• No results found

The effect of the customer experience on financial performance: a cross-industry study

N/A
N/A
Protected

Academic year: 2021

Share "The effect of the customer experience on financial performance: a cross-industry study"

Copied!
46
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)
(2)

2

The effect of the customer experience

on financial performance:

a cross-industry study

By Lisanne Venema S2531542 University of Groningen Faculty of Economics and Business

MSc Marketing Intelligence & Marketing Management

Master Thesis January 2019 Antillenstraat 1-80 9714JT Groningen +31 (0) 6 23 88 01 94 l.venema.3@student.rug.nl Student number: S2531542

First Supervisor: prof. dr. T. H. A. Bijmolt Second Supervisor: dr. H. Risselada

(3)

3

Management summary

Customers are key to companies since they determine how successful a company will be. In order to track the behaviour of customers, customer feedback metrics (CFMs) are used. Some examples of CFMs are customer satisfaction, Net Promotor Score and repurchase likelihood. Due to the growing amount of CFMs, companies are not sure which CFM to use and how to interpret them. Previous research highlighted the importance of for example customer satisfaction since this metric is an indicator of financial performance. Therefore, it is important to investigate the influence of different CFMs on financial performance. Eventually, this will create more clarity and appropriate use of CFMs. In this paper, another CFM was investigated thoroughly. This CFM is the customer experience, developed by the Dutch marketing research company SAMR. SAMR uses their CFM to track 218 companies in the Netherlands. Every year, questionnaires are distributed to Dutch respondents in an online panel. Eventually, respondents are asked to rate companies they visited on different aspects of the customer experience. Since the effect of this CFM on financial performance was not investigated, this became the main objective of this study.

Financial performance is a rather broad definition, and therefore this study focused specifically on the effect of the customer experience on sales growth in percentages. Besides the more general question whether the customer experience can be used in order to be an indicator of sales growth rate, it was interesting to obtain more detailed information about this CFM. Therefore, it was also examined which aspect of the customer experience has the largest effect on sales growth. Also, this CFM uses a multi-item measurement scale as opposed to most CFMs. Therefore, it was tested whether the use of this multi-item scale is preferred over the use of a single-item scale in testing the effect of the customer experience on sales growth. Literature also suggests that at a certain point customers can become saturated, where after an increase in the customer experience score will not lead to an increase in sales growth anymore. In the end, this can even cause a decrease in the sales growth. In other words, an inverted u-shape relationship is suggested. So, investigating that point was also an objective of this study. Furthermore, the dataset of SAMR includes industry indicators. Therefore, industry effects were also taken into account.

(4)

4 First of all, this research found no significant linear relationship between the customer experience and sales growth. Nevertheless, a significant inverted u-shape relationship between the customer experience and sales growth was found, but this only holds for specific values of the customer experience score. More specifically, companies obtaining a customer experience score between 6 and 6.6 will obtain an increase in sales growth rate. Moreover, companies with a score between 6.61 and 7.8 will suffer from a decrease in sales growth rate. However, a customer experience score above 7.8 will lead to an increase in sales growth rate. So, the supported inverted u-shape relationship only holds for the customer experience scores between 6 and 7.8. This nonlinear relationship was also supported by the significant finding that companies moving from a customer experience score of above a 7 to a score below 7 will obtain an increase in their sales growth rate. Even though theory suggests the use of a multi-item scale, this study found no significant difference in the use of a single-item scale opposed to a multi-item scale. At last, an industry effect is present. Namely, the customer experience has 154% more impact on sales growth in the leisure industry compared to the retail industry.

In addition, this study has implications for both academics and practitioners. Academics can use this research as an extension in the field of the effect of CFMs on financial performance. Also, it adds insights regarding industry effects. Practitioners can use this research in order to tailor their marketing strategy and eventually, obtaining an increase in sales growth rate.

(5)

5

Preface

This is my master’s thesis: The effect of the customer experience on financial performance: a cross-industry study. This thesis is the final work of my master’s degree in Marketing Intelligence and Marketing Management at the University of Groningen. It presents the results of the research I conducted from September 2018 until January 2019.

The research was based on a request by SAMR, a Dutch market research company. Together with Gerrit Piksen from SAMR and my supervisor prof. dr. T. Bijmolt a brainstorm session was conducted in order to find the right direction regarding this research. Therefore, I would like to thank Gerrit Piksen with providing my the right data for this research. I enjoyed using the data and analysing it. Also, I would like to thank prof. dr. T. Bijmolt for his feedback and useful discussions during my master’s thesis, it helped me a lot. During my thesis, I realised that I learned a lot, both during the courses of the master and while writing my thesis. This confirmed that I had made the right choice regarding the master Marketing Intelligence and Marketing Management. At last, I would like to thank my family and friends for their support during my master.

I hope you enjoy reading my master’s thesis. Lisanne Venema,

(6)

6

Contents

1. Introduction ... 8

1.1 Problem statement and research questions ... 8

1.2 Academic relevance... 10

1.3 Managerial relevance ... 10

1.4 Structure of the study ... 11

2. Theoretical framework ... 12

2.1 Customer Feedback Metrics ... 12

2.2 The customer experience ... 13

2.3 Financial performance and its relationship with the customer experience ... 13

2.4 Multi-item scale ... 15

2.5 Non-linear relationship between customer service and financial performance ... 15

2.6 Industry ... 16

3. Methodology ... 17

3.1 Data ... 17

3.1.1 The dependent variable ... 18

3.1.2 The independent variables ... 19

3.1.3 Data cleaning ... 19

3.1.4 Imputation of missing values ... 20

3.2 Descriptive analysis ... 21

3.3 Multiple regression... 22

3.4 Multicollinearity ... 22

3.4.1 Correlation matrix ... 23

3.4.2. VIF ... 23

3.4.3 Principal Component Analysis ... 24

3.4.4 Factor Analysis ... 24

3.5 Normality and heteroscedasticity ... 25

3.5.1 Heteroscedasticity, non-normality ... 25

3.6 The final models ... 26

4. Results ... 28

4.1 The effect of the customer experience on sales growth ... 28

4.2 The effect of the Golden Rules on sales growth rate ... 28

4.3 Single-item scales versus multi-item scales ... 28

(7)

7

4.5 The effect of the customer experience on sales growth rate across industries ... 30

4.6 Additional analysis ... 33

4.7 Validation ... 33

4.7.1 R-squared ... 33

4.7.2 Overall model fit ... 34

4.7.3 Information criteria ... 34 5. Conclusion ... 35 5.1 Discussion ... 35 5.2 Implications ... 37 5.2.1 Academic implications ... 37 5.2.2 Managerial implications ... 37 5.3 Recommendations... 38

5.4 Limitations and future research ... 38

References ... 40

Appendices ... 43

Appendix 1: Cook’s distance plot ... 43

Appendix 2: Histogram and boxplots of sales after data cleaning ... 43

Appendix 3: Importance of Principal Components ... 43

Appendix 4: Scree plot of Principal Components Analysis ... 44

(8)

8

1. Introduction

In 1954, more than 60 years ago, Peter F. Drucker already demonstrated that without customers companies are not able to exist. Since customers are that important to companies, they are at the cornerstones of every company. Accordingly, customers are able to determine the success rate of a company. In order to keep track of customers’ behaviour, companies make use of Customer Feedback Metrics (CFMs). Some examples of those CFMs are average satisfaction score, top 2 box customer satisfaction score, Net Promotor Score and repurchase likelihood (Morgan & Rego, 2006). CFMs are not only used to keep track of customer behaviour, but they are also indicators of financial performance. More specifically, it is widely known that companies who focus on increasing their customer satisfaction rate are rewarded with an increase in financial performance (Anderson, Fornell & Lehmann, 1994).

Even though literature shows the importance of CFM usage, companies lack the knowledge on which CFMs to use and how to interpret changes in those CFMs. This can lead to frustration and irritation, or even the abandonment of the utilized CFMs. Eventually, the accountability of the marketing department can be affected and can hinder companies from becoming customer centric (De Haan, Verhoef & Wiesel, 2015). Moreover, misuse of CFMs can even hinder companies from increasing their financial performance (Anderson et al., 1994). Therefore, companies should refocus again on their CFM use.

In 2007, SAMR introduced the so-called “Klantvriendelijkste Bedrijf van Nederland” award. This award put companies in the spotlight that are able to make appropriate use of CFMs and therefore are able to focus on their customers. The winner of the award is based upon the scores on a CFM developed by SAMR themselves. This CFM is one of a comprehensive kind since it contains different aspects. These aspects are overall satisfaction, convenience, honesty, obtrusiveness, expectation management and responsibilities towards people and society. More specifically, all these different aspects together capture the customer experience. The customer experience is an internal and subjective response of customers to factors directly or indirectly related to a company (Verhoef et al., 2009). Therefore, the CFM of interest in this study is the customer experience.

1.1 Problem statement and research questions

(9)

9 formulated:

To which extent is it possible to use the customer experience as an indicator of financial performance?

Financial performance is a rather broad definition, containing different accounting measures. Some examples of financial performance are sales growth, gross margin and cash flow (Van Doorn, Leeflang & Tijs, 2013). The relationship between CFMs and the aforementioned financial performance measures was examined before but achieved different conclusions. However, the CFM of SAMR was not examined before, but does contain measures of customer satisfaction. Since customer satisfaction measures show correlations with sales growth, this financial performance measure will be used in this paper (Van Doorn, 2013). Based on the proved correlation between customer satisfaction and financial performance, the first research question is as follows:

1. Is it possible to use the customer experience as an indicator of sales growth?

Most companies focus on one specific part of the customer experience, namely customer satisfaction. A question most likely related to: “Considering all your experience of company X, how satisfied are you,

in general?” is used to measure customer satisfaction (Grønholdt, Martensen & Kristensen, 2000).

Therefore, a single-item scale measures customer satisfaction. However, the customer experience is more comprehensive since it takes multiple aspects into account, such as honesty, obtrusiveness and customer satisfaction. Therefore, a multi-item scale is used to measure the customer experience. Literature shows that the predictive validity of the scale used depends upon the conditions under which the scale is used (Diamantopoulos & Sarstedt, 2012). This raises the second research question:

2. Is a multi-item scale better in measuring the effect of the customer experience on sales growth than a single-item scale?

Since the customer experience metric is a multidimensional construct, it is interesting from a managerial perspective to obtain more insights regarding the different aspects it covers. More specifically, it is interesting to know which one of those different aspects of the customer experience has the largest effect on sales growth. This raises the third research question:

3. Which CFM aspect has the most explaining power in its effect on sales growth?

(10)

10 performance will no longer increase due to customer satisfaction (Ittner & Larcker, 1998). Since the customer experience covers customer satisfaction as well, the fourth research question is as follows:

4. Does a nonlinear relationship between the customer experience and sales growth exist?

The award offered by SAMR is industry-wide, but they offer an award for every separate industry also. It can be assumed that the level of services offered to the customer is distinct per industry. In addition, the contact time between a company and its customers as well as the importance of the purchase differs between industries (Ou, Verhoef & Wiesel, 2015). Therefore, it can be suggested that the related industry of a company has an influence on the effect of the customer experience on financial performance. This raises the research question:

5. In which industry does the customer experience score have the largest effect on sales growth?

1.2 Academic relevance

This paper contributes to literature in several ways. First of all, the effect of several CFMs on different financial performance measures has been studied before. However, this research uses a CFM that has not been considered before, since it is developed by a Dutch company. Also, initially, where earlier studies used a CFM with a single-item measurement scale, whereas now, the CFM used in this study uses a multi-item scale (Morgan & Rego, 2006). Another literature gap exists regarding the importance of the customer experience in different industries. For example, it is suggested that focusing on the customer experience can only lead to a limited amount of growth since it is possible that only a specific market segment is interested in an enhanced customer experience (Verhoef et al., 2009). Therefore, this paper investigates the effect of the customer experience on sales growth in different industries. At last, a research gap exists regarding the development of the customer experience over time. Moreover, it is still unknown whether the effect of the customer experience changes over time due to for example changing expectations of customers. Consequently, this research contributes to this literature gap by taking different time periods into consideration.

1.3 Managerial relevance

(11)

11

1.4 Structure of the study

(12)

12

2.

Theoretical framework

In this section, the theoretical framework will be explained based upon the conceptual model, as visualised in figure 1. This section begins with a general explanation about CFMs and the customer experience. Thereafter, the relationship between the customer experience and financial performance will be explained accompanied by the first and second hypotheses. This will be accompanied by an elaboration upon the multi-item scale and the nonlinear relationship between the customer experience and financial performance complemented by related hypotheses. The last section of the theoretical framework is about the effect of the customer experience on financial performance in different industries.

Figure 1: Conceptual model

2.1 Customer Feedback Metrics

(13)

13

Study CFM Time focus Question Measurement scale

De Haan et al., 2015

Customer satisfaction

Present How satisfied or unsatisfied are you with [company x ]?

1 = very unsatisfied, 7 = very satisfied

De Haan et al., 2015

NPS Future How likely is it that you would recommend [company x] to a colleague or friend?

0= very unlikely, 10 = very likely, 9s and 10s are promotors

De Haan et al., 2015

Top-2-box customer satisfaction

Present How satisfied or unsatisfied are you with [company x ]?

1 = very unsatisfied, 7 = very satisfied, with 6s and 7s belonging to the Top-2-box

De Haan et al., 2015

Customer Effort Score

Past 1. Did you try to contact [company x] with any kind of request? 2. How much effort did you personally have to put in forth to handle your request?

1. Yes or no

2. 1 = very low effort, 5 = very high effort

Van Doorn et al., 2013

Multi-item satisfaction

Present Can you indicate how satisfied you are with the performance of [company x] regarding each of the following aspects:

- Portfolio of available products and services - Price/quality ratio of the products and services - Customized advice

- Clarity concerning relevant products and services

- General settlement of questions and complaints

1 = very dissatisfied, 10 = very satisfied

Table 1: CFM overview

2.2 The customer experience

Over the last years, the customer experience became more and more important to companies, and even to such an extent that it is incorporated in companies’ mission statements (Verhoef et al., 2009). Moreover, the customer experience is not only related to the purchase itself but considers the complete customer lifecycle. Therefore, it is about the interactions with a company before, during and after the purchase. Examples of the customer experience are the purchase itself, but also a company’s advertisements and news reports of a company (Meyer & Schwager, 2007). Therefore, the customer experience is a dynamic CFM taking into account different company-specific factors (Verhoef et al., 2009). Compared to other CFMs, the customer experience has a dynamic time frame. Where most CFMs are focused on one time frame (see table 1), the prior customer experience can influence future the customer experience. Therefore, the customer experience differs from the CFMs mentioned in table 1.

2.3 Financial performance and its relationship with the customer experience

(14)

14 satisfaction score does (Morgan & Rego, 2006). An overview of the CFMs being an indicator of several financial business performance measures can be found in table 2.

Study CFM Financial performance measure

Morgan & Rego, 2006 Average satisfaction score Net operating cash flows

Morgan & Rego, 2006 Average satisfaction score Annual sales growth

Morgan & Rego, 2006 Average satisfaction score Total shareholder returns

Morgan & Rego, 2006 Average satisfaction score Tobin’s Q

Morgan & Rego, 2006 Average satisfaction score Market share

Morgan & Rego, 2006 Average satisfaction score Gross margin

Morgan & Rego, 2006 Top-2-box satisfaction score Net operating cash flows

Morgan & Rego (2006), Williams & Naumann (2011) Top-2-box satisfaction score Annual sales growth

Morgan & Rego, 2006 Top-2-box satisfaction score Total shareholder returns

Morgan & Rego (2006), Williams & Naumann (2011) Top-2-box satisfaction score Tobin’s Q

Morgan & Rego, 2006 Top-2-box satisfaction score Market share

Van Doorn et al., 2013 Average satisfaction score Current sales growth

Van Doorn et al., 2013 Top-2-box satisfaction score Current sales growth

Van Doorn et al., 2013 Multi-item satisfaction score Current sales growth

Van Doorn et al., 2013 Net promoter score Current sales growth

Van Doorn et al., 2013 Alternative NPS Current sales growth

Table 2: Overview of different CFMs used as an indicator of financial business performance

As becomes clear from table 2, financial business performance can be measured in several ways. Nevertheless, one of the most reliable measures of financial performance is net operating cash flows since this measure is less sensitive to the accounting practices of a company than for example profits do. Moreover, customer loyalty metrics can be used as indicators of net operating cash flows. Since the CFM used in this paper is more related to customer satisfaction than customer loyalty, sales growth will be used as a measure of financial performance. The main reason for this is that customer satisfaction metrics perform better in predicting sales growth compared to its performance in predicting net operating cash flow (Van Doorn et al., 2013). Therefore, hypothesis 1 is created:

H1: The customer experience has a positive effect on sales growth

(15)

15 sustainability level of a company is flattening out over the last years. More specifically, around 79% of the Dutch customers state that sustainability affects their purchase decision in 2018, opposed to 81% in 2017 (Sustainable Brand Index, 2018). This can be explained by for example the presence of customers who do not care about the environment since this group represents 18% of the Dutch customers (Sustainable Brand Index, 2018). Therefore, hypothesis 2 is as follows:

H2: Compared to other the customer experience factors, the people and society factor has the smallest positive effect on sales growth

2.4 Multi-item scale

CFMs can be measured in multiple ways. For example, the NPS and Top-2-box satisfaction score are measured by one question but the multi-item satisfaction score is measured by multiple questions (for an overview, see table 1). However, the predictive validity of single- vs. multi-item scale is a point of discussion, since both contain advantages and disadvantages. The use of a single-item scale will reduce “oversurveying”, non-response rates and high costs of surveying additional items (Rogelberg & Stanton, 2007). However, a multi-item scale should not be disregarded easily. For example, the predictive ability of both scales depends upon the construct used, since constructs with an abstract nature should use multi-item scales opposed to single-item scales (Rossiter, 2002). Not only the nature of the construct is important in the choice between a single- or multi-item scale, but the predictive validity of both also differs between product categories and stimuli such as brands. Therefore, single-item scales should be used with caution and only in special circumstances (Diamantopoulos & Sarstedt, 2012). Another advantage of using a multi-item scale opposed to a single-item scale is related to the reliability and precision of the outcomes. Single-item scales contain a considerable amount of measurement error, leading to unreliable outcomes. The use of multi-item scales reduces this measurement error by summing the total scores of the single-items (Gliem & Gliem, 2003). This leads to more reliable and precise outcomes. The customer experience CFM used in this study has an abstract nature. It measures the customer experience, but it is difficult to grasp this complete construct in one question. Therefore, the following hypothesis is constructed:

H3: The multi-item scale CFM outperforms a single-item scale CFM as an indicator of sales growth

2.5 Non-linear relationship between customer service and financial performance

(16)

16 (Ittner & Larcker, 1998). Eventually, this will result in diminishing financial returns. At the saturation point, customers will not purchase more from a company. Even if the customer experience metric increases, the magnitude of sales growth will be less strong than the increase in the CFM (Streukens & De Ruyter, 2004). Moreover, this non-linear relationship can have an inverted u-shape. The saturation point can be seen as a super-saturation point in such a way that sales will decline after a certain point. This can be due to too much pressure put on creating the best the customer experience resulting in fewer sales (Manchanda & Chintagunta, 2004). Therefore, hypothesis 4 is as follows:

H4: The relationship between the customer experience and financial performance is non-linear with an inverted u-shape and therefore outperforms the linear model

2.6 Industry

The experiences customers encounter differs between industries, as it depends for instance on the complexity of the purchase (Ou et al., 2015). SAMR determines the customer experience score for companies from a whole range of industries. Even though the award is not industry-specific, winners within specified industries exist as well. As found by De Haan et al. (2015), the performance of CFMs as indicators of financial performance differs per industry. As, for example, in the online industry, customer satisfaction is of high importance. This is mainly due to the ease for customers to compare different companies and to switch to another when they are unsatisfied with their current choice. This will have a major influence on for example sales growth. Literature has found that several industry drivers have an effect on how loyal customers are towards companies, namely competition intensity, the innovativeness of the market, contractual settings, the visibility of product usage, the difficulty of evaluating product quality prior to consumption and the complexity of purchase decisions (Ou et al., 2015). Therefore, in this particular case, it can be assumed that the effect of the customer experience on sales growth will be larger for companies operating in an industry where purchases are of high importance to customers. These industry differences result in the following hypothesis:

(17)

17

3. Methodology

3.1 Data

In this paper, the dependent- and independent variables are based on different data sources. The main data source is the database of SAMR and is the provider of the independent variables. In 1985, the research agency Market Response was established due to the increasing demand of specific customers. Due to the increasing demand, the company Smart Agency was founded in 2000. In 2015, it was agreed to merge both companies, resulting in the company name SAMR (SAMR, 2018). SAMR operates in the Netherlands and collects the customer experience data in 12 different industries, such as banking, supermarkets and leisure.

The customer experience data are collected by distributing surveys in an online panel. Respondents are randomly assigned to three different industries. In the survey, respondents are asked which specific companies they have visited or whether they are a customer. If they did, respondents rate those companies based on the different aspects of the customer experience. One quarter later, respondents receive the survey again but choose four different industries. This cycle will be continued in the entire year. Those customer experience aspects are all related to a so-called by SAMR developed “Golden Rule”. According to SAMR, those Golden Rules can be seen as tools to deliver on those aspects customers value in a company (SAMR, 2018). Altogether, the different questions are the basis of the customer experience construct. Table 3 shows how the CFM of SAMR is exactly measured, and to which Golden Rule every question is related.

Question Golden Rule

1. To what extent are you satisfied with company X Overall satisfaction (no Golden Rule)

2. Company X does not bother me with irrelevant occasions Obtrusiveness

3. Company X’s employees are available when you need them Availability

4. Company X admits their mistakes and solves them Honesty

5. Company X delivers what it promises in advertisements Expectation management

6. Company X handles formalities such as product returns in a convenient way Convenience

7. Company X is involved with you as a customer Involvement

8. Company X cares about people and society People & society

Table 3: Survey questions and relationship to Golden Rules by SAMR

(18)

18 as possible, only companies with less than 20 respondents were deleted from the dataset. This resulted in a total of 216 companies.

3.1.1 The dependent variable

In this paper, the dependent variable is sales growth. Since the dataset of SAMR only includes data related to the customer experience score, other data sources are used to obtain companies’ sales numbers. In order to obtain those numbers, the main resource used is Orbis. Orbis is a database providing information on more than 300 million companies worldwide (Bureau van Dijk, 2018). The companies’ sales are stated in US dollars. Therefore, the sales amount is converted to euros. This is viable since the EUR/USD rate used is mentioned as well. However, not all the necessary data could be collected from this database. Therefore, online published financial reports of the companies themselves are used. Only in the case of no available company information in the aforementioned resources, online published news articles are used. In order to create the dependent variable, the absolute sales amount is used from 2013 onwards until 2017. After that, the change in sales is calculated and transformed into a percentage. By doing so, the effect of company size is excluded. An overview of the distribution of change in sales percentages can be found in figures 2, 3, and 4. The boxplots are split based on whether the change in sales concerns a decrease or increase.

Figure 2: Histogram of sales Figure 3: Boxplot of sales growth Figure 4: Boxplot of sales decline

(19)

19

Figure 5: Histogram of sales growth

What becomes clear from figure 5, is that the sales growth in percentages is far from normally distributed. Therefore, the dependent variable should be normalized by using a log transformation (Morgan & Rego, 2006).

3.1.2 The independent variables

Despite the addition of the seventh Golden Rule, the CFM of SAMR can be broken down into two sections: the first 6 Golden Rules, all related to the experience based on direct- or indirect contact with the firm. Therefore, they can be considered as customer service factors when they are combined into one concept. The other section contains the seventh Golden Rule. This Golden Rule is based on how customers perceive to what extent a company takes care about people and society in general. All Golden Rules are distinct independent variables. Besides those independent variables, the average satisfaction score, whether or not the specific company is part of a holding, the year the scores are obtained in, and in which industry the company operates are used as independent variables. All Golden Rules and the average satisfaction score are used as absolute numbers instead of a change in percentages since these kinds of CFM scores were used in literature before (Morgan & Rego, 2006, Van Doorn, 2013).

3.1.3 Data cleaning

(20)

20 to those abnormalities are deleted from the dataset. However, the boxplots in figures 3 and 4 do show more outliers than mentioned before, since growth percentages of more than 100% are present. So, the Cook’s distance measurement tool is also used. This tool highlights exact which data entry is an outlier. The corresponding plot can be found in Appendix 1. Using the outlierTest once again, 5 abnormal values are found. Four out of five values are way above the threshold line found in Appendix 1, and so, those entries will be deleted based upon the same reasoning as above. Other outliers are winsorized. By the use of winsorization the data remains in the dataset, but its value is rescaled to the nearest value within the valid distribution (Price et al., 2015). The values are rescaled by replacing their values with the 25th and 75th percentile. This has two advantages: 1) the dataset will not decrease

anymore and 2) the very small and very large values are still rather small/large and therefore retain their explaining power. In Appendix 2 the new histogram and boxplots can be found, which are more close to a normal distribution than the distribution showed figures 2, 3 and 4.

Regarding the dataset, there are also some problems with missing values. This is explained by the fact that not all sales numbers were available for every company. This led to missing values. First of all, some companies did not encompass any sales numbers. Those companies were deleted from the dataset since it was not possible to calculate the sales growth rate. A total of 125 companies remained. In addition, it turns out that the people and society variable is only available for 2017. Therefore, this variable cannot be used. This also holds for question 7, measuring whether the company is involved with the customer since this question is added since 2014 onwards. So, information on this variable is lacking for 2013. Moreover, the question about the availability of employees is lacking data as well. However, this is not the case for an entire year. In total, there is a grand total of 30 missing values for this question. Therefore, this question will not be deleted. How this missing data problem is treated will be explained next.

3.1.4 Imputation of missing values

(21)

21 matching (pmm) in case of numeric data, as is the case here. Pmm regresses the selected variables on the missing values in order to predict the value that should be imputed. This is done 5 times, where after 1 of those 5 values is chosen to be the imputed value (Azur et al., 2011). So, mice is used in order to impute the missing values.

The combination of the dependent- and independent variables leads to a data frame that is based on panel data. By using a panel dataset, all the data can be stored in one dataset since every entry is based upon a specific year. An overview of how the combination of this specific combination of variables looks like can be found in table 4.

Id Year Company # Respondents Industry Holding (1/0) Q1 Q2 Q3 Q4 Q5 Q6 Sales growth rate

1 2

Table 4: Dataset set-up

3.2 Descriptive analysis

After cleaning the data and deciding upon which variables are most important in this research, the variables were analysed. A summary of this analysis can be found in table 5. Moreover, this analysis is based only on the data entries related to a positive sales growth rate. At first sight, it becomes clear that there is a large difference in the number of respondents. The minimum number of respondents is 24, whereas the maximum number of respondents is 1960. Also, the standard deviation is rather large, 338.80. Accordingly, a histogram is created to investigate the distribution of the respondents variable. From this histogram (figure 6), it becomes clear that the number of respondents is far from normally distributed. Consequently, this distribution should be taken into account during further analysis since it might influence the customer experience score. Therefore, the models should be weighted by a log transformation of the respondents variable. Furthermore, the industry sizes do differ as well. For example, the retail industry is represented by a total of 59 companies, but telecom is only represented by 2 companies. Besides, the telecom industry is not represented in 2015, 2016 and 2017. One reason for the absence of the telecom industry in those years is that a negative sales growth rate was reported. Table 6 displays the total data available concerning the number of companies per year.

(22)

22

Variable Description Mean Median SD

Respondents Total amount of respondents for a specific company per year 268.36 131.67 338.80

Overall satisfaction Overall satisfaction score for a company in a specific year 7.53 7.57 .48

Obtrusiveness Obtrusiveness score for a company in a specific year 7.60 7.63 .40

Availability Availability score for a company in a specific year 7.55 7.62 .47

Honesty Honesty score for a company in a specific year 7.33 7.38 .53

Expectation management

Expectation management score for a company in a specific year

7.50 7.56 .46

Convenience Convenience score for a company in a specific year 7.56 7.64 .51

Growth 2017 The change in sales measured in percentages 7.21 4.83 6.41

Table 5: Descriptive statistics about the dependent variable and independent variables

Industry 2014 2015 2016 2017 Total Banking 6 5 2 3 16 Energy 1 1 2 2 6 Leisure 8 11 13 13 45 Online 4 5 5 3 17 Supermarket 14 17 12 13 56 Telecom 2 0 0 0 2 Insurance 13 15 10 8 46 Retail 9 16 19 15 59 Total 57 70 63 57 247

Table 6: Overview of the number of companies per industry per year

3.3 Multiple regression

In order to test the hypotheses, multiple regression will be used. This is an appropriate choice since the dependent variable consists of interval data and multiple independent variables are used to explain the dependent variable. The following formula represents the general model:

(1) y = β0 + β1Q1 + β2Q2 + β3Q3 + β4Q4 + β5Q5 + β6Q6 + β9Holdingi + β10No_holdingi +

β11Banking_industryi + β12Energy_industryi + β13Leisure_industryi + β14Online_industryi +

β15Supermarket_industryi + β16Telecom_industryi + β17Insurance_industryi +

β18Retail_industryi + β19Year_2017i + β20Year_2016i + β21Year_2015i + β22Year_2014i + ε

Where:

y = log-transformed dependent variable sales growth rate Q1-Q6: question 1 till question 6

i = 1 (yes) or 0 (no)

Moreover, all unknown parameters can be estimated by using ordinary least squares (OLS). This is one of the most common methods for estimating unknown parameters. The objective of OLS is to minimize the sum of squared residuals. By doing so, the estimated values should be close to the observed values across all observations (Leeflang, Wieringa, Bijmolt and Pauwels, 2013).

3.4 Multicollinearity

(23)

(Morrow-23 Howell, 2003). Since the independent variables are all related to each other since they all form the construct of the customer experience, there is a substantial possibility for multicollinearity. This is the case since 6 out of 9 variables included measure the same subject, namely the customer experience. Therefore, information captured by these variables can be redundant.

3.4.1 Correlation matrix

In order to test for multicollinearity, a correlation matrix is created first. This matrix determines the magnitude of correlation between every independent variable separately. Those correlations are found by regressing every independent variable on, in this research, every other independent variable (Morrow-Howell, 2013). Multiple guidelines are present on deciding upon which correlation is too high (Morrow-Howell, 2013). One rule of thumb is a cut-off point of .80 (Morrow-Howell, 2013, Lewis-Back, 1980). However, this threshold is extremely high, and no standard threshold exists since problems can also occur at lower correlation levels, such as .40 (Lewis-Back, 1980). As can be seen in table 7, the correlations range from .65 up to .89. Therefore, they indicate a strong positive correlation between the independent variables. This means that in this research a high score on question 4 will also result in a high score for question 6. Not only the correlation matrix shows the possible presence of multicollinearity, but all p-values also contribute to this finding. This is the case since all p-values are small and significant (p < .001).

Q1 Q2 Q3 Q4 Q5 Q2 .67 Q3 .88 .65 Q4 .85 .68 .78 Q5 .87 .70 .78 .87 Q6 .80 .66 .70 .88 .86 Table 7: Correlation matrix

3.4.2. VIF

Another method to detect multicollinearity is the use of the variance inflation factor (VIF). The Ri2

represents the proportion of variance in a specific independent variable that is associated with the variance of another independent variable in the same model (O’Brien, 2007). The VIF will be calculated as follows: 1/(1- Ri2). The VIF scores represent the proportion of variance shared with another

(24)

24 expected. However, the p-values of the correlation matrix are all significant (p < .001), and therefore, a Principal Component Analysis and Factor Analysis will be conducted in order to reduce the amount of correlating independent variables.

Q1 Q2 Q3 Q4 Q5 Q6

7.65 2.14 4.72 6.54 6.59 5.28

Table 8: VIF scores on sales growth

3.4.3 Principal Component Analysis

The goal of performing a Principal Component Analysis (PCA) is to extract the important information from the dataset. This information is transformed into linear combinations of the original variables and represents the so-called principal components (Abdi & Williams, 2010). From this Principal Components Analysis it becomes clear that 1 component explains 81.51% of the variance. Moreover, the other 5 components only explain between 6.97% and 1.5% of the variance. The summary of the PCA can be found in Appendix 3.

Moreover, in 1966, Cattell developed the scree test. The eigenvalues are plotted, after which the plot is examined for breaks or discontinuities. The underlying theory of this test is that a couple of major factors should account for the most variance. This is displayed in the scree plot by a deep cliff (the major factors), followed by the so-called scree representing the factors accounting for a small proportion of the variance. Therefore, the rule of thumb is to retain the number of components that do not belong to the scree (Hayton, Allen & Scarpello, 2004). A visual representation of the scree test based on the eigenvalues of question 1 till question 6 can be found in Appendix 4. Based on the aforementioned theory, only 1 principal component should remain in this particular case.

3.4.4 Factor Analysis

After the PCA, a Factor Analysis (FA) is performed in order to determine the final amount of factors used. Based on this FA, both 1 and 2 factors are appropriate to use in the final model (p < 0.001). However, in the case of using two factors, there are multiple variables with relatively high loadings on both factors. This can be seen in table 9 about factor loadings, where variables with high loadings on both factors are displayed in bold. Based upon the PCA, scree test, and FA, questions 1 till 6 can be combined into one overarching independent variable.

Q1 Q2 Q3 Q4 Q5 Q6

(25)

25 The scores mentioned in table 9 are all factor loadings. Those factor loadings explain the amount of correlation between the independent variable and the latent factor. In order to calculate the amount of variance of a specific variable the factor accounts for, the factor loading should be squared. Therefore, the created factor accounts for 86.12% of the variance of Q1 (.928^2 = 86.12%). Since all factor loadings differ from each other, the created factor is able to account for differences in the importance of the underlying variables. The higher the factor loading, the higher is the importance of that specific variable. An overview of the explained variances can be found in table 10.

Q1 Q2 Q3 Q4 Q5 Q6

Variance explained 85.75% 54.61% 72.93% 87.05% 87.98% 80.28% Table 10: Factor analysis - total variance explained

Even though it is possible to create a new variable that takes the different weights from table 10 into account, a new variable is created without taking into consideration the different weights of every component. The main reason is the ease of use. A new variable based on table 10 will also generate negative values, and hence cannot be used in further analysis. This new variable is called the average customer experience score (Avg_CE), and represents the unweighted average of questions 1 till 6. In table 11, an overview of the descriptive statistics regarding this new variable can be found.

Min. 1st Qu. Median Mean 3rd Qu. Max.

Avg_CE 5.322 7.150 7.467 7.435 7.726 8.571

Table 11: Descriptive statistics of Avg_CE

3.5 Normality and heteroscedasticity

Besides the assumption of the absence of multicollinearity, there are three other relevant assumptions when OLS is used. Namely, a homoscedastic error term, normality in the residuals and the absence of autocorrelation. However, autocorrelation will not be checked. In panel data, the presence of autocorrelation is common since it involves time series data. As mentioned before, the years available in this study are also independent variables in order to check for a trend and therefore the residuals can exhibit autocorrelation (Halcoussis, 2005).

3.5.1 Heteroscedasticity, non-normality

(26)

26

3.6 The final models

Based upon the multicollinearity problem discussed before, the final model will look as follows: (2) y = β0 + β1Avg_CE + β2Holdingi + β3No_holdingi + β4Banking_industryi + β5Energy_industryi +

β6Leisure_industryi + β7Online_industryi + β8Supermarket_industryi + β9Telecom_industryi +

β10Insurance_industryi + β11Retail_industryi + β12Year_2017i + β13Year_2016i + β14Year_2015i

+ β15Year_2014i + ε

Where:

y = log-transformed dependent variable sales growth rate Q1-Q6: question 1 till question 6

i = 1 (yes) or 0 (no)

Model (2) is used to test the first and fifth hypothesis. In order to test the second hypothesis, model (1) will be used. Hypothesis 2 takes into consideration the influences of all questions separately. Since this model suffers from multicollinearity, the interpretation of the results should be done with caution. Hypothesis 3 can be tested with a model almost similar to the model of hypothesis 1. The only difference is that the Avg_CE variable will be replaced by question 1 (Q1). Question 1 is the variable measuring overall satisfaction and differs from Avg_CE to such an extent that the other questions related to the customer experience are not taken into account. This results in model (3), and looks as follows:

(3) y = β0 + β1Q1 + β2Holdingi + β3No_holdingi + β4Banking_industryi + β5Energy_industryi +

β6Leisure_industryi + β7Online_industryi + β8Supermarket_industryi + β9Telecom_industryi +

β10Insurance_industryi + β11Retail_industryi + β12Year_2017i + β13Year_2016i + β14Year_2015i

+ β15Year_2014i + ε

Where:

y = log-transformed dependent variable sales growth rate Q1-Q6: question 1 till question 6

i = 1 (yes) or 0 (no)

(27)

27 (4) y = β0 + β1Avg_CE + β2Avg_CE2 + β3Holdingi + β4No_holdingi + β5Banking_industryi +

β6Energy_industryi + β7Leisure_industryi + β8Online_industryi + β9Supermarket_industryi +

β10Telecom_industryi + β11Insurance_industryi + β12Retail_industryi + β13Year_2017i +

β14Year_2016i + β15Year_2015i + β16Year_2014i + β17D_=<7i + β18D_>7i + ε

Where:

y = log-transformed dependent variable sales growth rate Q1-Q6: question 1 till question 6

i = 1 (yes) or 0 (no)

D_=<7 = the customer satisfaction score below 7 D_>7 = the customer experience score of above 7

In order to test the industry effect, an interaction should be added. This interaction will take place between every industry dummy and the Avg_CE. The last model of this paper is as follows: (5) y = β0 + β1Avg_CE + β2Holdingi + β3No_holdingi + β4Banking_industryi + β5Energy_industryi +

β6Leisure_industryi + β7Online_industryi + β8Supermarket_industryi + β9Telecom_industryi +

β10Insurance_industryi + β11Retail_industryi + β12Year_2017i + β13Year_2016i + β14Year_2015i

+ β15Year_2014i + β16(Banking_industryi * Avg_CE) + β17(Energy_industryi * Avg_CE) +

β18(Leisure_industryi * Avg_CE) + β19(Online_industryi * Avg_CE) + β20(Supermarket_industryi

* Avg_CE) + β21(Telecom_industryi * Avg_CE) + β22(Insurance_industryi * Avg_CE) +

β23(Retail_industryi * Av_CE) + ε

Where:

y = log-transformed dependent variable sales change Q1-Q6: question 1 till question 6

i = 1 (yes) or 0 (no)

Below, in table 12, an overview can be found regarding which model is used to test a specific hypothesis. Model Hypothesis/hypotheses 1 2 2 1 3 3 4 4 5 5

(28)

28

4. Results

This chapter presents all the outcomes related to the tested hypotheses. In table 14, an overview with all the estimation results and their significance levels can be found. All results discussed next are based upon this table. At the end of the first section, a final overview of the supported and rejected hypotheses can be found. This part will be followed by a validation of the models used.

4.1 The effect of the customer experience on sales growth

The overarching question in this research is whether the customer experience has an effect on sales growth rate. First of all, the model used to test this, is model (2). According to the outcomes of this model the effect of the customer experience on sales growth rate is not significant (β = - .14283, p = .3841). Therefore, hypothesis 1 is rejected. Moreover, an interesting effect is present. Even though the aforementioned relationship is not significant, the parameter estimate is negative (-0.14283). So, if significant, an increase in the customer experience would lead to a decrease in sales growth rate. This effect is opposed to the effect suggested by literature. Therefore, a nonlinear effect will be investigated later on.

4.2 The effect of the Golden Rules on sales growth rate

Since the customer experience is based on 6 different questions in this study, it is interesting to investigate which one of those questions has the largest influence on sales growth rate. In order to test this, model (1) is used. Accordingly, it becomes clear that the average satisfaction score and every Golden Rule have an insignificant effect on sales growth rate. Therefore, based on the output in table 14, hypothesis 2 is rejected. This can be caused by the presence of multicollinearity. According to section 3.4, multicollinearity is present between the 6 different questions used in this model. Multicollinearity can lead to two situations: inflated standard errors and a reduction in the magnitude of the parameter estimates. First of all, a high correlation between variables causes larger standard errors what reduces statistical power. Also, the parameter estimates are reduced since multiple variables try to claim the same share of the dependent variable, sales growth rate (Morrow-Howell, 1994). Therefore, the results of model (1) cannot be interpreted appropriately.

4.3 Single-item scales versus multi-item scales

(29)

29 negative. Therefore, the kind of relationship between the average satisfaction score and sales growth rate is the same as for the customer experience. Since the average satisfaction score is found insignificant, hypothesis 3 is rejected.

4.4 The nonlinear relationship between the customer experience and sales

growth rate

As already became clear from testing the relationship between the customer experience and sales growth rate, a nonlinear relationship between the customer experience and sales growth rate might be present. In order to test this, model (4) in table 14 is used. First of all, this model uses the unweighted average customer experience score as well as the squared values of this score. The unweighted average customer experience score is used since it is prevented from multicollinearity. It becomes clear that both have a significant effect on sales growth rate. The non-squared values of the customer experience score indicate a positive relationship (β = 9.1430, p = .02259). The squared-values of the customer experience have a negative sign (β = -.5984, p = .02538). This can be interpreted as follows: an increase in the customer experience will also lead to an increase in sales growth rate. However, due to the negative sign of the squared values, the sales growth rate will decrease at a certain point. This results in an inverted u-shape. Figure 7 visualizes the aforementioned relationship. Figure 7 also indicates an inverted u-shape relationship. A customer experience score between 6 and approximately 6.6 will lead to an increase in the sales growth rate. At a score of 6.6 the top of the inverted u-shape can be found. Therefore, a customer experience score between 6.61 and 7.8 will lead to a decrease in sales growth rate. Nevertheless, as figure 7 displays, when the customer experience score is higher than a 7.8 an increase in sales growth rate is realised again. However, this effect was not taken into account during the hypotheses stage. Due to the use of a log-transformed growth sales rate variable and squared customer experience score, the coefficients cannot be interpreted. Moreover, another indicator of nonlinearity is present. When companies move from a customer experience score of above 7 towards a score below 7, they will obtain an increase in their sales growth rate. In the end, it can be concluded that hypothesis 4 is partially accepted since the inverted u-shape relationship is present, but only in the case of a customer experience score between 6 and 7.8.

(30)

30

4.5 The effect of the customer experience on sales growth rate across industries

In order to fill the gap regarding the industry effect, the different industries present in the database are also taken into consideration. This is tested by the use of model (5) in table 14, since this model includes interactions between the customer experience score and the different industries. Based on this model, it can be concluded that an interaction effect is present. More specifically, this interaction effect is significant in the leisure industry (β = .931777, p = .0448). Therefore, the effect of the customer experience is exp(.931777)-1*100% = 154% higher in the leisure industry as opposed to the retail industry. Also, it might be assumed that purchases in the leisure industry is of higher importance compared to purchases in the retail industry. Therefore, hypothesis 5 is accepted.

Table 13 represents a summary of the results discussed above.

Supported Remarks

Hypothesis 1 ✘ The non-significant

relationship between the customer experience and sales growth rate is negative

Hypothesis 2

Hypothesis 3

Hypothesis 4 Partially supported Partially supported. An

inverted u-shape relationship is present for only a specific range of the customer experience score

Hypothesis 5 ✔ There is only a statistically

significant interaction found for the leisure industry

(31)

31

Model 1 Model 2 Model 3 Model 4 Model 5

Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|) Intercept 2.90417 .0481 * 2.98625 .0180 * 2.97153 .00677 * -33.1837 .02727 * 6.746740 .0191 * Independent variables The customer experience Avg_CE -.14283 .3814 9.1430 .02259 * -.659075 .0816 . Avg_CE2 -.5984 .022538 * Grade ≤7 .8780 .00436 ** Grade > 7 NA NA Q1 -.40070 .3867 -.14154 .31809 Q2 -.23474 .3589 Q3 .16180 .6742 Q4 -.10626 .7733 Q5 .30913 .4210 Q6 .11318 .7413 Industry Banking -.31021 -.3079 -.30063 .2944 -.30408 .28836 -.3997 .16503 -4.561656 .3126 Energy .16373 .7089 .15586 .7104 .14415 .73158 .3064 .46610 -42.756365 .1008 Leisure -.16170 .4963 -.20777 .3184 -.19141 .35635 .0402 .85200 -7.179819 .0390 * Online .40890 .2042 .44511 .1280 .46197 0.46197 .4800 .09910 . -.282012 .9515 Supermarket -.48268 .0145 * -.47138 .0134 * -.47248 .01284 * -.4289 .02254 * -1.874965 .7197 Telecom -.77984 .2762 -.77357 .2720 -.77493 .27080 -.6746 .33247 -1.457361 .9716 Insurance -.32078 .2110 -.32093 .1214 -.32470 .11717 -.2447 .23323 -3.527040 .3728 Retail NA NA NA NA NA NA NA NA NA NA Years 2014 -.17890 .4081 -.24571 .2127 -.22604 .24112 -.2327 .23618 -.318608 .1118 2015 .03075 0.8672 .02278 .9000 .01928 .91533 -.0105 .95320 -.006511 .9714 2016 -.19591 .2954 -.19673 .2896 -.19604 .29113 -.2515 .17158 -.239336 .2016 2017 NA NA NA NA NA NA NA NA NA NA

(32)

32

Model 1 Model 2 Model 3 Model 4 Model 5

Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|) Estimate Pr(>|t|)

Holding -0.17332 -.17064 .2595 -.16820 .26509 -.1229 .41195 -.145114 .3627 No holding NA NA NA NA NA NA NA NA NA Interactions Banking:Avg_CE .569407 .3545 Energy:Avg_CE 5.860772 .1005 Leisure:Avg_CE .931777 .0448 * Online:Avg_CE .098034 .8734 Supermarket:Avg_CE .192234 .7789 Telecom:Avg_CE .077699 .9888 Insurance:Avg_CE .426614 .4223 Retail:Avg_CE NA NA

Residual std. error 2.28, 229 degrees of freedom 2.264, 234 degrees of freedom 2.263, 234 degrees of freedom 2.229 232 degrees of freedom 2.26 227 degrees of freedom R2 0.0887 0.08133 0.08223 0.1172 0.1122 Adj. R2 0.02105 0.03421 0.03517 0.06395 0.03791 P-value 0.1864 0.06219 0.05814 0.06619 0.08318

Table 14 Estimation results

(33)

33

4.6 Additional analysis

Even though no hypotheses are developed regarding the year of measurement or whether the company belongs to a holding or not, this will still be tested. The model used is the same as model (5), but also adds interactions between the year of measurement, the unweighted average customer experience score and between whether or not the company belongs to a holding. The output can be found in Appendix 5. However, it becomes clear that there are no extra significant interactions apart from the interactions tested under hypothesis 5. So, there is no difference in magnitude of the effect of the customer experience on sales growth when a company belongs to a holding or not. Also, there are no significant interaction effects for the different years used in this study. Therefore, the size of the effect of the customer experience on sales growth rate does not differ across years.

4.7 Validation

In order to demonstrate that the assumptions made by the models are reasonable and that the models are an accurate representation of reality, the models should be validated (Hillston, 2003). Therefore, r-squared and adjusted r-squared will be discussed, the total significance of the model as well as the information criteria.

4.7.1 R-squared

The r-squared and adjusted r-squared values of all 5 models can be found in table 15. From this table, it becomes clear that based on the r-squaredvalues, model (4) has the best fit. However, the r-squared does not take into account the number of parameters estimated. Since models (1-5) differ from each other regarding the number of parameters they capture, the adjusted r-squared is a better measure of model fit. Based on the adjusted r-squared, it becomes clear that model (5) has the best fit. In other words, model (5) has the highest explanatory power. Nevertheless, all squared and adjusted r-squaredvalues are rather low. This is due to the fact that more independent variables are needed to explain the total variance and to increase the explanatory power of the model. Indeed, an increase in sales growth rate is caused by more than the customer experience only. For example, there are multiple firm-level characteristics not taken into consideration, such as advertising intensity and market share. Also, multiple industry characteristics are not accounted for, such as demand growth (Morgan & Rego, 2006).

R-squared Adjusted r-squared

Model 1 0.0877 0.02105

Model 2 0.08133 0.03421

Model 3 0.08223 0.03517

Model 4 0.1172 0.06395

Model 5 0.03791 0.08318

(34)

34

4.7.2 Overall model fit

In order to test whether the models are significant overall, the F-test is used. The corresponding p-values of every model can be found in table 16.

P-value Model 1 0.1864 Model 2 0.06219 Model 3 0.05814 Model 4 0.008493 Model 5 0.08318

Table 14 P-values of overall model fit

Based on table 16, it becomes clear that only model (4) is significant at a significance level of .05 (p=.008493). Moreover, under a significance level of 0.1 models (2, 3 & 5) are significant. However, model 4 has a p-value much lower than the other models. The fact that the other models are only significant using a higher significance level compared to model (4) is related to the estimation results in table 14. The F-test assess the significance of all coefficients together and it is due to the fact of few significant parameters in models (2, 3 & 5) that the F-tests do not perform very well.

4.7.3 Information criteria

Another method for model comparison is the so-called information criteria. Information criteria are used to find the model that is the best approximation to reality (Leeflang et al., 2013). Multiple information criteria exist, such as the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). The difference between AIC and BIC is the weight of the parameter penalty. For BIC, this penalty is larger. As a result, BIC will opt for the most parsimonious model (Leeflang et al., 2013). An overview of the AIC and BIC scores of models (1-5) can be found in table 17.

AIC BIC Model 1 424.3913 799.9494 Model 2 416.3825 774.3937 Model 3 416.1386 774.1498 Model 4 410.5381 775.5680 Model 5 421.9346 804.5115

Table 15: Overview of information criteria AIC and BIC per model

(35)

35

5. Conclusion

5.1 Discussion

In 1954, Peter F. Drucker already emphasised on the importance of customers. Customers are necessary for companies in order to survive and are therefore able to determine the success rate of a company. Later on, CFMs were developed in order to keep track of customers. They are able to measure how satisfied customers are with a company, or how likely they are to repurchase. In 1994, it was already proven that companies who focus on the customer satisfaction CFM are rewarded with superior financial performance (Anderson, 1994). However, over the years, more and more CFMs were created. Due to the existence of so many CFMs, companies lack knowledge on which CFM to use. Also, frustration and confusion sets in and eventually, the accountability of the marketing department can be affected. In order to keep companies back on track by refocusing on their customers again, SAMR introduced a new CFM. This CFM comprises the customer experience. Since it is interesting to know whether this CFM can also be used as an indicator of financial performance, the following problem statement came to mind: To which extent is it possible to use the customer experience as an indicator

of financial performance?

In order to answer this question, a dataset on the customer experience owned and created by SAMR was used. This dataset was complemented by a dataset containing sales numbers of the companies included in the dataset of SAMR. Those sales numbers were obtained through Orbis, a financial database containing information of more than 300 million companies around the world. The financial performance measure used was sales growth rate, developed by calculating the percentage change in sales numbers. Overall, it was found that the customer experience could be used as an indicator of sales growth rate only under specific circumstances.

This conclusion is based on several research questions, all related to the problem statement. The first research question was very broad: Is it possible to use the customer experience as an indicator of sales

growth? The main answer to this question is no. Out of the 5 performed models, only models (4) and

(5) showed a significant effect. Besides, the effect found in model (5) is only significant under the p <.1 level and both models have a low adjusted r-squared value (0.06395 and 0.08318, respectively). Many existing CFMs use a single-item measurement scale, such as for example customer satisfaction (Grønholdt et al., 2000). However, it was suggested by literature to use multi-item scales in order to increase the predictive validity and to reduce measurement error (Diamantopoulos & Sarstedt, 2012, Gliem & Gliem, 2003). Therefore, the following research question was stated: Is a multi-item scale

(36)

36 From this study, it became clear that a multi-item scale does not perform better than a single-item scale since both models were insignificant. So, nothing can be said about which model performs better. Since the customer experience exists of several components, it was interesting to find out which customer experience aspect was the most important one. Therefore, another research question was stated: Which CFM aspect has the most explaining power in its effect on sales growth? The model used to test this came out insignificant. Therefore, no one aspect has an effect on sales growth rate. However, in this model, multicollinearity is present. All questions are highly correlated with each other, and therefore, a factor analysis proved that it was sufficient to combine all questions into one unweighted average score. Also, multicollinearity could have been the cause of no significant estimates.

According to literature, the relationship between the customer experience and sales growth should be nonlinear. At a certain point, a saturation effect is suggested (Ittner & Larcker, 1998). This can be caused by for example putting too much pressure on creating the best the customer experience. Moreover, oversaturation can take place at this point, even leading to a decrease in sales. So, another research question became: Does a nonlinear relationship between the customer experience and sales

growth exist? At this point, it became clear that the customer experience has an influence on sales

growth. As was suggested, a nonlinear relationship between the customer experience and sales growth exist. This was confirmed by both the statistical significance of the customer experience score and the addition of its squared values. Since the estimate of the customer experience has a positive sign, it indicates that an increase in the customer experience leads to an increase in the sales growth rate. Also, the negative sign of the estimate corresponding to the squared values of the customer experience represents a concave shape of the relationship. In addition, what became clear was that moving from a customer experience score of above a 7 to a score below 7 increases the sales growth rate. Even though the presence of an inverted u-shape relationship is accepted, based upon the corresponding plot it became clear that this relationship is not present over the full range of the customer experience score. After the downward slope of the inverted u-shape, an increasing trend is observed again. Therefore, companies with a customer experience score between 6.61 and 7.8 will face a decrease in sales growth rate opposed to companies scoring between a 6 and 6.6 or above an 7.8.

Referenties

GERELATEERDE DOCUMENTEN

The index of frailty - The KPS scale - can be used as a predictor of 1-year post-transplant outcomes (patient and graft survival) in a national United States liver

e moderator effect of ADHD was only signi�cant for the teacher data; children with higher levels of ADHD showed a weaker relationship between peer problems and prosocial

Hierbij zal in het bijzonder in worden gegaan op de grondslag, de duur, de mogelijkheid van het opnemen van alimentatie in huwelijkse voorwaarden en de beëindiging

Moreover, the market betas of the portfolios with high customer satisfaction results (both based on relative and absolute ACSI scores) are considerably lower compared

Predictors: (Constant), Inkomenssituatie, Supermarkttrouw (0-10), Winkel, CHOCTASTE, PENPICK, Dag, Boodschapper, Aantal personen in huishouden, NULCONDITIE, Frequentie

When performing a regression analysis on the whole model, with all the factors and their components and the moderating effects included, only responsiveness of

In addition, we therefore analyzed the effects a more hedonic brand attitude has on the individual components of Customer Performance, which showed that a brand store with a