Online order fulfillment failures and subsequent shopping behavior

(1)

Online order fulfillment failures and subsequent shopping behavior

Jantina Henrieta Steen

Rijksuniversiteit Groningen

(2)

Online order fulfillment failures and subsequent shopping behavior

Master thesis

Jantina Henrieta Steen

Rijksuniversiteit Groningen

Faculty of Economics and Business

MSc Business Administration – Operations and Supply chains

First reader: Prof. Dr. K.J. Roodbergen

(3)

MSc Thesis J.H. Steen August 31st, 2012 Page i

Abstract

The purpose of this research is to analyze the impact of online order fulfillment failures on actual shopping behavior of customers. The internet as a channel for distribution of goods is growing significantly. This means that more often customers order their products online. Since this channel is growing, customers can easily compare offerings of internet retailers in order to get the best delivery options. This competition puts pressure on the internet retailers to improve their logistical

performances. In order to research the impact of order fulfillment failures on actual shopping behavior two separate groups of customers are analyzed. All customers need to have bought at least once from the online retailer. Either experiencing a failure or received their goods on the promised delivery date. The differences between these two groups can give insight into the impact of order fulfillment failures. Secondly, this research does not focus solely on failures but also on the order magnitude. Depending on when a late order arrives at the destination it can have differential impact on the actual shopping behavior. Lastly, e-purchasing experience is taken as a moderating variable on the relationship between fulfillment failures and subsequent shopping behavior.

Two main methods of research are used to analyze the data, Change Score Analaysis (CSA) and regression analysis. Conducting a CSA analysis includes performing two steps. Step 1 is to test the differences per group by means of an unpaired t-test. The outcomes of this test show how the groups behave separately. The second step is to test if a significant difference is found between the two groups. This will tell whether a fulfillment failure impacts the shopping behavior of customers. Regression analysis will tell if a significant relationship is found between the magnitude of failure and subsequent shopping behavior. In the second stage of this regression analysis e-purchasing

experience will be taken as a moderating variable.

The first and most important conclusion from this research is that there are no significant differences found between the two groups. This means that when customer experience a failure or when their order is received on the promised delivery date their shopping behavior remains the same. Different from what is found in literature, the outcome in the current study were completely different. Secondly, the magnitude of failure did not have a significant relationship with order frequency or order size after failure. E-purchasing experience has a significant relationship with order frequency after failure for Company A but not for Company B. Moreover, the regression analysis shows a significant relationship between e-purchasing experience and order size after failure for both companies. E-purchasing experience did not have a moderating effect on the relationship between magnitude of failure and subsequent shopping behavior.

(4)

MSc Thesis J.H. Steen August 31st, 2012 Page ii

Preface

This report is my graduation thesis that I performed for my study MSc Operations and Supply Chain. Over the period of 6 months I have investigated the problem of fulfillment failures and their impact on actual shopping behavior of customers. Data for this research was provided by two internet retailers. Both companies gave me the opportunity to visit their warehouses to improve my

understanding of the logistical operations that are involved in this channel of distribution. This gave me some essential insights that were very helpful in executing the research. Since, both companies are not mentioned by name, via this way I want to give them special thanks for all their inputs to the research.

From the University of Groningen a special thanks goes to Prof. Dr. K.J. Roodbergen for his guidance and support during this process. I want to thank him for his assisting and supervising role in

successfully concluding this thesis. Secondly, I would like to thank my second supervisor Dr. Tudor Bodea for all the feedback and useful input for my thesis.

At last, I would like to thank everybody else who was not mentioned by name but supported me during this period. I want to thank them for their advice, support and critical comments when writing my thesis.

J. Henrieta Steen

(5)

MSc Thesis J.H. Steen August 31st, 2012 Page iii

Content

Abstract ... i Preface ...ii Content ... iii List of Tables ... iv List of Figures ... iv 1. Introduction ... 1 1.2 Research questions... 2 2. Literature research ... 3 2.1 E-purchasing experience ... 4 3. Hypotheses development ... 6

3.1 Order frequency and order size ... 6

3.2 Failure magnitude ... 6 3.3 E-purchasing experience ... 7 3.4 Conceptual model ... 8 4. Methodology ... 9 5. Results ... 12 5.1 Results Company A ... 12 5.2 Results Company B ... 17 6. Discussion ... 21

6.1 Discussion company A & B ... 21

6.2 Comparison Company A & B ... 22

7. Conclusion ... 23

7.2 Limitations and future research ... 24

(6)

MSc Thesis J.H. Steen August 31st, 2012 Page iv

List of Tables

Table 1 Descriptive statistics of the treatment group Page 12

Table 2 Descriptive statistics of the control group Page 12

Table 3 Statistic outcomes of the paired t-test Page 13

Table 4 Statistic outcomes of the unpaired t-test Page 13

Table 7 Statistic outcomes of paired t-test Page 14

Table 8 Statistic outcomes of the unpaired t-test Page 15

Table 9 Resulting levels of significance from regression analysis Page 15 Table 10 Resulting levels of significance from regression analysis Page 16 Table 11 Resulting levels of significance from regression analysis Page 16

Table 14 Statistic outcomes of paired t-test Page 18

Table 15 Statistic outcomes of unpaired t-test Page 18

Table 16 Resulting levels of significance from regression analysis Page 19 Table 17 Resulting levels of significance of regression analysis Page 19 Table 18 Resulting levels of significance from regression analysis Page 20

List of Figures

(7)

MSc Thesis J.H. Steen August 31st, 2012 Page 1

1. Introduction

Nowadays the use of Internet as a channel for the sale and distribution of goods from business to customers (B2C) has grown significantly. The internet provided business with a new channel of distribution. Before, customer bought their product in a traditional brick-and-mortar shop and brought the product home themselves. In the channel of e-commerce the delivery of products is the responsibility of the e-tailer. Moreover, many web shops are available on the internet and customers can compare offerings of sellers worldwide. This makes it difficult to achieve customer retention since customers can easily switch to other retailers. The above issues lead to many new challenges for logistics decision makers in B2C companies that sell over the internet (Esper et al., 2003). When customers order a product online they expect that the product will be delivered on the promised delivery date. These deliveries of products to the customers, also called online order fulfillment, are important for the business to keep customers satisfied. As said earlier, this new channel of distribution leads to many new challenges for the e-tailers. Either the product can be delivered upon the agreed day or customer can experience a failure in the delivery of their order. Both, effective and ineffective order fulfillment can have differential effects on customer behavior. If customers experience a delivery failure they can become dissatisfied with the e-tailer and switch easily to one that will deliver on time. If customer gain more experience with e-purchasing, which makes them become more familiar with the internat as a shopping channel, they feel more in control. This experience can lead to either a loyal customer that will stay with the same e-tailer or a bad experience which leads to customers switching to other e-tailers.

Previous research focuses on the relationship between successful order fulfillment and customer behavior. In service failure research, many different authors have concluded that positive and negative outcomes related distinctly to satisfying and dissatisfying experiences (e.g., Bitner, 1990; Boulding et al., 2003; Kelley et al., 1993). Examination of the relationship between ineffective order fulfillment and customer behavior is important as it holds the potential to evoke a customer reaction more complex than merely the inverse of successful service engagement (Rao et al., 2011). The above leads us to the main objective of this paper.

The principal objective of this paper is to examine the relationship between online order fulfillment failures and subsequent shopping behavior in an online retailing environment through CSA.

(8)

1.2 Research questions

1. What previous research is done on the topic online order fulfillment?

2. What factors influence the relationship between order fulfillment failures and subsequent shopping behavior?

3. What research methods did previous studies use to research order fulfillment failures? 4. What alternative research methods are suitable for the research?

5. Does an order fulfillment failure decrease order frequency and order size? 6. Does the magnitude of failure impacts order frequency and order size?

7. Does e-purchasing experience mediates the relationship between order fulfillment failure and subsequent shopping behavior?

In order to examine the relationship between order fulfillment failures and subsequent shopping behavior two sets of data are provided by two online retailers. The data includes orders of customers who experienced an order fulfillment failure, and orders which are delivered at the agreed upon time. The data tracks the shopping behavior of a 12 month period (six months before the failure and six months after the failure). To see what influence the failure has on future purchasing behavior, the data is compared to a group of customers who did not experience an order fulfillment failure. Secondly, data will be gathered on the e-purchasing experience of customers who bought from the online retailer in the past. This data will give insight into the moderating effect of e-purchasing experience on the relationship between online order fulfillment failures and subsequent shopping behavior.

(9)

2. Literature research

In this chapter the current literature is reviewed to provide insight into the studies that have been done previously on the topic of order fulfillment failures. Since, literature on order fulfillment failures and actual shopping behavior is limited other studies that discuss important aspects are included as well. This literature research defines the area of research and how the current study fits into this. First, the relationship between order fulfillment failures and subsequent shopping behavior is discussed. Secondly, the e-purchasing experience, as one of the variables on the relationship between order fulfillment failures and subsequent shopping behavior, is discussed separately. It used to be that location was one of the key determinants of success for the bricks-and-mortar retailers, but for online retailers location became less important. While location remains a key determinant of success for a conventional bricks-and-mortar retailer, the importance of location in an online context is substantially reduced (Esper et al., 2003). Rather than location, online retail success often hinges upon how efficiently and effectively the online retailer is able to deliver products to the end customer (Alba et al., 1997).The main focus of online retailers is the delivery of products to end consumers or also called online order fulfillment. In fact, it has been argued that order fulfillment is the most critical operation for Internet retailers and that those online retailers who outperform the competition in this regard have much to gain (De Koster, 2003; Grewal et al., 2004). Studies have found that customers generally considered physical delivery as a very important factor (Esper et al., 2003).This brings new aspects to the business that determine the key success factors of online retailers.

One of these key success factors is the logistics performance of a firm in the e-commerce market. The study of Cho et al. (2008) examines the impact of logistics capability and logistics outsourcing on firm performance in an e-commerce market environment. The results of the study reveal that logistics capability is positively associated with firm performance in the computer and consumer electronics retail industry. The study of Ramanathan (2010) explored how the relationship between logistics performance and customer loyalty are affected by risk characteristics of products and efficiencies of the websites. Their results show that efficiency, but not risk, is a significant moderator of the impact of logistics performance on customer loyalty. The above mentioned studies, from Cho et al. (2008) and Ramanathan (2010), both seek a relationship between logistics performance and customer loyalty. These studies give a basic understanding of what consumers consider important when buying from internet retailers.

(10)

MSc Thesis J.H. Steen August 31st, 2012 Page 4 The study of Bejou & Palmer (1998) investigates the effects of service failure on relationship

breakdown within the context of a buyer-seller relationship life cycle. In their research they used a questionnaire and face-to-face interviews in order to collect data. Their results indicate that any given level of service failure resulted in reduction in commitment and trust which was dependent on the duration to date of their relationship.

The main purpose of the study of Wang et al. (2011) was to explore the relationship of service failure severity, service recovery justice and perceived switching costs to customer loyalty, and the

moderating relationship of service recovery justice and perceived switching costs on the link between service failure severity and customer loyalty in the context of e-tailing. In this particular study service recovery justice is defined as the attempt to correct a service failure (Kelley et al., 1993). Their results indicate that service failure severity was one of the factors that have a significant relationship with customer loyalty.

The above studies focus on failures in order fulfillment and the impact these failures have on the customer loyalty. Only a few studies have researched the effect of order fulfillment failure on the actual shopping behavior. In this research this relationship will be tested in a different market segment and on fulfillment failures with higher magnitude compared to previous research. These results can also be compared among different businesses which can give a more broad perspective on the effect of fulfillment failures on subsequent shopping behavior. In this study, the moderating effect of e-purchasing experience on the relationship between online order fulfillment failures and subsequent shopping behavior will be researched. The literature on e-purchasing experience will be discussed in the next paragraph.

2.1 E-purchasing experience

Researchers in several fields have evaluated the role of experience in traditional purchase situation. They conclude that consumers update their expectations and perceptions over time and

continuously blend prior beliefs with new information (Boulding et al., 1993).

Kuan et al. (2008) examined the relationship between website quality and customer retention. The framework of Kuan et al. (2008) demonstrates the impact of website quality on intention to purchase on the Web. Different quality constructs exerts a different impact on intention of initial purchase and intention of continued purchase. Research has hardly considered the differences generated by customers' e-purchasing experiences on their behavior because it does not usually differentiate between initial and continued purchasing intentions (Kuan et al., 2008).

(11)

(12)

3. Hypotheses development

This chapter will develop hypotheses that will help to answer the objective of this research. The focus of this research will be on two post-purchase customer behavior measures. Both these measures, order frequency and order size, are economic. The psychological measures are not included in this research. Since research on actual shopping behavior in an online environment after a delivery failure is limited, research on logistics performance and service failures provides more insights.

3.1 Order frequency and order size

Several researchers (e.g., Bitner et al., 1990; Boulding et al., 1993; Kelley et al., 1993; Smith et al., 1999; McCollough et al., 2000;) show that when customers are satisfied (dissatisfied) with the service encounter, this leads to greater (lower) perceived service quality, which, in turn, leads to repeat (reduced) purchase activity. Consequently, we expect that after customers experience an order delivery glitch their perception of an unequal distribution of benefits will cause them to take action to restore balance between themselves and the retailer (Oliver and Swan, 1989; Kelley and Davis, 1994). One way in which customers compensate for such imbalances is by reducing their business with the firm.

Some studies also suggest that service failure acts as one significant motivator in customer switching behavior (McCollough, 2000; Roos, 1999). Moreover, Bolton (1998) found that consumer perceptions of losses experienced during transactions will reduce the customer relationship duration. As found in the service failure literature customers which are satisfied with the service encounter repeat

purchase activity. In this research we argue that in an online retailing environment this will be similar. If customers experience a failure in order fulfillment which leads to dissatisfaction with the service encounter, this leads to reduced purchase activity. The above observations lead to the following hypotheses.

H1. When customers experience a delivery failure their order frequency will decrease. H2. When customers experience a delivery failure their order size will decrease.

3.2 Failure magnitude

Goodwin and Ross (1992) showed that customers perceive the loss from a large service failure as higher than that from a smaller failure. Similar results have been demonstrated in a restaurant context by researchers who demonstrated that the magnitude of failure in a prior service’s

(13)

MSc Thesis J.H. Steen August 31st, 2012 Page 7 Different to previous research this study suggests that order fulfillment failure magnitude has a differential negative relationship with the actual shopping behavior of customers. In this study not the psychological elements like loyalty are measured but economic elements. With analyzing data instead of using a survey actual shopping behavior subsequent to a failure can be tracked. These arguments lead to the following hypotheses.

H3. Online order fulfillment failure magnitude has a differentiated negative relationship with order frequency.

H4. Online order fulfillment failure magnitude has a differentiated negative relationship with order size.

3.3 E-purchasing experience

Past work has shown that customers are more forgiving of service failures by a firm as the length of their relationship with the firm increases (Anderson and Weitz, 1989; Bitner and Hubbert, 1994). Loyal customers appear to be careful to base long-term consumption behavior with the firm on multiple experiences rather than a single transaction (Folkes, 1984; Hess et al., 2003).As the number of past service encounters increases for a particular customer, that customer becomes more familiar with the service provider, its offerings, and its processes (Solomon, Surprenant, Czepiel, and Gutman 1985). As a result, this customer perceives less risk when purchasing the service than do customers possessing less experience with the organization. Due to the growing number of encounters the risk is reduced which should increase the customer’s desire to continue the relationship. Similar to this is that it has been argued that a failure occurring early in the customer’s relationship with a supplier will be perceived more adversely than one which occurs later in a relationship because the customer has less experience of successful service experiences to counterbalance the failure (Boulding, 1993). From the above service failure literature we find that customer are more forgiving of service failures by a firm if the length of the relationship increases. Moreover, failures that happen early in a

customer relationship are more adversely. In this research we assert that in an online retailing environment the process will be similar. If e-purchasing experience increases it will positively affect the shopping behavior of customers. Since the customer has experience more successful service fulfillment to counterbalance the failure. According to the above assumptions it is important to see the effect of the e-purchasing experience of customers. These arguments drive us to the following hypothesis:

H5. E-purchasing experience positively affects the relationship between the magnitude of failure and order frequency.

(14)

3.4 Conceptual model

To give a better insight into the variables that are studied, figure 1 is developed. This model visualizes the relations that are tested in the research, as already described in the previous paragraphs.

Hypotheses one and two are not in this picture since these only test the differences between the two groups. The figure shows the relation between failure magnitude and the expected decrease in order frequency and order size. This relationship is described in hypothesis 3 and 4. Furthermore,

e-purchasing experience is depicted as a moderating variable. This visualizes the possible moderating effect of e-purchasing experience on the relationship between failure magnitude and decrease in order frequency and order size. When proceeding to the results section of the report figure 1 would clarify the discussion.

Figure 1

Summary of the research model

(15)

4. Methodology

In this chapter of the report describes the method of research into more detail. First, an overview is given on how the current research is performed. Secondly, the method of research for each of the hypotheses developed in chapter 3 is described from literature. After given a brief overview of the description in literature the approach of the current study will be described. This will provide guidance for the analysis of the results.

To examine the relationship between order fulfillment failures and future customer purchasing behavior, a data set with actual shopping behavior is used. Both companies promised a next day delivery if the product was ordered before 22.00 the previous day. This means a next day delivery for all products that are on stock. In case the product cannot be delivered from stock then it depends on the product what delivery time it has. Rao et al. (2011) defined delivery failure as “a failure to deliver an order within a previously promised time limit” (p. 696). In the current study a delivery is

considered a failure when the promised delivery date is not met or those orders that were in the system longer than 24 hours. Severity of failure can be defined as the magnitude of loss that

customers experience due to the failure. Such losses can be either tangible (e.g., a monetary loss) or intangible (e.g., anger, frustration) (Smith et al., 1999). In this study, the magnitude of failure per order can be measured according to the number of days late orders were past the promised delivery date. One customer can experience more than one failure during the month November 2011. Those customer numbers are excluded from the research since the focus is on the impact of a failure on actual shopping behavior.

The data as described above is obtained from two online retailers and tracks the shopping behavior of those customers who experienced a delivery failure delivery over a twelve-month period (six months before failure and six months after failure). The data from the six months period before the failure is compared to the data of the six months after a failure. Shopping behavior will be measured by two post-purchase customer behaviors. The first, order frequency can be measured as the number of times in a given time period that a customer places an order (Waller et al., 1999). In the current study we measure the number of orders that are placed 6 months prior to the online order fulfillment failure and 6 months after the failure. Secondly, order size has been operationalized Lewis et al. (2006) as the average dollar value of orders placed per purchase. In this study, the order size is measured as the average value per purchase over the six months periods being examined.

To analyze the influence a failure has on future purchasing behavior, the data is compared to a group of customers who did not experience an order fulfillment failure. Additionally, data will be gathered on the e-purchasing experience of customers of shopping at the online retailer in the past. Adoption can be defined as the first purchase carried out by a potential customer from a website, and

repurchasing behavior as the return to the channel for shopping purposes (Davis, 1989; Kuan et al., 2008). In the current study, e-purchase experience is the number of times the customer placed an order with the e-tailer before the month November 2011. This data will be gathered in order to test the moderating effect of e-purchasing experience on the relationship between online order

fulfillment failures and subsequent shopping behavior. This can give more insight in whether

(16)

MSc Thesis J.H. Steen August 31st, 2012 Page 10 The research consists of three different parts. Each part of the study first was described from

literature and then described as applied in the current study.

1. Estimating the effect of the intervention (failure in order fulfillment) requires a comparison between what happened after the intervention was implemented and what would have happened if the intervention had not been happened. In randomized experiments, different interventions are assigned to individuals at random. In this experiment intervention conditions are not assigned to units at random. This makes this design a quasi-experiment. To analyze the differences between the two groups a changed score analysis (CSA) is used, which is recommended for quasi-experimental designs (Rao et al., 2011; Achen, 1986). In CSA the dependent variable is measured as the post-failure order frequency minus the pre failure order frequency. CSA involves comparing the change in the dependent variable’s score for the treatment group for the time periods. The same is calculated for the control group. Both of these changes can be calculated by using a paired sample t-test. The scores for these two groups can be compared using an unpaired t-test (Kenny, 1975). The next paragraph explains step wise how this analysis will be performed.

1. Hypothesis one and two are analyzed according to the changed score analysis (CSA) as described in the above literature. Step 1 is to compare the change in shopping behavior six months prior to the failure and six months after failure. It is tested if there is a significant difference between the two periods in order frequency. The same is done for order size. These changes will be tested by means of a paired sample t-test. Step 2 is to compare the treatment group with the control group by means of an unpaired t-test. This will test the significance difference between the two groups for and after failure. If a significant difference is found between the two groups, then it can be concluded that one group changed more than the other group in the given time period. This can help to answer the question whether order fulfillment failures influence future purchase behavior and if the magnitude of order fulfillment failures also changes subsequent purchase patterns.

2. Most users of hierarchical multiple regression analysis, also called moderated regression analysis (Saunders. 1956; Zedeck, 1971), have used the incremental R2 to assess the magnitude of the interaction effects of their moderator variables. In moderated regression, a dependent variable is regressed on an independent variable, a moderator variable, and a product term between the independent and the moderator variables (Hair et al., 2006). The impact of the moderator variable is assessed using a two stage regression (Sanchez and McKinley, 1998; Li et al., 2001). In the first stage, the dependent variable is regressed with the independent variable, moderator variable and control variables (if any). In the second stage, a product term (independent*moderator variable) is added. The impact of moderator is assessed based on the improvement in R2 in the second stage regression over the first stage. If this change is statistically significant, then a significant moderator effect is predicted (Hair et al., 2006).

Hypothesis three and four will be tested by means of regression analysis. In the same time this is the first stage of the two stage regression analysis for hypotheses five and six. First the dependent variable, in this case the magnitude, is regressed with the independent variable (order frequency and order size). This regression analysis will test whether the magnitude of failure can explain the

(17)

MSc Thesis J.H. Steen August 31st, 2012 Page 11 3. In the second stage of the regression analysis the moderating effect of e-purchasing experience on the relationship between failure magnitude and order frequency is tested. In this phase of the test a product term (order frequency * e-purchasing experience) is added. The impact of the moderator is assessed based on the improvement in R2 in the second stage over the first stage. The same process is done to test the moderating effect of e-purchasing experience on order size after failure.

(18)

5. Results

In this section, the data analysis is described in detail and results are presented. The outcomes of the analysis provide arguments for approving or rejecting the hypotheses that were developed in chapter 3. The names of the online retailers, who provided the data, cannot be mentioned in the report. In the report we will refer to them as Company A & B. Company A is market leader in the online sales of ink cartridges, toners and paper. The company was launched in 2000 and is one of the fastest

growing and most successful merchants of the Netherlands with still a huge growth. Company B is a web-based department store with an extensive range of branded products in the areas of living, cooking, garden, design, gifts and wellness. The company consists of more than 100 online stores and brand shops.

5.1 Results Company A

This section presents the outcomes of the analysis for Company A. This company provided data with recent purchase orders. One month, in this case November 2011, was selected to analyze the failed and successful deliveries. The data is a random sample of 2500 orders from the month November 2011. This random sample contains both orders with fulfillment failures and successful deliveries. Henceforth, the group with an order fulfillment failure will be called treatment group, those who did not experience a failure will be called control group. Actual purchasing behavior of both groups will be analyzed six month prior to the failure/success and six month after failure/success.

The treatment group contains 206 orders and the control group contains2096 orders. The total of 206 orders was delivered after the promised delivery date, which means that 8.9% (approximately 9%) of the orders could not be fulfilled on the promised delivery day. On average an order was delivered 3.95 (4) days after the promised delivery day. The magnitude of failure ranges from 1 day up to 26 days late. Table 1 (treatment group) and table 2 (control group) will provide descriptive statistics for a general overview of the data.

Table 1

Descriptive statistics of the treatment group

Measurement May 2011 – October 2011 December 2011 – May 2012 Order frequency Order size Order frequency Order size

Mean 1,95 482,32 2,03 496,09

Standard dev. 7,46 2707,50 5,46 1728,90

Table 2

Descriptive statistics of the control group

It can be concluded from table 1 that order frequency is increasing after failure to 2,03 (period of December 2011 till May 2012). The order frequency increases in the period after failure compared to the period before failure.

Measurement May 2011 – October 2011 December 2011 -May 2012 Order frequency Order size Order frequency Order size

Mean 0,67 85,41 0,64 81,89

(19)

MSc Thesis J.H. Steen August 31st, 2012 Page 13 Order size is increasing from 482,32 to 496,09 after a delivery failure is experienced.These results are unexpected and not in line with the hypothesis development. Table 2 shows that the order frequency for the control group is decreasing, from 0,67 to 0,64. The order size is decreasing as well, from 85,41 to 81,89 after successful delivery. Again, these are unexpected results. To test if these differences are significant, a paired t-test is performed. The paired sample t-test is conducted to test if significant relations are found between the two time periods for both the order frequency and order size. The results of this test are presented in the first two columns of table 3.

Table 3

Statistic outcomes of the paired t-test

Variable Treatment group Control group

Order frequency M=-0,087, T=-0,359, P=0,720 M=0,02, T=1,110, P=0,267 Order size M=-13,77, T=-0,137, P=0,891 M=3,517, T=0,503, P=0,615 The two time periods, before and after failure, are compared to each other for order frequency and order size. Table 1 shows that order frequency after failure increased compared to the order frequency before failure. The results in table 3 show that this increase is not a significant difference, since the P-value is larger than 0.05. Order size shows an increase after failure but this was not a significant different from the first period as the P-value is 0,891. The control group shows a decrease for both order frequency and order size. Again, table 3 shows that these decreases are not significant because the P-value is not smaller than 0,05.

The above results show that when customers experience a failure their order frequency and order size increased compared to the period before failure. The control group, who got delivered

successfully, decreased their order frequency and order size. These results are unexpected and not in line with the reasoning of previous research. Company A sends a card with their apologies to the customers after a delivery failure. It can be assumed that this would increase the order frequency and size even after a failure. The effect of these apology cards can be tested by means of interviews. This research is limited to actual purchasing behavior data, and cannot test the effect of this variable on future purchasing behavior of customers.Table 4 continues with the unpaired t-test, to test for significant differences between the treatment and control group.

Table 4

Statistic outcomes of the unpaired t-test

Variable Treatment vs. control group Implications

Order frequency M= 0,620, T=0,590, P=0,555 No significant difference between the failure or success group in order frequency

Order size M=10,25, T=0,266, P=0,790 No significant difference between the failure or success group in order size

(20)

MSc Thesis J.H. Steen August 31st, 2012 Page 14 For the unpaired t-test no equality of variances is assumed for both variables. The P-value for order frequency is 0,555. This means that no significant difference is found between the two groups when comparing order frequency. In practice this would mean that delivery failures or successful deliveries do not impact the order frequency afterwards. Comparing this outcome to table 1 and 2, which show the descriptive statistics, it would be expected to find a significant difference between the two groups. Order size is not significant either with a P value of 0,790. In this case the expectations were even higher to find a significant difference. In table 1 and 2 the standard deviations are very high, indicating large outliers in the data. The data shows a few outliers which could potentially influence the results of the t-test. The same tests are therefore repeated but all customers who bought more than 10 times in the six months previous or after failure are excluded. After this the two groups are compared again to see the influence of the outliers. Table 5 and 6 provide descriptive statistics of the treatment group and control group to give a general overview of the data.

Table 5

Mean 0,80 139,49 0,94 196,04

Standard dev. 1,52 496,48 1,32 498,53

Table 6

Table 5 shows that order frequency increases from 0,80 before failure to 0,94 after failure and order size increases from 139,49 to 196,04. These findings are in line with the descriptive statistics as presented in table 1. This data set shows a decrease in the standard deviations. This should have an impact on the outcomes of the paired sample t-test. Table 6 shows that order frequency and order size are decreasing after a successful delivery, respectively from 0,63 to 0,61 and from 69,82 to 65,14. This is consistent with the statistics as presented in table 2. As said previously, these results are unexpected for the both groups. Table 7 presents the results of the paired t-test.

Table 7

Statistic outcomes of paired t-test

Order frequency M=-0,15, T=-1,65, P=0,101 M=0,025, T=1,180, P=0,238 Order size M=-56,55, T=-2,68, P=0,008 M=4,68, T=0,838, P=0,402 Similar to what has been done previously, order frequency before and after failure are compared to each other for the treatment and control group. Order frequency shows no significant difference, with a P-value of 0,101, for the treatment group.

Mean 0,63 69,82 0,61 65,14

(21)

MSc Thesis J.H. Steen August 31st, 2012 Page 15 This is in line with the descriptive statistics which show only a small increase in frequency. The order size is significantly different after experiencing a failure because the P-value is smaller than 0,05. The control group shows different results. Both, order frequency and order size are not significantly different when the two time periods are compared. This was expected from the descriptive statistics that show a small decrease in order frequency and order size. To test the differences between the treatment and control group an unpaired t-test is performed. The outcomes of this test will show if significant differences are found between the two groups. Table 8 presents the results of this test. Table 8

Statistic outcomes of the unpaired t-test

Variable Treatment vs. control group Implications

Order frequency M=0,123, T=1,626, P=0,104 No significant difference between the treatment and control group. Order size M=51,87, T=2,68, P=0,007 Significant difference between the

treatment and control group Table 8 shows that order frequency is not significantly different for the control group compared to the treatment group since the P-value is larger than 0,05. This result is in line with previous

descriptive statistics which indicate only a small difference between the two groups. The order size does show a significant difference between the two groups. This shows that the control group, customers who experienced a successful delivery, had significant smaller order sizes in the period of December 2011 and May 2012. As can be learned from previous done literature, it would be

expected that the control group would increase their order size instead of decreasing. Table 4 through 8 show that the outliers in the data do have an influence on the outcomes. When continuing the analysis the potential effect of these outliers on the data should be considered.

The second part of the study concentrates solely on the treatment group, since the control group did not experience a failure. To answer hypotheses three to six regression analysis is performed.

Regression analysis is a statistical tool for the investigation of relationships between variables. The statistical significance is assessed of the estimated relationship, that is, the degree of confidence that the true relationship is close to the estimated relationship. In this study, the effect of magnitude of failure on the order frequency and order size is tested. Where magnitude of failure is the

independent variable and the order frequency or order size is the dependent variable. First, the relationship between failure magnitude and order frequency after failure is tested. The outcome of the regression analysis is found in table 9.

Table 9

Resulting levels of significance from regression analysis (dependent variable: Order frequency after failure)

Constant 0,034

Magnitude 0,304

Experience 0,004

Order frequency before failure 0,000

R² 0,817

(22)

MSc Thesis J.H. Steen August 31st, 2012 Page 16 Table 9 shows that experience and order frequency have a significance level below 0,05. This means that experience and order frequency before failure have a significant relationship with order

frequency after failure. The sign is positive, indicating that e-purchasing experience and order frequency before failure have a positive effect on the order frequency after failure.No evidence is found for either a positive or negative relationship between failure magnitude and order frequency after failure. In this study e-purchasing experience is also a moderating variable on the relationship between failure magnitude and order frequency after failure. The impact of the moderator (e-purchase experience) is assessed based on the improvement in R² in the second stage regression over the first stage. If this change is statistically significant, then a significant moderated effect is predicted. Table 10 presents the results of this test.

Table 10

Resulting levels of significance from regression analysis(dependent variable: Order frequency after failure)

Stage 1 Stage 2

Constant 0,034 0,226

Magnitude 0,304 0,837

Experience 0,004 0,003

Order frequency before failure 0,000 0,000

Moderating variable (Experience x Magnitude) - 0,219 R² 0,817 0,818 R² adjusted 0,814 0,815 Δ R² - 0,001

Table 10 shows that the moderating variable does not present a significant difference, as the level of significance is above 0,05. The variable does not provide a moderating effect on the relationship between magnitude of failure and order frequency after failure. I.e. when a customer bought a product from company A in the past, this does not necessarily mean that their order frequency will remain similar after experience a fulfillment failure. The same process of regression can be repeated when testing hypothesis 6 which states that e-purchasing experience positively affects the

relationship between the magnitude of failure and order size. In the first stage order size after failure is regressed on the magnitude of failure and experience. In the second stage the moderating variable e-purchasing experience is added to the regression analysis. The change inR² shows if this variable has a moderating effect on the relationship between magnitude of failure and shopping behavior after failure.

Table 11

Resulting levels of significance from regression analysis (dependent variable: order size after failure)

Stage 1 Stage 2

Worth before failure 0,000 0,000

(23)

MSc Thesis J.H. Steen August 31st, 2012 Page 17 When focusing on stage 1 it can be concluded that the magnitude of failure has no significant

relationship with order size after failure. Magnitude of failure shows a significance level of 0,693 which is larger than 0,05. Furthermore, from the table it can be concluded that the e-purchasing experience does not have a moderating effect on the relationship between magnitude of failure and order frequency after failure. The change in R² is such a small number that is can be neglected.

5.2 Results Company B

This section presents the outcomes of the analysis on the data from Company B. Similar to the analysis of Company A, November 2011 is taken for analysis. Company B did not provide a sample but provided a complete group of failures and successes. In total, 1310 failures were reported and 5578 successful deliveries. As mentioned in section 5.1 the delivery failures will be called treatment group and the successful deliveries will be called control group. From both groups data is analyzed six month prior to failure or success and six month after. On average the orders were delivered 3,38 (3) days later than promised.The magnitude of failure ranges from 1 day up to 32 days. Table 12 and 13 present the descriptive statistics for the treatment and control group, to provide some general insights into the data.

Table 12

Mean 0,14 14,33 0,16 13,37

Standard dev. 0,582 75,06 0,507 55,93

Table 13

Order frequency increased for both the treatment group and control group. As can be seen in table 12 the order frequency increase from 0,14 to 0,16 for the treatment group. For the control group the increase is from 0,12 to 0,14. Even when a delivery failure was experienced, the same increase in order frequency is reported. For the variable order size, the treatment group decreased from 14,33 to 13,37. After the a successful delivery the order size increased from 10,44 to 11,45. These are results are in line with the expectation, except the order frequency of the treatment group. Order frequency was expected to decrease after an unsuccessful delivery. Table 14 presents the results of the paired t-test for the treatment and control group.

Mean 0,12 10,44 0,14 11,45

(24)

MSc Thesis J.H. Steen August 31st, 2012 Page 18 Table 14

Statistic outcomes of paired t-test

Order frequency M=-0,015, T=-0,0823, P=0,411 M=-0,023, T=-1,130, P=0,258 Order size M=0,95, T=0,417, P=0,677 M=-1,007, T=-0,991, P=0,322 The treatment group shows a significant but not a strong correlation between the periods before and after failure. The paired t-test shows that order frequency for the treatment group is not significant (P=0,411) with a 95% confidence interval. Order size is not showing a significant difference between the two time periods, because the P-value is 0,677. The control group shows different results for the paired t-test. Order frequency and order size both are not strong correlated and do not show a significant difference before and after success. The P-value is respectively, 0,258 and 0,322. The paired t-test, for both groups, is performed with a 95% confidence interval. To find the difference between the two groups an unpaired t-test is performed. The results of this test are summarized in table 15.

Table 15

Statistic outcomes of unpaired t-test

Variable Treatment vs. control group Implication

Order frequency M= 0,38, T= 0,871, P=0,384 No significant difference between the treatment and control group Order size 0,053, T=0,022, P=0,982 No significant difference between

the treatment and control group. The Levene’s test for equality for variances will show whether equality of variances can be assumed. Both groups show no significant result for the test, which means that equal variances are assumed. As can be seen in table 15, the treatment group and control group are not significantly different from each other in order frequency and order size. Both p-values are larger than 0,05, which means no significant relation. It can be concluded that when customers experienced a delivery failure in November 2011 their shopping behavior does not change compared to those customer that experienced successful deliveries. From the descriptive statistics that are presented in table 12 and 13 it would be expected that order size would be significantly different for both groups. Since, both group do not show very large differences in order size and the sample is large this could explain why the statistical test does not show a significant difference.

(25)

MSc Thesis J.H. Steen August 31st, 2012 Page 19 Table 16

Resulting levels of significance from regression analysis (dependent variable: Order frequency after failure)

Constant 0,000

Magnitude 0,350

Experience 0,103

Order frequency before failure 0,000

R² 0,105

R² adjusted 0,103

The results of table 16 show that there is approximately 10% (R² = 0,105) change that a relation exist between the variables mentioned in the table and order frequency after failure. This means that it cannot be predicted if order frequency after failure will decrease or remain similar. Order frequency before failure is the only variable that has a significant effect on this relationship. Because this is the only variable with a significance level below 0,05. Magnitude of failure has no impact on the order frequency after failure. This would indicate that small failures, orders delivered only a few days after the promised delivery date, have the same impact as a high failure magnitude. The diversity in products can explain the unexpected results found in this analysis. The e-tailer offers an extensive range of products. Because of the extensive range of products the purchasing patterns are difficult to detect. Furthermore, this e-tailer has to consider that seasonality of products influence the buying patterns. In stage two of the regression analysis the moderating effect of e-purchasing experience is measured. The change in R² will show what effect the moderating variable has on the relationship between failure magnitude and order frequency after failure.

Table 17

Resulting levels of significance of regression analysis (dependent variable: Order frequency after failure)

Stage 1 Stage 2

(26)

MSc Thesis J.H. Steen August 31st, 2012 Page 20 The same process is repeated to analyze the effect of magnitude of failure on the order size after failure. The same independent variables are used but in this specific regression the dependent variable is order size after failure. A summary of stage 1 and 2 of the regression analysis are presented in table 18.

Table 18

Resulting levels of significance from regression analysis (dependent variable: order size after failure)

Stage 1 Stage 2

Approximately 5% of the results in order size after failure can be explained by the independent variables order size after failure and experience. As said earlier the e-tailer offers a wide range of products this can be a possible explanation why no clear purchase pattern can be detected. In stage 1 of the regression it can be found that experience and order size before failure have a significant relationship with order frequency after failure. It is likely that when customers bought from the e-tailer before their order size increase, as the trust and loyalty towards the e-e-tailer increases as well. In stage two of the regression analysis is can be concluded that the R² did not increase. Moreover the moderating variable does not have a significant effect on the relationship between failure magnitude and order size after failure.

(27)

6. Discussion

In this section of the research the results as presented in chapter 5 are discussed and conclusions are made. Furthermore, in this chapter the hypotheses that were developed in chapter 3 are answered. The answers to the hypotheses can help to formulate a conclusion to the main objective of this study.

6.1 Discussion company A & B

From the above result section it can be conclude that both hypotheses 1 and 2 are rejected since no evidence is found that the two groups behave differently. These hypotheses predicted that the order frequency and order size would decrease when customers would receive their products after the promised delivery date. In previous research the outcomes were different from this study. Several researchers (I.e. Bitner et al., 1990, Boulding et al., 1993, Kelley et al., 1993, Smith et al., 1999, McCollough et al., 2000) show that when customers are dissatisfied with the service encounter, this leads to lower perceived service quality, which, in turn, lead to reduced purchase activity. Even though this research was performed in the service industry a similar outcome could be expected for the e-commerce industry. This research, as said before, shows different findings compared to

previous literature. The results of the current study show that even when a customer experienced an order fulfillment failure this does not necessarily mean that purchase activities will be decreased. Moreover, the results of Company A show an increase in order frequency and order size for the treatment group and a decrease for both variables for the control group. Company B reports an increase in order frequency for both the treatment and control group. Only the treatment group experienced a decrease in order size after failure, which is more in line with previous studies. Hypothesis 3 which states that online order fulfillment failure magnitude has a differentiated negative relationship with order frequency is rejected. From the regression analysis that was performed for both companies no significant relationship is found between the two variables. This result is very unusual and was not expected. From the study of Rao et al (2011, p.700) it can be found that, “as orders are delivered later and later, the customer reacts to this treatment, and reduces the frequency with which they purchase from the internet retailer in following periods.” For this sample in a different market segment but with even a higher magnitude of failure (28 days) no such

conclusions can be made. Again, as found in the study of Wang et al. (2011), when customers experience a high magnitude of service failure they are more likely to exhibit low customer loyalty. This study does not find such a relationship but results show that magnitude of failure does not have an impact on order frequency of customers. Hypothesis 4 which states that online order fulfillment failure magnitude has a differentiated negative relationship with order size is rejected. Again these results differ from the findings found previous in literature. Overall, it can be conclude about hypothesis 3 and 4 that the magnitude of failure does not impact the order frequency nor order size of customers after failure. I.e. a customer can get a very late delivery of about 10 days, the difference in order frequency and order size will not be different from a late delivery of only 1 day.

For hypothesis 5 e-purchasing experience is used as a moderating variable between the magnitude of failure and order frequency. In stage one of the process e-purchase experience was used as

(28)

MSc Thesis J.H. Steen August 31st, 2012 Page 22 Here no significant relationship is found for either Company A or B. This means that e-purchasing experience has no moderating effect on the relationship between magnitude of failure and order frequency after failure. Hypothesis 5 is rejected. Similar results are found when concerning order size as dependent variable, instead of order frequency. In the first stage a significant relationship is found between e-purchasing experience and order size after failure for Company A and B. In the second stage, where e-purchasing experience is used as moderated, no significant relationship is found between failure magnitude and order size after failure. The study of Boulding (1993) states that failure occurring early in the customer’s relationship with a supplier will be perceived more adversely than one which occurs later in a relationship because the customer has less experience of successful service experiences to counterbalance the failure.

6.2 Comparison Company A & B

(29)

7. Conclusion

The principal objective of this research is to examine the relationship between online order fulfillment failures and subsequent shopping behavior in an online retailing environment. Previous research has shown that the internet as a channel for distribution of goods is growing significantly. This means that more often customers order their products online. Since this channel is growing, customers can easily compare offerings of internet retailers in order to get the best delivery options. This competition puts pressure on the internet retailers to improve their logistical performances. Therefore it is important to research the impact of delivery failures on future purchasing behavior of customers. This research is done by comparing two groups of customers, either experiencing a delivery failure or experiencing a successful delivery. By comparing their actual shopping behavior after the delivery failure or successful delivery, the impact of failures can be measured. Secondly, the regression analysis is used to test the impact of failure magnitude on the shopping behavior after failure. The second stage of this regression analysis determines the moderating effect of e-purchasing experience. The research presents three main outcomes that will be discussed below. This study focuses upon the relationship between online order fulfillment failures and subsequent shopping behavior of customers. From the CSA analysis we did not find significant differences between the treatment and control group. This means that the two groups behave similar to each other. From this research it can be concluded that experiencing a failure in order fulfillment does not result in a decrease in purchase activity. This research extends the examination beyond percentage of late order or the magnitude, and presents results on the e-purchase experience of customers. The results do not suggest there is a significant relationship between magnitude of failure and

subsequent shopping behavior of customers. Spending extra effort to minimize the lateness of orders, given the observed reactions of customers, would not be considered as useful. Furthermore, the experience of customers is significant related to the order frequency and order size after failure for Company A. For Company B the experience of customers is only significant related to order frequency. Whether or not this result is positive or negative is not found from the results. But when a customer has e-purchase experience it impacts the order frequency and size. The analysis from both companies shows that this variable has no moderating effect on the relationship between magnitude of failure and subsequent shopping behavior.

(30)

MSc Thesis J.H. Steen August 31st, 2012 Page 24 Lastly, this study focuses on the actual shopping data from customers. This illustrates the effect of operations management on marketing in online retailing.

7.2 Limitations and future research

(31)

References

Achen, C.H., 1986. The Statistical Analysis of Quasi-experiments. University of California Press.

Alba, J., Lynch, J.J., Weitz, B., Janiszewski, C., Lutz, R., Wood, S., 1997. Interactive home shopping: Consumer, retailer, and manufacturer incentives to participate in electronic marketplaces. Journal of Marketing 61 (3), 38– 53

Anderson, E., Weitz, B., 1989. Determinants of continuity in conventional industrial channel dyads. Marketing Science 8 (4), 310–323.

Bejou, D., & Palmer, A. (1998). Service failure and loyalty: An exploratory empirical study of airline customers.

Journal of Services Marketing, 12(1), 7–22

Bitner, M. J., Booms, B. H.,&Tetreault, M. S. (1990). The service encounter: Diagnosing favorable and unfavorable incidents. Journal of Marketing, 54(1), 71–84.

Bitner, M.J., Hubbert, A.R., 1994. Encounter satisfaction versus overall satisfaction versus quality: the

customer’s voice. In: Rust, R.T., Oliver, R.L. (Eds.), Service Quality: New Directions in Theory and Practice. Sage, Thousand Oaks, CA, pp. 72–94.

Bolton, R. N. (1998). A dynamic model of the duration of the customer’s relationship with a continuous service provider: The role of satisfaction. Marketing Science, 17(1), 45–65.

Boulding W., Kalra A., Staelin R., Zeithaml V.A. 1993. A dynamic process model of service quality: from expectations to behavioral intentions. Journal of Marketing Research. 30(1):7-27.

Buttle, F.,&Burton, J. (2002). Does service failure influence customer loyalty? Journal of Consumer Behavior, 1(3), 217–227.

Cho, J.J.-K., Ozment, J., Sink, H., 2008. Logistics capability, logistics outsourcing and firm performance in an e-commerce market. International Journal of Physical Distribution and Logistics Management 38 (5), 336–359. Davis FD. Perceived usefulness, perceived ease of use and user acceptance of information technology. MIS Quarter 1989;13(3):319–39.

De Koster, R., 2003. Distribution strategies for online retailers. IEEE Transactions on Engineering Management 50 (4), 448–457.

Esper, T.L., Jensen, T.D., Turnipseed, F.L., Burton, S. (2003). The last mile: an examination of effects of online retail delivery strategies on consumers. Journal of business logistics, 24 (2)

Folkes, V.S., 1984. Consumer reactions to product failure: an attributional approach. Journal of Consumer Research 10 (4), 398–409.

Goodwin, C., Ross, I., 1992. Consumer responses to service failures: influence of procedural and interactional fairness perceptions. Journal of Business Research 25 (2), 149–163.

Grewal, D., Iyer, G.R., Levy, M., 2004. Internet retailing: enablers, limiters, and market consequences. Journal of

Business Research 57 (7), 703–713.

Hair Jr., J.F., Black, W.C., Babin, B.J., Anderson, R.E., Tatham, R.L., 2006. Multivariate Data Analysis. Pearson-Prentice Hall, New Jersey.

(32)

Hess, R., Ganesan, S., Klein, M., 2003. Service failure and recovery – the impact of relationship factors on service satisfaction. Journal of the Academy of Marketing Science 31 (2), 127–145.

Kelley, S.W., Davis, M.A., 1994. Antecedents to customer expectations for service recovery. Journal of the Academy of Marketing Science 22 (1), 52–61.

Kelley, S. W., Hoffman, K. D., & Davis, M. A. (1993). A typology of retail failures and recoveries. Journal of Retailing, 69(4), 429–452.

Kenny, D.A., 1975. A quasi-experimental approach to assessing treatment effects in the nonequivalent control group design. Psychological Bulletin 82 (3), 345–362.

Kuan HH, Bovk G-W, Vathanophas V. Comparing the effects of website quality on customer initial purchase and continued purchase at e-commerce websites. Behav Inf Technol 2008;27(1):3-16.

Lewis, M., Singh, V., Scott, F., 2006. An empirical study of the impact of nonlinear shipping and handling fees on purchase incidence and expenditure decisions. Marketing Science 25 (1), 51–64.

McCollough, M. A., Berry, L. L., & Yadav, M. S. (2000). An empirical investigation of customer satisfaction after service failure and recovery. Journal of Service Research, 3(2), 121–137.

Oliver, R.L., Swan, J.E., 1989. Consumer perceptions of interpersonal equity and satisfaction in transactions: a field survey approach. Journal of Marketing 53 (2), 21–35.

Ramanathan, R. (2010). The moderating roles of risk and efficiency on the relationship between logistics performance and customer loyalty in e-commerce. Transportation Research Part E 46 950–962

Rao, S., Griffis, S.E., Glodsby, T.J. (2011). Failure to deliver? Linking online order fulfillment glitches with future purchase behavior. Journal of operations Management. 29, p. 692-703.

Roos, I. (1999). Switching process in customer relationships. Journal of Service Research, 2(1), 68–85. Saunders. D. R. (1956), Moderator variables in prediction. Education and Psychological measurement. 16, 209 222.

Smith, A. K., Bolton, R. N., &Wagner, J. (1999). Amodel of customer satisfaction with service encounters involving failure and recovery. Journal of Marketing Research, 36(3), 356–372.

Solomon, M.R., Suprenant, C.F., Czepiel, J.A., Gutman, E.G., 1985. A role theory perspective on dyadic interactions: the service encounter. Journal of marketing 49, 99-111.

Waller, M., Johnson, E., Davis, T., 1999. Vendor managed inventory in the retail supply chain. Journal of Business Logistics 20 (1), 183–203.

Wang, Y.S., Wu, S.C., Lin, H.H., Wang, Y.Y. (2011). The relationship of service failure severity, service recovery justice and perceived switching costs with customer loyalty in the context of e-tailing. International Journal of

Information Management 31 350–359

Weun, S., Beatty, S. E., & Jones, M. A. (2004). The impact of service failure severity on service recovery evaluations and post-recovery relationships. Journal of Services Marketing, 18(2), 133–146.