. The average age of the customers is 27, 40% of them are revealed as females, and 16% of them are gold members18. On the marketing side, the average weekly price is $35.71, and customers rece ive at least one coupo n/advertisement via email every 2.5 months. Table 4-2 presents a simple comparison of three focus transaction/return measures before and after the policy change (which took place in Month 54). The numbers show that, after the policy change, the purchase frequency decreases from 0.23 to 0.16; the return rate increases from 27.4% to 33.3%, and the average return duration decreases from 9.72 days to 7.27 days. While the simple comparison shows some interesting patterns, we must model t he three relevant decisions jointly and also 18 There is no requirement for customers to register as a standard member. As a gold member, however, a customer must spend at least $200 annually.
98 control for variations in other marketing variables and customer characteristics to examine the impact of the policy change. 4.3 Model In this secti on we describe our modeli ng approach. For the analysis, we develop a model to capture three elements of a cust omers decision, whether the customer will make a purchase, whether a customer will make a return for each transaction, and when the customer will make the return. The model is an extension of Hess and Mayhew (1991), which proposed a split hazard model to capture w hether the customer will make a return and when to make the re turn. Our model extends theirs by also incorporating the purchase decis ion in order to have a more complete picture of the impact of the changed return poli cy. Our model also captures the return time constraint set up by the retailer, which is not consi dered by Hess and Mayhew (1991), because in their data customers are allowed to return the product at any time. We also consider unobserved consumer heterogeneity that is neglected in their model. We describe the three decisions sequentially and at the end derive the joint distribution of the three decisions and the maxi mum likelihood function. 4.3.1 Purchase Incidence Model We begin by considering consumers purc hase incidenc e decisions. In each time period (month) we assume that customers make a decision about whether to make a purchase. We letim I equal to 1 if customeridecides to make a purchase in month m, and 0 otherwise. The utility derived from customeri through making a purchase in month m is given by
99 0 1, with ., and 1,2, .,imimim iUXmmMiN (4-1) where im X is a vector of independent variables. T he X variables include factors that influence the utility of a pur chase, such as marketing mix elements, household demographics, transaction characte ristics, and the indicator of the return policy change. By assuming that im follows a logistic distribution wit h location parameter 0 and scale parameter 1, the prob ability for customeri to make an order in month mcan be written as in equation (4-2). 1 1Pr(1)Pr(0) 1im imX imim Xe IU e (4-2) 4.3.2 Return Rate Model We next consider the return decision. For each transaction the consumer made with the retailer, we model wh ether the customer wants to make a return. We assume that imV in equation (4-3) is the latent va riable that determines whether customeri wants to make a return on the transaction made in monthm19. 2imimimVX (4-3) By assuming that the conditional (on t he fact that a transaction is made in monthm) distribution ofim is a logistic distribution with location parameter 0 and scale parameter 1, the prob ability that customeriwill want to make a return *1imY for that transaction can be written as in equation (4-4). 2 2*Pr(1)Pr(01) 1im im imX imimim Xe YIVI e (4-4) 19 We can view it as the difference between the net utility of returning versus keeping the product.
100 Notice that whether the customer wants to return the product (*imY) may not always be the same as whether we observe a return (imY) in the dataset, as we will explain below. 4.3.3 Return Duration Model On the condition that a customer wishes to do so, we model when the customer intends to make a return (or the return duration)*imt. We adopt the proportional hazard model to capture the intended return durat ion and use a flexible Box-Cox formulation that captures several possible shapes (G onul, Kim, Shi 2000) of the baseline hazard function. More specifically, the hazard rate of the return duration (i n days) satisfies the following distribution. **** 03 ***2 12343(1,1)(1,1)*exp() expln*exp()im imimim imim im im imim imhtIYhtIYX tttX (4-5) We also define imtas the observed return duration as per the following equation: ** 0imim imttT t otherwise (4-6) Where Tis the return time limit (which equals 90 days before month 54 and 30 days afterwards). The above equation states t hat we will observe a return in the dataset only if the customer wants to make the return within the time limit. In other words, we may observe a non-return either because the customer does not want to return the product, or the customer wants to retu rn it but the required time limit Thas passed. Accordingly, the probability we observe a non-return 0imY in the dataset is ** *Pr(01)Pr(01)Pr(1,1)im im imim im imimYIYIYtTI (4-7)
101 Therefore, combining all th ree elements, we derive the joint probability for the three types of observations in the datase t: First, the probability that we observe customerimake a purchase in month mand return it afterimtT days is: 12 12 12 12* **Pr(1,1,)Pr(1,1,) *(1,1) 11 (1,1)*(1,1) 11im im im it it im it it it it im im it itim im im XX imim XX XX imim imim XXIYtIYt ee ftIY ee ee htIYStIY ee (4-8) where *(1,1)imimimftIY is the density function of the return duration imt, and *(1,1)imimimStIY is the corresponding survival rate (i.e., the probability that the customer wants to return the purchased item afterimt). Second, the probability that we observe customerimake a purchase in month mwhich is not returned afterwards is the sum of the probability that she makes the purchase and does not want to return it, and the probability t hat she makes the purchase and wants to return it after the time limit T: 12 122** *Pr(1,0)Pr(1,0)Pr(1,1,) 1 *(1,1) 111im im it it im it it itimim im im im XX im XXXIYIYIYtT ee STIY eee (4-9) Third, the probability that customeridoes not make a purchase in month m is: 11 Pr(0) 1imim XI e (4-10) 4.3.4 Consumer Heterogeneity and the Log-Likelihood Function We adopt a latent class model to c apture unobserved consumer heterogeneity (Kamakura and Russell 1989). The latent class model treats the population as a mixture
102 of unobserved types. A vector of parameters will be estima ted for each type in the population, and the likelihood function is a weighted average of the type-specific likelihoods. If we a ssume that there are K segments of consumers, the log-likelihood function is: 01** ( 1 ) 11logPr(0)Pr(1,1,)Pr(1,0)im imim imim iM NK IIYIY kkimkimimimkimim ik mmLL I IYtIY (4-11) wherek is the proportion of typekcustomers in the population. 4.4 Empirical Analysis We now detail the full set of covariat es used in the demand model. Table 4-3 includes the set of covariates that is ex pected to influence the customers purchase and return behavior. These covariates are di vided into three categories: Marketing Variables, Customer and Trans action Characteristics, a nd The Return Policy Change Indicator. Marketing variables We include two types of ma rketing variable, PRICE and COUPON. Because the retailer carries an enormous number of SKUs that change frequently each season, we construct a price index to describe the price environment the consumer faces. Specific ally, PRICE is defined as the weighted average price of items purchased by customers in the reward s program, over the ent ire data collection period, in each month. The covariate used in the model NPRICE is the mean-centered value of the price index. In addition to product pricing, the retailer is also actively involved in direct marketing promotions. During t he data collection period, the retailer constantly sent out different types of coupon, e.g., seasonal promotion coupons, new customer acquisition coupons, birthday coupons, and new product in troduction coupons, to consumers via
103 snail mail or email. We construct a bi nary variable COUPON, which equals 1 if the customer receives a c oupon and 0 otherwise. Customer and transaction characteristics In addition to ma rketing variables, we also include customer-level information in the model. Demographic variables include the customers AGE, the binary variable FE MALE, which equals 1 if the customer is revealed as female and 0 otherwise; and GMEM which equals 1 if the customer is a gold member and 0 otherwise; DIST is the distance from the customers home address to the closest offline store. We also include some variables that de scribe transaction characteristics. For example, we include GIFT, which equals 1 if the customer makes the purchase using a gift card and 0 otherwise; ONLINE which equal s 1 if the transaction is made online and 0 otherwise. Return policy change indicator. The focus of this study is to examine the impact of return policy change on customers purchase and return behavior. The retailer restricted the return policy from 90 days to 30 days in Month 54. Accordingly, we construct a binary variable AFT, which equa ls 1 after and 0 before the policy change. We also include the interactions betw een the policy change indicator AFT and other marketing and customer/transaction variables to see whether and how the change of return policy influences the effectiveness of these variables on customers purchase and return behavior. Because of the product cat egories offered by the retailer, Christmas and back-to-school seasons are of special importance. We include variable HOL to control the seasonality.
104 4.4.1 Estimation Results Estimation results are provided in Tables 4-4 and 4-5. Table 4-4 provides the number of parameters, log-likelihoods, and Bayesian in formation criteria (BIC) fit measures for our model with a different number of unobserved types within the population. The two-segment model provides the best fit in terms of BIC and yields reasonable parameter estimates. Table 4-5 provides the par ameter estimates and the standard errors for the two-s egment model. We first look at segment 1 customers, because they account for the majority (67. 7%) of the total population, and then discuss the similarity and difference between them and segment 2 customers. The results show that, after the return po licy becomes more restrictive, customers return faster but do not return significantly less; furthermore, consumers purchase rate decreases significantly after the policy change. This set of results is interesting since an intuitive prediction about the impact of rest ricting the return policy (e.g., shortening the return time limit) is that customers shoul d return less because of the increased hassle cost associated with product returns. Howeve r, the results show that this may not always be the case when customers react by returning faster. A possible explanation is that the hassle cost associated with returning the product faster is smaller than the net loss of keeping a product that would be re turned under the more lenient policy. The negative impact on purchase rate is interest ing and important. Considering the product category (footwear) offered by the retailer, the leniency of the return policy may be viewed as an important signal of the retaile rs service and/or product quality and will not only influence return behavior but also pur chase behavior (Zeithaml et al. 1996). The finding that a more restrictive return policy does not decreas e return rate but decreases the purchase rate is also identified by Wood (2001) using lab experiments. While the
105 experiments mostly focus on the remote purchase environment (i.e., online purchases), the explanation offered ther e, i.e., the endowment effect (customers may like the product better when they are allowed to k eep the product longer with the more lenient return policy), may also be valid in our context. In addition to the return policy, we also identify some interesting effects on marketing variables. For example, a higher price environment leads to a lower purchase rate, a higher return rate and slower product returns. The negative impact on the purchase rate and the positive impact on the retu rn rate are consistent with the existing literature (Anderson 2008). The positive impact on return duration, however, is somehow counter-intuitive since we would expect that higher-priced items be returned more quickly because of the higher financia l cost associated with keeping these products (Hess and Mayhew 1991). The fact that Hess and Mayhew (1991) find this expected negative effect to be insignificant and our results illustrate an opposite positive effect shows that other factors may be involv ed. For example, the customer may like the product itself better when it is purchased at a higher price (e.g., without any price discount) and it may take her a longer perio d of time to decide to not keep it. For consumer and transactional characteristic variables, the results show that gold members buy more frequently and return less, but their return duration is not significantly different from that of regular members. Female customers buy more frequently, return less, and retu rn more slowly compared wit h other customers. These results may be related to the fact that, for the product category the retailer offers, female customers tend to have a higher purchase preference (thus buy more frequently and if they do return, return more slowly) and greater expertise (t hus return less). The age of
106 the customer does not significantly affect the purchase rate, return rate, or return duration under the lenient re turn policy, but older customers purchase less frequently than younger customers when the return poli cy become more restrictive. This shows that older customers respond more negatively when the return policy becomes more restrictive. Intuitively, transactions that are made using gift cards have a significantly lower return rate (Peterson and Kumar 2009), but we do not find that these items are returned more slowly. Furthermore, products purchased from an online store have a significantly higher return rate and are re turned more slowly. This may be explained by the higher risk and transaction cost (e.g., a longer delivery time) associated with online transactions (including returns). For the interactions between the return policy change and other marketing and/or consumer variables, we find that, when the return policy becomes more restrictive, customers become less price sensitive (price discounts are less effective in increasing the purchase rate) but more sensitive to direct marketing efforts (mailed/emailed coupons are more effective in increasing the purchase rate). The salient difference between direct in-store price promotion and emailed coupons is that the latter are often more personalized, e.g., the coupons are often sent out to specific customer segments based on their demographics, RFM characteri stics, and customer preference. The customization effort may compensate for the decrease in service quality (or higher hassle cost associated with product returns) and make the direct marketing promotion activities more effective in encouraging purc hase. The difference between the purchase (and return) rates of gold and regular mem bers increased after the returned policy was
107 restricted. These results suggest that gold members may be less influenced by the change in return policy, e.g., because of a higher loyalty level. Compared with segment 1 cust omers, the results for s egment 2 customers have some similarities and also some significant differences. First, the main effects of the policy change indicator on customers purchas e and return behavior are similar overall. The only difference is that s egment 2 customers return rate increases when the return policy is tightened. For these customers, the leniency of t he return policy may be a very important indicator of the retailers service/product quality, and tightening the return policy may further decrease their willingness to keep the product. Second, the results show that segment 2 customer s purchase and return behavior is more sensitive to the location of a physical store. More specific ally, they purchase more frequently, return more, and return faster when they are located closer to a brick-and-mortar store. Such effects do not exist for the ma jority of segment 1 customers when a lenient policy is offered. However, after the return policy becomes more re strictive, a similar pattern appears for segment 1 customers also. This result indicates that, when the retailer increases the hassle cost associated with produc t returns, other transaction costs, such as travel cost (which is related to the dist ance to the store), may play a more important role in customers purchase and return dec isions. Finally, there is also a notable difference in the shape of the baseline hazard function of the return duration for the two segments of customers. Within the 90-day time interval, t he hazard rate of the return duration for segment 1 custom ers decreases over time, while the hazard rate first increases and then decreases fo r segment 2 cust omers.
108 4.4.2 Validation To validate our model, we used a random sample of 700 customers who were not included in the estimation sample. We compar e the predicted purchase rate, return rate, and return duration wit h the observed values from the holdout sample. For the purchase incidence and return rate validation we also compute the Hit Rate for each observation in the holdout sample. We calculated t he above measures by first obtaining the segment-specific predicted purchase incide nce probability, conditional return rate, and return duration, and then computing the predicted values for each customer by averaging the segment-specific quantities weighted by the households posterior segment probabilities. The expec ted purchase rate, return ra te, and return duration are calculated using the following formula. 1 1()Pr(1) 1im imX imim Xe EII e (4-12) 2 22** **** *(1)Pr(11)1Pr(01) 1Pr(01)Pr(1,1) 1Pr(01)Pr(11)*Pr(1,1) 1 1*(1,1) 11im im im im im it im it itimim imim imim im imim im im imim X im XXEYIYI YI YIYtTI YIYItTIY e STIY ee (4-13) **** 0 ****** 0(1,1)(1,1) (1,1)(1,1)imim im im imim im im im imT imimimim T imimEtIYtftIYdt thtIYStIYdt (4-14) where T is the return time limit. The results are shown in Table 4-6. T he predicted value of the average purchase rate, return rate, and return duration is very close to the observed value. For the
109 purchase incidence, the overall hit rate is around 75.3%, and for the return rate, the overall hit rate is 61.3%. 4.5 Conclusion This paper takes the fi rst step toward ex amining the impact of different return policies on consumer purchase and return behavior using data from a real business context. While we observe considerable variat ion in the leniency of the return policy across different retailers or across different time periods or product categories for the same retailer, how to choose the optimal le vel of leniency is mostly an open question both for practitioners and in the literature (Petersen and Kumar 2009, Anderson et al. 2009). Our analyses reveal that re stricting the return policy by shortening the return time limit creates an overall negative impact for the retailer. Consumers purchase less when the return hassle cost increases, but do not return less; instead, they react to the shortened return time limit by returning faster. These results emphasize the importance for retailers to conduct detailed analyses of consumer behavior that could be influenced directly or indirectly by the return policy when choos ing the leniency level. It is important to note that retailers may change the leniency of the return policy in different ways. For example, the retailer may charge a re stocking fee or require the original receipt. While we only look at one dimension of the retu rn policy, future research may address other potential ways to adjust the return policy leniency. Also, we conduct the study and examine the impact of restricting re turn policies on consumers purchase incidence only at the store level. Furt her research may look at other consumer decision variables such as basket size, and conduct analyses at the product category, brand, or SKU level. Another interesting issue to consider is how the return policy change influences customer relationship matr ices such as customer retention,
110 acquisition, and cross-selling opportuniti es. Furthermore, examining the long-term impact of product returns and the change of return policy may help the retailer to design a return policy that improves not only shor t-term profitability but al so long-term customer equity.
111 Table 4-1. Descriptive statistics Mean Standard Deviation Order Amount $61.68 $41.75 Order Cost $35.43 $25.02 Return Rate 28.1% 0.45 Return Duration 9.18 days 12.23 days Order Frequency 0.19 0.15 Cumulative Orders 4.82 4.35 Cumulative Returns 1.38 0.93 Online Order Rate 6.7% 0.16 Distance 6.82 miles 8.40 miles Age 27.30 18.52 Female 39.9% 0.49 Gold Member 16.0% 0.37 Price $35.71 $2.47 Coupon 39.0% 0.49 Table 4-2. Descriptive stat istics pre vs. post policy change Month 54 Month>54 Mean Standard Deviation Mean Standard Deviation Return Rate 27.4% 0.45 33.3% 0.47 Return Duration 9.72 days 13.22 days 7.27 days 7.43 days Order Frequency 0.23 0.19 0.16 0.24 Table 4-3. Variable definitions Variable Names Definitions A FT 1 before the p olic y chan g e and 0 otherwise. DIST Distance to the closest store. ONLINE 1 if the order is made online and 0 otherwise. PRICE Monthl y p rice index constructed as the wei g hted p rice of all items sold to customers in the lo y alt y p ro g ram. NPRICE Mean-centered value of PRICE COUPON 1 if the customer receives a mailed/emailed cou p on and 0 otherwise. GMEM 1 if the customer is a g old member and 0 a standard member of the lo y alt y p ro g ram FEMALE 1 if the customer is rev ealed as a female and 0 otherwise. A GE Ag e of the customer. HOL 1 if the month is duration holida y s and 0 otherwise. A FT PRICE The interaction between AFT and PRICE. A FT COUP The interaction between AFT andCOUPON. A FT GMEM The interaction between AFT and GMEM. A FT FEM The interaction between AFT and FEMALE. A FT AGE The interaction between AFT and AGE. A FT DIST The interaction between AFT and DIST.
112 Table 4-4. Fit statistics Model Parameters LL BIC Single Segment 45 -53847.9 108205.2 Two Segments 91 -53194.1 107418.2 Three Segments 137 -53070.5 107691.7 Table 4-5. Estimation results Segment 1 Segment2 Coefficient SE Coefficient SE Purchase Incidence Parameters INTERCEPT -2.134*** 0.031 -2.167*** 0.088 AFT -0.421*** 0.078 -0.903*** 0.216 DIST/10 0.022 0.020 -0.102** 0.048 NPRICE -0.223*** 0.021 -0.275*** 0.047 COUPON 0.797*** 0.031 0.687*** 0.068 GMEM 0.647*** 0.038 0.963*** 0.077 FEMALE 0.144*** 0.029 -0.114 0.071 AGE/10 -0.006 0.008 0.020 0.018 HOL 0.378*** 0.034 0.563*** 0.074 AFT_NPRICE 0.074** 0.034 0.162* 0.083 AFT_COUP 0.398*** 0.068 0.660*** 0.183 AFT_GMEM 0.607*** 0.075 0.765*** 0.176 AFT_DIST/10 -0.090** 0.041 -0.252* 0.141 AFT_AGE/10 -0.041** 0.016 -0.035 0.038
113 Table 4-5 Continued Segment 1 Segment2 Coefficient SE Coefficient SE Return Rate Parameters INTERCEPT -0.723*** 0.094 -0.240 0.184 AFT -0.096 0.113 0.504** 0.231 DIST/10 0.061 0.040 -0.230*** 0.089 NPRICE 0.207*** 0.049 0.008 0.087 COUPON -0.198*** 0.068 -0.538*** 0.115 GMEM -0.484*** 0.077 -0.799*** 0.130 FEMALE -0.116* 0.066 -0.355*** 0.129 AGE/10 -0.002 0.014 0.028 0.028 HOL 0.225*** 0.065 0.107 0.124 GIFT -0.654** 0.303 -0.340 0.604 ONLINE 0.407** 0.203 0.759* 0.399 AFT_NPRICE -0.098* 0.058 -0.080 0.129 AFT_COUP -0.050 0.142 0.745** 0.337 AFT_GMEM -0.234** 0.112 -0.249 0.311 AFT_FEM 0.293** 0.131 0.487* 0.260 AFT_DIST/10 -0.157* 0.089 0.000 0.252 Return Duration Parameters AFT 0.177** 0.085 0.386*** 0.140 DIST/10 0.036 0.034 -0.181** 0.086 NPRICE -0.083** 0.042 -0.190*** 0.049 COUPON 0.107* 0.065 0.122 0.082 GMEM -0.121 0.083 0.159* 0.091 FEMALE -0.113** 0.054 -0.263*** 0.088 AGE/10 -0.011 0.014 0.031 0.020 HOL -0.010 0.066 0.014 0.092 GIFT -0.454 0.350 0.254 0.531 ONLINE -0.412** 0.203 -0.910** 0.403 AFT_DIST/10 -0.191** 0.087 0.013 0.190 ALPHA1 -2.740*** 0.147 -0.051 0.265 ALPHA2 -0.054*** 0.139 -0.061** 0.279 ALPHA3 0.341*** 0.106 1.723*** 0.294 ALPHA4/10 0.003** 0.001 -0.003 0.003 Segment size (w) 0.739*** 0.156 ***0.01 p ; **0.05 p ; *0.1 p Note: We estimated the segment size par ameter using a logit formulation such at1exp()/1exp() ww.
114 Table 4-6. Holdout sample validation Actual Predicted Mean Orders 0.162 0.155 Mean Returns 0.292 0.3 Mean Return Duration 8.668 days 8.164 days Hit Rate (Overall) 75.3% Hit Rate (Purchase) 21.7% Hit Rate (Non-purchase) 85.7% Hit Rate (Overall) 61.3% Hit Rate (Return) 35.7% Hit Rate (Non-return) 72.3%
115 CHAPTER 5 CONCLUSION Advances in information technologies create new opportunities and challenges for retailers to coordinate marketing and operat ions manag ement by utilizing customer database and database marketing techniques to customize interface decisions. On the opportunity side, many decisions made by operations managers directly influence consumers purchase behavior in both the shortand the long term. Operations managers, by ex amining the impact and impr oving their understanding of consumer behavior, can improve the quality of their decision making e.g., to decrease the operations cost and/or to increase retu rn on the operations in vestment. Marketing managers, by collecting new dat a and analyzing the new variables that are traditionally managed by other function areas, can identify new marketing mixes to improve customer relationships and increase custom er equity. Chapters 2 and 4 illustrate the idea of using detailed consumer transaction data to improve inventory management and product return managementtwo areas t hat are traditionally managed by the operations department. Many research opportunities exist along this line,. First, the general issue of how retailers can benefit fr om customer database and database marketing techniques to match th e demand and supply is an interesting topic for consideration. For exampl e, in Chapter 2 we propose t hat the retailer may implement customized inventory management by direct ly rationing inventory or indirectly customizing product availability information sent to different customer segments to adjust customer arrival time. Another choice is to customize the price, e.g., charge a premium price when demand is higher than the supply. It will be interesting to study the
116 differences among these tactics and fac ilitate the sellers choice among these alternatives. Second, the management of product returns by integrating the perspectives of relationship marketing and operations managem ent is an important issue. An open and interesting question concerns how retailers can use the information from the product return data to improve customer relations hips, e.g., how to re tain a dissatisfied customer who want to return a product and ev en turn her into a satisfied customer. Third, other operations decisions such as logistic management, shipping and delivery for online retailers, and assortment management may also benefit from the customer database and database marketi ng techniques. Further research might examine the potential value of customized logistic and assortment management systems. Fourth, another important issue concer ns whether the co mpetitive advantage created by customized interface decisions is sustainable. The existing literature on customization generally finds that, when seller s implement customizat ion or targeting on traditional marketing mixes (especially pric e), the benefit disappears when consumer strategic reaction or market competition is considered. It will be interesting to examine similar issues for customized interface or operations decisions. On the challenge side, the main questi on is how consumers react to these customization and targeting activities especially given that many of them can be viewed as discriminatory. Chapter 3 examines c onsumers reactions to customized price promotions. Future research should exami ne how consumers react to other targeting and personalization efforts. For example, t he following issues may be interesting:
117 First, it will be interesting to examine em pirically how consumers react to the personalization/customizati on of other marketing and operations mixes such as personalized/customized service. Despite the importance of the topic and its prevalence in practice, existing literature has insuffici ently examined the relevant issues. Second, while technology has signific antly decreased the cost for firms to implement customization/per sonalization for different marketing and operations management decisions, it is not clear how retailers should choose among different customization tools. For example, should the retailer offer preferential service or exclusive deals to reward better customer s? Should the retailer customize product availability and ration inventory or adjust marketing variables such as price to balance demand and supply? Third, the majority of this dissertation is based on the observation that firms are increasingly capable of collecting individual-level data and using the data to improve decision making, but it is not clear how consumers will react to firms data collection activities. Some important ques tions for future consideration are: 1) How should firms deal with consumers who have privacy c oncerns? 2) How should consumers decide whether or not to reveal personal informa tion? and 3) How should firms decide whether or not to collect certai n specific information?
118 LIST OF REFERENCES Adams, J. (1965), Inequity in Social Exchange, in Advances in ex perimental social psychology 21(2), L. Berkowitz, ed. New York, NY: Academic Press, 267. Agrawal, Narendra and Stephen A. Smith (2009), Retail supply chain management: quantitative models and empirical studies New York, NY: Springer. Amato, Deena (2009), Taking Stock in OOS, Chain Store Age March 2009, 38. Andersen Consulting (1996), Where to Look fo r Incremental Sales Gains: The Retail Problem of Out-of-Stock Merchandise, The Coca-Cola Retailing Research Council, Atlanta. Anderson, Eric and Duncan Simester (2004), Does Promotion Depth Affect Long-Run Demand? Marketing Science 23(4), 4. Anderson, Eric, Gavan Fitzsimons and Duncan Simester (2006), Measuring and Mitigating the Costs of Stock-outs, Management Science 52(11), 1751. Anderson, Eric, Karsten Hansen, and Duncan Simester (2009), The Option Value of Returns: Theory and Empirical Evidence, Marketing Science 28(3), 405. Anderson, Eric, Karsten Hansen, and Duncan Simester, and Lei K. Wang (2008), How Are Demand and Returns Related? working paper, Kellogg School of Management, Northwestern University. Andrade, Eduardo (2005), Behavioral Consequences of Affect: Combining Evaluative and Regulatory Mechanisms, Journal of Consumer Research 32(Dec), 355. Anupindi, Ravi, Maqbool Dada and Sachin Gupta (1998), Estimat ion of Consumer Demand with Stock-Out Based Substitution: An Application to Vending Machine Products, Marketing Science 17(4), 406. Anupindi Ravi, Sachin Gupta, and Venkataramanan M. A. (2009), Managing Variety on the Retail Shelf: Using Household Scanner Panel Data to Rationalize Assortments, in Retail supply chain managemen t: quantitative models and empirical studies, New York, NY: Springer, 155. Bagozzi, Richard, Mahesh Gopinath and Pras hanth Nyer (1999), The Role of Emotions in Marketing, Journal of the Academy of Marketing Science 27(2), 1846. Bell, David, and James Lattin (1998), S hopping Behavior and Consumer Preference for Store Price Format: Why Large Basket Shoppers Prefer EDLP, Marketing Science 17(10), 66. Belobaba, Peter (1989), Application of Probabi listic Decision Model to Airline Seat Inventory Control, Operations Research 37(2), 183.
119 Bitran, Gabriel R. and Susana V. Mondschein (1996). Mailing Decisions in the Catalog Sales Industry, Management Science 42(9), 1364. Black, Sam (2001), E-Fulfillment Be coming a Priority for Retailers, City Business, April 13. Blattberg, Robert and John Deighton ( 1996), Manage Marketing by the Customer Equity, Harvard Business Review July-August, 136. Boatwright, Peter and Joseph C. Nunes (20 01), Reducing Assortment: An A ttributeBased Approach, Journal of Marketing 65(3), 50. Bolton, Lisa, Luk Warlop, and Joseph Alba (2003), Consumer Perceptions of Price (Un)Fairness, Journal of Consumer Research 29(March), 474. Bolton, Ruth (1998), A Dynamic Model of th e Duration of the Customer's Relationship with a Continuous Service Provider : The Role of Satisfaction, Marketing Science 17(1), 45. Borle, Sharad, Peter Boatwright, Joseph B. Kadane, Joseph C. Nunes, and Shmueli Galit (2005), The Effect of Product Assortment Changes on Customer Retention, Marketing Science 24(4), 616. Boston University (2004),Grocery Store Loyalty Card Use is Strong Despite Privacy Concerns, December 28, available at www.i-newswire.com/pr1371.html, Pre ss Release. Bougie, Roger, Rik Pieters and Marcel Z eelenberg (2003), Angr y Customers Dont Come Back, They Get Back: The Exper ience and Implications of Anger and Dissatisfaction in Services, Journal of the Academy of Marketing Sciences 31(Fall), 377. Brown, Stephen, Pamela Homer and Jeffrey Inman (1998), A Meta-analysis of Relationships between Ad-Evoked Feelings and Advertising Response, Journal of Marketing Research 35(February), 114. Business 2.0 (2007), Zappos Has Become the No. 1 Footwear Retailer on the Web by Making Customer Service a Competit ive Weapon, March 15, available at http://money.cnn.com/magazines/business2 /business 2_archive/2006/12/01/83949 93/index.htm Business Week (2000), Why Service Stinks, October 23. Campbell, Margaret (1999), Perceptions of Price Unfairness: Antecedents and Consequences, Journal of Marketing Research, Vol. 36, May, 187. Che, Yeon-Koo (1996), Customer Return Policies for Experience Goods, Journal of Industrial Economics 44(1), 17.
120 Cross, Robert (1997), Revenue management: hard-core ta ctics for market domination New York, NY: Broadway Books. Consumer World (2008), Holiday Return Policies: Some Retailers Naughty, Others Nice, December 15, available at http://www.consumerworld.org/pages/returns.htm. Cox, D.R. (1972), Regressi on Models and Life Tables Journal of the Royal Statistical Society Series B 34(2), 187. Culnan, Mary (1993), How Did They Get My Name? An Exploratory Investigation of Consumer Attitudes Toward Secondary Information Use, MIS Quarterly, 17(3), 341. Darke, Peter and Darren Dahl (2003), Fairness and Discounts: The Subjective Value of a Bargain, Journal of Consumer Psychology 13(3), 328. Davis, Scott, Eitan Gerstner, and Michael Hagerty (1995), Money Back Guarantees in Retailing: Matching Products to Consumer Tastes, Journal of Retailing 71(1), 7 22. Davis, Scott, Michael Hagerty, and Eitan Gers tner (1998), Return Policies and Optimal Levels of Hassle, Journal of Economics and Business 50(5), 445. Dhar, Ravi and Klaus Wertenbroch (2000), Consumer Choice Between Hedonic and Utilitarian Goods, Journal of Marketing Research 37(Feb), 60. Dhar, Sanjay K., Stephen J. Hoch, and Nanda Kumar (2001), Effective Category Management Depends on the Role of the Category, Journal of Retailing 77(2), 165. Dickson, Peter and Alan Sawyer (1990), T he Price Knowledge and Search of Supermarket Shoppers, Journal of Marketing 54(July), 42. Dowling, Grahame and Mark Uncles (1997), Do Customers Loyalty Programs Really Work? Sloan Management Review Summer, 71. Dreze, Xavier and Stephen J. Hoch (1998), Exp loiting the Installed Base Using CrossMerchandising and Category Destination Programs International Journal of Research in Marketing 15(5), 459. Dreze, Xavier and Joseph Nunes (2008), F eeling Superior: The Impact of Loyalty Program Structure on Consumers Perceptions of Status, Journal of Consumer Research, 35(April), 890. Egan, Cathleen (2001), Vending-Machine Te chnology Matures, Offering Branded Foods, Convenience, Wall Street Journal December 13, B13.
121 Evan, Porteus (2002), Foundations of stochast ic inventory theory Stanford, CA: Stanford Business Books. Fader, Peter and Leonard Lodish (1990), A Cross-Category Analysis of Category Structure and Promotional Acti vity for Grocery Products, Journal of Marketing 54(4), 52. Fader, Peter, Bruce Hardie and Ka Lok Lee (2005), Counting Your Customers the Easy Way: An Alternative to the Pareto/NBD Model, Marketing Science 24(2), 275 284. Feinberg, F., Aradhna Krishna, and Z. John Zhang (2002), Do We Care What Others Get? A Behaviorist Approach to Targeted Promotions, Journal of Marketing Research, 39(3), 277. Fitzsimons, Gavin (2000), Consum er Response to Stock-outs, Journal of Consumer Research, 27(2), 249. Forbes (1999), Santa Flaws, December 27. Forbes (2006), Out Of Sto ck, December 14. Fournier, Susan (1998), C onsumers and Their Brands: Developing Relationship Theory in Consumer Research, Journal of Consumer Research volume 24 (March), 343. Fudenberg, Drew, and Tirole (1998), Upgrades, Trade-ins, and Buybacks, RAND Journal of Economics 29(2), 235. Fudenberg, Drew, and Tirole (2000), C ustomer Poaching and Brand Switching, RAND Journal of Economics 31(4), 634. Gaur, Vishal and Young-Hoon Park ( 2007), Asymmetric Consumer Learning and Inventory Competition, Management Science 53(2), 227. Grocery Manufacturers of Am erica (2002), Full-shelf Satisfaction Reducing Out-ofStocks in the Grocery Channel: An In-Depth Look at DSD Categories. Grocery Headquarters (2006), Holes on t he Shelf, April 10, available at www.groceryheadquarters.com Gupta, Sunil and Do nald Lehmann (2005), Managing Cu stomers as Investments: The Strategic Value of Customers in the Long Run Philadelphia, PA: Wharton School Press. Gustafsson, Anders, Michael Johnson and Inger Roos (2005), The Effects of Customer Satisfaction, Relationship Commitment Dimensions, and Triggers on Customer Retention, Journal of Marketing, 69(4), 610.
122 Hall, Joseph and Evan Porteus (2000), Customer Service Competition in Capacitated Systems, Manufacturing & Service Operations Management 2(2), 144. Hamilton, David (2001) E-Commerce S pecial Report: The Price Isnt Right, Wall Street Journal, Feb. 12, R.8. Haws, Kelly and William Bearden (2006), D ynamic Pricing and Consumer Fairness Perceptions, Journal of Consumer Research 33(Dec.), 304. Heskett, James, Earl Sasser and Leonard Schlesinger (1997), The Service Profit Chain. New York, NY: Simon and Schuster. Hess, James D. and Glenn E. Mayhew (1997), Modeling Merchandise Returns in Direct Marketing, Journal of Direct Marketing 11(2), 20. Hess, James, Wujin Chu, and Eitan Gerstn er (1996), Controlling Product Returns in Direct Marketing, Marketing Letters 7(4), 307. Ho, Teck H. and Christopher S. Tang (2004). Introduction to the special issue on marketing and operations management interfaces and coordination, Management Science 50(4), 429. Homans, G. (1961), Social behavior: its elementary forms New York, NY: Harcourt, Brace & World. Hoch, Stephen J. (2002), How effe ctive is category management? ECR Journal 2(1), 26. Hughes, Arthur (2000 ), Strategic database marketing. New York, NY: McGraw Hill. Iyer, Ganesh, David Soberm an, and J. Miguel Villas-Boas (2005), The Targeting of Advertising, Marketing Science 24(2), 461. Jing, Xiaoqing and Michael Lewis (2010) Stockouts in Online Retailing, Journal of Marketing Research Forthcoming. Jones, Mary (1991), Privacy: A Signif icant Marketing Issue for the 1990s, Journal of Public Policy and Marketing 10(1), 133. Kahneman, Daniel, Jack L. Knetsch and Richard Thaler (1986), Fairness as a Constraint on Profit Seeking Entitlements in the Market, American Economic Review, 76(40), 728. Kahneman, Daniel, Kamakura, Wagner and Gary Russell (1989), A Probabilistic Choice Model for Market Segmentation and Elasticity Structure, Journal of Marketing Research 26(4), 379.
123 Khan, Romana, Michael Lewis and Vis hal Sing (2009), Dynamic Customer Management and the Value of One-to-One Marketing, Marketing Science forthcoming. Kimes, Sheryl (1994), Perceived Fairness of Yield Management, Cornell Hotel and Restaurant Administration Quarterly 35, 22. Kivetz, Ran and Itamar Simonson (2002), Earni ng the Right to Indul ge: Effort as a Determinant of Customer Preferences toward Frequency Program Rewards, Journal of Marketing Research 39(2), 155. Kivetz, Ran (2003), The Idiosyncratic Fit Heur istic: Effort Advantage as a Determinant of Consumer Response to Loyalty Programs, Journal of Marketing Research 40(4), 454. Kivetz, Ran, Oleg Urminsky and Yuhuan Zheng (2006), The Goal-Gradient Hypothesis Resurrected: Purchase Acceleration, Illusionary Goal Progress, and Customer Retention, Journal of Marketing Research 43(1), 39. Lal, Rajiv and David Bell (2003), The Impact of Frequent Shopper Programs in Grocery Retailing, Quantitative Marketing and Economics 1, 179. Lam, Shun, Mark Vandenbosch, John Hulland, and Michael Pearce (2001), Evaluating Promotions in Shopping Environments: Decomposing Sales Response into Attraction, Conversion, and Spending Effects, Marketing Science 20(2), 194. Lewis, Michael (2004), The Influence of Loyal ty Programs and ShortTerm Promotions on Customer Retention, Journal of Marketing Research 41(3), 281. Lewis, Michael (2005), Research Note: A Dynamic Programming Approach to Customer Relati onship Pricing, Management Science, 51(6), 986. Littman, Alan (2002) The Technology Split in Customer List Interpretation, The University of Chicago Law Review 1901. Liu, Yuping (2007), The long-Term Impact of Loyalty Pr ograms on Consumer Purchase Behavior and Loyalty, Journal of Marketing 71(October), 19. Mano, Haim and Richard Oliver (1993), Assessi ng the Dimensionality and Structure of the Consumption Experience: Evaluat ion, Feeling and Satisfaction, Journal of Consumer Research 20(3), 451. Moorthy, Sridhar and Kannan Srinivasan (1 995), Signaling Quality with a Money-Back Guarantee: The Role of Transaction Costs, Marketing Science 14(4), 442. Neslin, Scott (2004), Sales Promotion Cambridge, MA: Marketi ng Science Institute.
124 Noone, Breffni M., Sheryl E. Kime s and Leo M. Renaghan ( 2003), Integrating Customer Relationshi p Management and Revenu e Management: A Hotel Perspective, Journal of Revenue and Pricing Management 2(1), 7. Nunes, Joseph and Xavier Dreze (2006), The Endowed Progress Effect: How Artificial Advancement Increases Effort, Journal of Consumer Research 32(4), 504. Oliver, Richard and John Swan (1989), Consumer Perceptions of Interpersonal Equity and Satisfactions in Transactions: A Field Survey Approach, Journal of Marketing 53(April), 21. Ordonez, Lisa, Terry Connolly and Richard Cough lan (2000), Multiple Reference Points in satisfaction and Fairness Assessment, Journal of Behavioral Decision Making 13(3), 329. Padmanabhan, V. and I.P.L. P ng (1997), Manufacturers Returns Policies and Retail Competition, Marketing Science 16(1), 81. Pazgal, Amit and David Soberman (2008), B ehavior-Based Discrimination: Is It a Winning Play, and If So, When? Marketing science 27(6), 977. Petersen, J. Andrew and V. Kumar (2009), Are Product Returns a Necessary Evil? Antecedents and Consequences, Journal of Marketing, 73(3), 35. Peppers, D. and M. Rogers (1993), The one to one future New York, NY: Doubleday. Phelps, Jospeh, Glen Nowak, and Elizab eth Ferrell (2000), Privacy Concerns and Consumer Willingness to Provide Personal Information, Journal of Public Policy and Marketing 19(1), 27. Porteus, Evan (2002), Foundations of stochast ic inventory theory Stanford, CA: Stanford University Press. Roehm, M. E. Pullins, and H. Roehm (2002), Designing Loyalty-Building Programs for Packaged Goods Brands, Journal of Marketing Research, May, 202. Rust, Roland, Katherine Lemon and Valarie Zeithaml (2004), Return on Marketing: Using Customer Equity to Focus Marketing Strategy, Journal of Marketing 68(1), 109. Rust, Roland and Tuck Siong Chung (2006), Marketing Models of Service and Relationships Marketing Science 25(6), 560. Schmittlein, David, Donal d G. Morrison and Richard Colom bo (1987), Counting Your Customers: Who Are They and What Will They Do Next? Management Science 33(1), 1.
125 Sharp, Byron and Anne Sharp (1997), Loy alty Programs and their Impact on RepeatPurchase Loyalty, International Journal of Research in Marketing 14, 473. Shugan, Steven and Ramarao Desiraju (1999), Strategic Service Pricing and Yield Management, Journal of Marketing 63(1), 44. Simester, Duncan, Peng Sun an d John Tsitsiklis (2006), Dynamic Catalog Mailing Policies, Management Science 52(5), 683. Sinha, Indrajit and Rajeev Ba tra (1999), "The Effect of Consumer Price Consciousness on Private Label Purchase," International Journal of Research in Marketing ,16 (3), 237. Sloan Management Review (2006), Managing Product Re turns for Competitive Advantage, October 1, available at http://sloanreview.mit.edu/themagazine/articles/2006/fa ll/48112/managing-product-returns-for-competitiveadvantage/. Sloot, Laurens, Peter Verhoef and Philip Ha ns Franses (2005), The Impact of Brand equity and the Hedonic Level of Products on Consumer Stock-Out Reactions, Journal of Retailing 81(1), 15. Smith, Stephen (2009a), Optimizing Retail Assortments for Diverse Customer Preferences, in Retail supply chain managem ent: quantitative models and empirical studies, New York, NY: Springer, 183. Smith, Stephen (2009b), Clearance Pricing in Retail Chains, in Retail supply chain management: quantitative models and empirical studies New York, NY: Springer, 271. Stevens, Amy (1996), Tempest in a C Cup: W hats Underneath Victoria s Secret Suit, Wall Street Journal 227(86), PA1. Talluri, Kalyan and Garrett Van Ryzin (2004), The theory and practice of revenue management. Boston, MA: Kluwer Academic Publishers. Thaler, Richard (1985), Mental A ccounting and Consumer Choice, Marketing Science 4(3), 199. Tice, Dianne, Ellen Bratslavsky and Roy Ba umeister (2001), Emotional Distress Regulation Takes Precedence over Impulse Control: If You Feel Bad, Do It! Journal of Personality and Social Psychology 80(January), 53. Urbany, Joel E, Peter R. Dickson, and William L. Wilkie (1989), B uyer uncertainty and information search, Journal of Consumer Research 16(2), 208. Villas-Boas, Miguel J. (1999) Dynamic competition wit h customer recognition, Rand Journal of Economics 30(4), 604.
126 Villanueva, Julian, Shijin Yoo and Domi nique M. Hanssens (2008), The Impact of Marketing-Induced Versus Word-of-Mout h Customer Acquisition on Customer Equity Growth, Journal of Marketing Research, 45(1), 48. Wall Street Journal (2008), The War on Returns, May 8. Walters, Rockney and Scott MacKenzie (1988), A Structural Equations Analysis of the Impact of Price Promotions, Journal of Marketing Research 25(1), 51. Wangenheim, Florian v. and Tomas Bayon (2007), Behavioral Consequences of Overbooking Service Capacity, Journal of Marketing 71(4), 36. Wood, Joanne (1989), Theory and Research C oncerning Social Comparisons of Personal Attributes, Psychological Bulletin 106(2), 231. Wood, Stacy L. (2001), R emote Purchase Environments: The Influence of Return Policy Leniency on Two-Stage Decision Processes, Journal of Marketing Research, 38 (May), 157. Xia, Lan, Kent Monroe and Jennifer Cox (2004), The Price Is Unfair! A Conceptual Framework of Price Fairness Perceptions, Journal of Marketing 68(October), 1 15. Zeithaml, Valarie A., Leonard L Berry, and A. Parasuraman (1996), The Behavioral Consequences of Service Quality, Journal of Marketing 60 (April), 31. Zhang, Jie and L. Krishnamurthi (2004), Customizing Promotions in Online Stores, Marketing Science 23(4), 561. Zipkin, Paul (2000), Foundations of inventory management Boston, MA: Irwin McGrawHill.
127 BIOGRAPHICAL SKETCH Xiaoqing Jing received her Ph.D. in market i ng from the University of Florida in 2010. She also holds a bachelors degree in mechanics, a second bachelors degree in economics, and a master's degree in management science from Beijing University, China. Xiaoqing's research interests lie in the marketing/operations management interface, customer relationship managem ent, and supply chain management. She is also interested in marketing strategies for managi ng consumer social interactions.