The Impact of Channel Usage on Customer Churn in the Telecommunications Industry

The Impact of Channel Usage on

Customer Churn in the

Telecommunications Industry

The Impact of Channel Usage on

Customer Churn in the

Telecommunications Industry

Ilona Bosma

Barestraat 21-8

9725 CM Groningen

Tel: (+31) 06 24 35 25 75

E-mail: I.N.Bosma@student.rug.nl

Student number: S2744023

MSc Marketing Intelligence and MSc Marketing Management

June 17, 2019

First supervisor:

Prof. dr. J.E. Wieringa

Second supervisor:

Dr. J.T. Bouma

University of Groningen

Faculty of Economics and Business

Department of Marketing

Abstract

Customer churn is negatively impacting firms by reducing market shares and increasing costs. Therefore, many firms are active in churn management and churn prediction. In order to make accurate predictions and successfully manage customer churn, identifying and understanding churn drivers is crucial. Moreover, it is important to understand how a customer’s channel choice impacts these drivers. However, little is known about these effects and these gaps in knowledge prevent firms from successfully managing churn. Therefore, this paper conducts a systematic investigation on the drivers of customer churn and the impact of channel choice on these drivers. In doing so, telephone surveys for qualitative insights are combined with recent transaction data from a large retailer in the Dutch telecommunications industry. A binary logistic regression model is used to predict customer churn and its underlying drivers.

Results show that it is important to include channel choice in churn prediction models for firms. Next to the relevance of characteristics related to firm’s marketing efforts, the customer-firm relationship and the customer itself, the paper proves the importance of channel choice when it comes to predicting and understanding churn and its drivers. This provides opportunities for firms that know how to account for this in their churn management and prediction strategy. Doing so increases the predictive ability of the models and enables firms to better identify which individuals to target with their retention efforts.

Preface

This thesis represents the end of the master Marketing Management and Intelligence at the University of Groningen. The research that is described in this thesis took place between February and June 2019 and is combined with a master thesis internship at an anonymous retailer in the Netherlands. I would like to express my gratitude to several individuals that contributed to the successful completion of my master thesis project.

First, I would like to thank the focal retailer of this research for the opportunity to combine writing my master thesis with an internship at the firm. The firm provided all the help and data necessary for a successful completion of this research. Moreover, I would like to thank all my colleagues at the firm for their help and support in writing this thesis.

Furthermore, I would like to thank my first supervisor prof. dr. Jaap Wieringa for his guidance and constructive feedback during the process of completing this thesis. Finally, I would like to thank my second supervisor dr. Jelle Bouma for his help in finding a suitable master thesis internship and his guidance on how to successfully complete it.

I hope you enjoy reading this thesis as much as I enjoyed writing it.

Ilona Bosma

Table of Content

2.1 Customer Churn Definition 7

2.2 Literature Review 7

2.2.1 Morgan and Hunt (1994) 8

2.2.2 Dick and Basu (1994) 9

2.2.3 Keaveney (1995) 10

2.2.4 Bolton, Lemon and Verhoef (2004) 12

2.2.5 Bansal, Taylor and St. James (2005) 13

2.2.6 Literature summary 15

2.3 Drivers of Customer Churn 16

2.3.1 Relationship aspects 17

2.3.2 Customer characteristics 19

2.3.3 Marketing instruments 21

2.4 Conceptual Model 25

3. Data Collection 26

3.1 Telecommunications Market in the Netherlands 26

3.2 Outcomes Qualitative Research 27

3.3 Alterations to Conceptual Model 27

4. Methodology 39 4.1 Sample 39 4.2 Model Estimation 40 4.3 Model Specification 41 4.4 Model Comparison 42 4.5 Multicollinearity 43 5. Results 44

5.1 Results Logistic Regression 44

5.1.1 Relationship aspects 45 5.1.2 Customer characteristics 48 5.1.3 Marketing instruments 49 5.1.4 Channel choice 50 5.2 Classification Techniques 52 5.2.1 Logit model 53 5.2.2 Classification tree 53

5.2.3 Ensemble learning methods 53

5.2.4 Neural network 54 5.2.5 K-nearest neighbors 55 5.2.6 Conclusion 55 6. Discussion 57 6.1 Findings 57 6.1.1 Relationship aspects 57 6.1.2 Customer characteristics 59 6.1.3 Marketing instruments 59 6.2 Academic Implications 62 6.3 Managerial Implications 63

7. Study Limitations and Suggestions for Future Research 65

References 67

1. Introduction

1.1 Introduction

Customer churn has become an important topic for researchers and practitioners in the marketing field. Nowadays, many firms face the problem of a growing number of defecting customers. Given that firms derive a large part of their revenues from creating and sustaining long-term relationships with their customers, this is problematic (Kumar & Shah, 2009).

Customer churn is unambiguously an important issue in the telecommunications market, in which annual churn rates range between 15-30% (Ascarza, Iyengar & Schleicher, 2016). The market is characterized by fierce competition, new technologies, low switching costs and deregulation by governments, which lead to churn being a continuous problem for managers (Kirui, Hong, Cheruiyot & Kirui, 2013). Moreover, the market is getting increasingly saturated which further intensifies the competition and puts firms under pressure to retain customers (Baker, Sciglimpaglia & Saghafi, 2010). As a result, the churn-related costs for firms active in the market are up to $10 billion annually (Ascarza et al., 2016).

At the same time, the interaction between the customer and the firm has changed dramatically. Nowadays, customers have access to an increased number of information and interaction channels that are often used interchangeably or even simultaneously (Verhoef, Kannan & Inman, 2015). Especially the rise of the online channel empowers customers and makes them more informed about the products and services that they are purchasing (Arora, Singha & Sahney, 2017).

Moreover, consumers have increased awareness about alternative options and switching opportunities in the market, given that online competitive information is only one mouse click away. These developments lead to a growing number of defecting customers and increasing churn rates (Holtrop, Wieringa, Gijsenberg & Verhoef, 2017; Lejeune, 2001). That is, consumers’ greater access to information and growing capacity to choose the best option, makes the customer churn problem more serious for firms (Antón, Camarero & Carrero, 2007).

However, complete knowledge about these churn drivers is currently lacking. That is, even though many firms start to realize the importance of actively managing churn, this cannot be done optimally due to inadequate knowledge about what drives churn. Also, there are gaps in knowledge on whether and how customer churn drivers differ per channel, while prior research shows that customer characteristics and customer behavior are significantly different per channel (e.g. Nesar & Sabir, 2016).

Ignoring these differences and failing to accurately identify the drivers of customer churn leads to erroneous conclusions and suboptimal spending of incentive money. This could have detrimental effects on firm performance, given that churn-related costs are substantial (Neslin et al., 2006; Ascarza et al., 2016). Additionally, if a firm fails to understand why its customers churn, it is not able to take the appropriate actions to prevent this (Braun & Schweidel, 2011).

Therefore, this paper tries to fill the existing gaps in knowledge by addressing the topic of channel choice in relation to customer churn prediction. Understanding these issues is assumed to be a prerequisite for successful churn management.

1.2 Research Purpose

This study aims to find information on the drivers of churn in the telecommunications industry, and how these drivers are influenced by different off- and online channels. A better understanding of how churn drivers are influenced by channel usage enables telecommunication firms to allocate their marketing resources and activities more effectively. Therefore, this paper answers the following research question:

Which drivers of churn are important to consider in the Dutch telecommunications market and how does the usage of different off- and online channels affect this churning behavior?

To be able to answer this research question, the following sub questions are answered:

- What are important determinants of customer churn in the Dutch telecommunications market?

- How is customer churn behavior different between online and offline customers?

both qualitative and quantitative data to have maximum information about consumer motivations and derive the most valuable insights.

The performed analyses provide evidence for several direct drivers of customer churn, such as different relationship aspects, customer characteristics and marketing instruments. Most importantly, the results show that the impact of most of these drivers differs significantly between online and offline customers.

1.3 Contribution to Literature

It is not surprising that the topic of customer churn has been extensively researched in marketing literature, given the topic’s managerial and theoretical relevance. There are different studies that focus on customer switching or customer churn in the telecommunications industry (e.g. Gerpott, Rams & Schindler, 2001; Kim, Park & Jeong, 2004; Wieringa & Verhoef, 2007). However, none of the above-mentioned papers considers customer channel choice as a possible explanatory factor of customer churn.

The drivers of channel choice have been researched separately by several researchers (e.g. Arora et al., 2017; Chang & Zhang, 2016; Neslin et al., 2006). Most importantly, this previous research identifies that there are significant differences in customer characteristics of customers using a specific channel (e.g. Nesar & Sabir, 2016). Therefore, it is likely that there are differences in customer churn behavior between different channels as well.

This paper tries to fill the gap in knowledge about the relation between channel choice and churn. It tries to get a thorough understanding of customer behavior and its drivers by using various data sources. In this manner, useful insights related to motivational drivers and customer considerations can be derived. Therefore, this paper has several contributions to the existing literature.

First, the current paper is the first to focus on channel choice in relation to customer churn and its drivers. In doing so, the current research builds on prior research findings that focus on churn drivers or channel choice separately. This paper extends and combines these findings into an encompassing framework in which both drivers and channels are included. Consequently, a more comprehensive customer churn framework is built.

should certainly be related, the strength of the relationship between churn intentions and churn behavior is a much-debated subject (Chandon, Morwitz & Reinartz, 2005).

Given this, it is questionable that previous research often does not distinguish between intentions and behavior or assumes that both have the exact same drivers. This paper directly examines and predicts the drivers of customer churn behavior irrespective of the strength of the intentions-behavior link.

Additionally, existing literature often proxies customer churn behavior by focusing on concepts such as customer loyalty. While churn behavior might be a part of customer loyalty, the concept of customer loyalty is much broader defined. This means that the drivers of loyalty (e.g. measured as word-of-mouth spreading behavior) are not necessarily the same as the drivers of churn behavior.

The current paper’s main contributions are therefore that its research framework is more extensive and that it can measure churning behavior directly rather than deriving it from a proxy or an intention-related construct. That is, actual churn drivers can be derived that help firms to predict future churning customers.

1.4 Managerial Relevance

Given the impact that customer churn has on firm performance through reduced market share, impaired profitability and increased costs (Reichheld & Sasser, 1990), it is important for managers to have an efficient and adequate churn management strategy. Of main interest is customer retention, since retaining current customers is less expensive and more beneficial for firm performance than acquiring new customers (Verbeke, Dejaeger, Martens, Hur & Baesens, 2012; Ascarza et al., 2016).

Additionally, in a saturated market such as the telecommunications industry, attracting new customers is difficult and costly, which puts further emphasis on customer retention (Kim et al., 2004; Lee, Kim & Lee, 2017). In light of this insight, many firms active in the industry recognize the need to move from a product-centric strategy to a customer-product-centric one (Verbeke et al., 2012), realizing that their current customer database is their most valuable asset (Athanassopoulos, 2000; Jones, Mothersbaugh & Beatty, 2000; Thomas, 2001). Therefore, the spending on focused and targeted marketing actions is increasing with the goal of maximizing customer satisfaction and loyalty to increase retention (Burez & Van den Poel, 2007).

customers have certain characteristics that are preferential, such as a lower price sensitivity (Turel & Serenko, 2006), higher amount of positive word-of-mouth generation (Sasser, Schlesinger & Heskett, 1997; Reichheld & Sasser, 1990), greater usage of the firm’s services (Bolton & Lemon, 1999) and greater resistance to competitors’ persuasion attempts (Dick & Basu, 1994).

Moreover, Gupta, Lehmann and Stuart (2004) conclude that focusing on retention is more beneficial in terms of firm value than, for example, increasing profit margins or lowering acquisition costs. More specifically, Roofthooft (2010) identifies that an increase in retention equal to 5% can improve the company’s bottom line by 25-80% and that a retention rate increase of 2% has the same effect on firm performance as a 10% cost decrease. Additionally, Gupta and Lehmann (2003) show that a small increase in retention leads to significant increases in customer lifetime value (CLV). Given the benefits that customer retention can bring to the firm, customer churn management programs are increasingly used, and their successful implementation is crucial.

Previous research by Burez and Van den Poel (2007) identified two targeted churn management approaches that firms commonly use, being a reactive and proactive response, respectively. A reactive strategy implies that the firm provides the customer incentives to stay as soon as the customer requests the cancellation of the contract. On the other hand, a proactive approach relates to identifying customers that are likely to churn and target these customers before they cancel their contract.

Especially in the case of proactive churn management, it is important that churn predictions are accurate to avoid that incentive money is spent on the wrong customers. This study focuses on identifying the drivers for churn and the customer behavior that reflects these drivers to improve the accuracy of firms’ churn management.

Special attention is paid to how the decision to churn is impacted by the channel that a customer uses in the purchasing process, thereby distinguishing between online and offline channels. Research by Verhoef and Donkers (2005) indicates that customer characteristics and churn rates differ depending on the channel a customer uses. Consequently, channel usage might be an important explanatory factor to consider in churn management, that, to my knowledge, has not been used in research before.

how to most effectively allocate scarce marketing resources across individuals and channels. In this manner, the benefits that successful retention management offers to the firm can fully be used.

1.5 Thesis Structure

2. Theoretical Framework

This chapter focuses on the theoretical background and explanation of the concepts that are used in this research. It also clarifies the boundaries and focus of the current research. First, customer churn, the dependent variable of this research, is defined. Afterwards, this chapter provides the hypotheses and research framework of this study based on a review of the current literature.

2.1 Customer Churn Definition

In this research, customer churn is investigated in a contractual setting, given that customers are only able to terminate their contract with a service provider after the expiration date of the contract. Customer churn in this setting is defined as ‘the termination or defection of the contract between the customer and the company’ (Leeflang, Wieringa, Bijmolt & Pauwels, 2015, p. 321).

In further classifying types of churn, the approach as initiated by Lee et al. (2017) is followed. That is, churn is subdivided into internal and external churn, since these have different drivers and different degrees of influenceability by the firm. In this light, external churn is explained as customer churn that is either due to non-voluntary variables such as death or voluntary variables such as a change of a service provider (Kaya & Williams, 2005). On the other hand, internal churn is defined as a change of service type, such as moving from a prepaid to a postpaid mobile subscription (Lee et al., 2017).

For the purpose of this research, the focus is on voluntary external churn. With this type of churn, the customer voluntarily terminates the contractual relationship that it has with the service provider. Given that the service provider has an influence on the customer decision to voluntarily leave the company, this type of churn is most relevant and most interesting from a churn management perspective.

Furthermore, churn is measured based on actual churn behavior rather than intentions. Customer churn relates to a binary classification problem, since customers either churn or do not churn. In this sense, the churn definition in this thesis differs from related constructs such as repeat purchase, where the absolute number of purchases is often considered.

2.2 Literature Review

only, these are still considered important as a starting point. The main goal of this section is to structure the enormous amount of churn-related literature and to identify the most prominent drivers of customer churn.

2.2.1 Morgan and Hunt (1994)

Morgan and Hunt (1994) focus on loyalty and churn behavior from the perspective of relationship marketing, which is a type of marketing that aims at establishing, developing and maintaining long-term relationships. The framework as proposed in the paper relates to the relationship between a customer and a firm as well as to relationships in a broader sense, reflecting the broad application opportunities of the framework. The authors consider customer churn to be an element of the relationship between customers and firms.

Morgan and Hunt (1994) argue that there are two important characteristics that can be used to distinguish successful relationships from unsuccessful ones. These concepts are trust and relationship commitment. The authors define trust as ‘a willingness to rely on an exchange partner in whom has confidence’ (Morgan & Hunt, 1994, p. 23). Relationship commitment is defined as ‘an enduring desire to maintain a valued relationship’ (Morgan & Hunt, 1994, p. 23). The authors stress that both trust and commitment should be present - not just one or the other - to lead directly to cooperative behaviors that are crucial for relationship success.

Figure 1. Morgan and Hunt (1994) framework

a form of commitment related to emotions, calculative commitment is of a more financial nature. Again, this stresses the wide applicability of the Morgen and Hunt (1994) framework.

2.2.2 Dick and Basu (1994)

Dick and Basu (1994) discuss the concept of customer loyalty and its components. The authors note that there are two prerequisites of real loyalty, being repeat purchase behavior and favorable attitudes. It is argued that true loyalty can only occur when both repeat purchase behavior and favorable attitudes are present.

The authors identify four different types of loyalty, in a 2-by-2 matrix with either high or low attitude and high or low repeat patronage. For example, they distinguish latent loyalty, a situation in which a customer holds favorable attitudes, but does not display repeat purchases. The authors argue that this type of loyalty may be caused by social norms or situational constraints, such as inconvenience. Further, promotions or discounts at competing firms may incentivize customers to buy at a competitor. Additionally, the authors identify social norms as a moderator between the attitude-behavior relationship. For example, a peer or relative might convince a customer to eat at a specific restaurant or buy a specific product, regardless of the customer’s own attitudes.

Figure 2. Dick and Basu (1994) framework

centrality (degree to which an attitude is related to the individual’s value system) and clarity (how well-defined an individual’s attitude is). Second, affective antecedents relate more to emotions (states of arousal), moods, and satisfaction (matching expectations and perceived performance). Third, conative antecedents include switching costs (one-time costs when switching), sunk costs (costs already incurred) and expectations (current and expected fit between the customer needs and market offerings).

The affective antecedents and situational influences presented in the framework by Dick and Basu (1994) are closely related to commitment as presented in the framework by Morgan and Hunt (1994). As mentioned previously, recent relationship marketing often distinguishes between affective and calculative commitment. The affective antecedents as discussed by Dick and Basu (1994) closely relate to affective commitment and the situational constraints that Dick and Basu (1994) include could be compared to calculative commitment. Therefore, the two frameworks agree that both emotional and financial considerations play a role in continuation of the customer-firm relationship.

Importantly, the Dick and Basu (1994) paper distinguishes loyalty attitudes from actual loyalty behavior, recognizing that these are related but clearly different constructs. The paper identifies several antecedents of loyalty attitudes that can be classified into cognitive, affective or conative antecedents. Most of these attitude antecedents are considered in subsequent studies in the field of customer loyalty as well. Further, the framework includes moderating variables such as social norms and situational influences (e.g. discounts).

2.2.3 Keaveney (1995)

The theoretical framework as proposed by Keaveney (1995) focuses directly on customer switching behavior as the outcome variable. The exploratory framework has been established using qualitative data by asking customers for their reasons to switch firms. Afterwards, the author classified these qualitative outcomes into several mutually exclusive categories. This explains why the variables in this framework are of a more short-term and behavioral nature compared to the other frameworks that are discussed.

failures are considered. These relate to the personal interactions between customers and employees of the service firm. Examples include uncaring, impolite or unresponsive service employees.

The fifth category relates to the firm’s response to service failures. A company’s poor response to a service failure (e.g. not responding, reluctantly responding) is considered an important determinant of customer’s switching behavior. The sixth category focuses on attraction by competitors, for example when the competitor is considered more reliable, more personable or its products are perceived to be of higher quality. Another important antecedent of customer switching are ethical problems. For example, customers reported that dishonest or intimidating behavior on the side of the firm is a reason to switch. The final category relates to involuntary switching. Customers identified that they switched firms for reasons out of their own and the firm’s control.

That is, the paper proposes a framework in which six out of the eight categories are under the firm’s control. The author does not specify how these different switching antecedents are related to customer perceptions. Still, the framework is considered helpful in identifying possible churn antecedents. Other researchers have underscored the importance of the variables in the framework after adopting it into their own empirical work (e.g. Athanassopoulos, 2000).

2.2.4 Bolton, Lemon and Verhoef (2004)

Bolton, Lemon and Verhoef (2004) introduce the customer asset management of services (CUSAMS) framework, which is closely related to the relationship marketing literature. The framework identifies how different marketing instruments influence customer behavior and financial firm outcomes via their influence on customer perceptions about the customer-firm relationship. The authors identify three different types of customer behavior, being relationship length (duration of the relationship), relationship depth (depth of the relationship e.g. via increased usage or upgrading) and relationship breadth (expansion of the relationship).

There are six different marketing instruments considered in the CUSAMS framework. These are the following: price of a product or service, distribution channel, advertising effects, direct marketing promotions (e.g. direct mailings), relationship marketing instruments (e.g. reward programs), and service quality. The authors argue that these marketing instruments affect customer perceptions and subsequently customer behavior. The customer perceptions included in the framework are price perceptions, satisfaction, and commitment. Similar perception-related variables are included in the previous frameworks.

Bolton and colleagues argue that different marketing instruments have different effects on behavior. More specifically, the authors distinguish between marketing instruments that are effective in the short term (e.g. price promotions) and instruments with a long-term effect (e.g. service quality). That is, the model proposes that the degree to which customer behavior is impacted depends on the type of marketing instrument. By identifying six separate marketing instrument, the CUSAMS framework isolates and specifies the effects that different marketing activities have to a greater extent than the previous frameworks did.

Figure 4. CUSAMS framework

2.2.5 Bansal, Taylor and St. James (2005)

Bansal, Taylor and St. James (2005) categorize churn drivers according to the comprehensive push-pull-mooring (PPM) framework, which originates from migration literature. The framework divides factors influencing consumer switching in the following categories: push, pull and mooring factors. Essentially, the PPM framework suggests that there are negative factors at the origin that push people away, while there are positive factors at the destination that pull people toward them. Additionally, there are mooring variables with which these push and pull variables interact. These mooring factors represent personal and social factors that either strengthen or hinder migration to the new destination (Moon, 1995).

Further, the pull effects are represented by the attractiveness of alternatives (positive characteristics of competitors). The mooring variables included in the PPM framework are the following: attitude towards switching, subjective norms (social pressures), switching costs (one-time costs when switching), prior switching behavior, and variety seeking tendency.

Figure 5. PPM framework

2.2.6 Literature summary

This section compares the frameworks in the previous subsections in order to offer a comprehensive literature overview. It presents the most important differences and similarities between the frameworks.

First, the frameworks differ considerably in how broadly defined the outcome variable of the respective research is. Whereas the Morgan and Hunt (1994) framework could be applied to any relationship, the other frameworks are more directly concerned with the specific nature of relationship between a customer and a firm. Dick and Basu (1994) take a relatively broad focus by taking loyalty as the outcome variable in their research. On the other hand, the frameworks by Keaveney (1995) and Bansal et al. (1995) are more directly concerned with actual consumer switching behavior. Bolton et al. (1994) extend these models by including financial impacts for the firm as well.

Furthermore, there are differences in whether the frameworks focus on positive behavioral outcomes such as loyalty or focus on the negative outcome of switching behavior. Whereas these are assumed to be each other’s opposite, this is relevant for the identification and framing of the drivers of the respective behavior.

What the frameworks have in common is that they recognize that the customer churn (or loyalty) decision is not made in isolation. That is, both the framework by Dick and Basu (1994) and the Bansal et al. (2005) framework include the social environment of the customer as an important driver of customer churn (either directly or indirectly). More specifically, both studies argue that customers are susceptible to social influence and that the decision to churn partly depends on the behavior of others.

Further, the influence of the competitive environment of the firm is recognized in the different studies. As argued by Keaveney (1995), Bolton et al. (2004) and Bansal et al. (2005), the attractiveness of competitive offerings influences the customer churn decision. Important is that this attractiveness is subjective in nature and depends on the perceptions of the customer about how different offers relate to each other. As stressed by Keaveney (1995) even satisfied customers might display churning behavior due to occasional competitive actions.

Irrespective of the differences between the frameworks, all authors agree that loyal customers possess some characteristics that makes them preferential over non-loyal customers. For example, loyal customers have a lower price sensitivity (Keaveney, 1995) and a greater usage of the firm’s services (Bolton et al., 2004). Additionally, Dick and Basu (1994) identify that loyal customers are less motivated to search for information about alternatives, are more resistant to persuasion by competitors and more likely to engage in positive word-of-mouth communications.

2.3 Drivers of Customer Churn

Based on the different prominent theoretical frameworks that are presented in the previous section, several potential drivers of customer churn are identified. That is, the existing literature and its pioneering frameworks are used to identify the most prominent drivers of customer churn and subsequently focus on these in this research. An important selection criterium is that a certain category of drivers is included in multiple existing frameworks. Taking this into account, this paper makes a distinction between the following different categories of churn drivers: relationship aspects, customer characteristics and marketing instruments.

The first category of drivers includes aspects related to the relationship between the customer and the firm. The frameworks that are presented in the previous section distinguish several variables related to the customer-firm relationship. Examples of aspects that are part of this category are relationship length, relationship breadth, and relationship depth. A main argument for including different relationship aspects into the current research framework is that prior research argues that the relationship between the customer and the firm should be considered worth maintaining to prevent the customer from churning (e.g. Morgan & Hunt, 1994). That is, existing literature shows that the characteristics of the customer-firm relationship influence the continuation of the specific relationship to a large extent.

Finally, existing literature stresses the importance of the marketing instruments that the firm and its competitors can use in attracting and retaining customers, confirming comprehensive nature of the customer churn decision. Examples include the price of a product or service and the channel through which it is offered. Also, the quality of the product or service offered is considered part of the marketing instruments that firms use. In line with general marketing mix theory, firms can use these factors to influence consumers and to convince them to buy or keep buying its products. Especially the CUSAMS framework by Bolton et al. (2004) is a useful starting point in identifying which marketing instruments are relevant for the customer churn decision.

The remainder of this section focuses on further specifying the different categories of churn drivers and how these are expected to influence customer churn based on existing theory. Afterwards, the conceptual model of this research is presented.

2.3.1 Relationship aspects Relationship breadth

Bolton et al. (2004) define that customer cross-buying is an important indicator of the breadth of the relationship between the customer and the firm. In a contractual setting, cross-buying refers to buying additional products and services from the existing service provider, in addition to the products that a customer currently has (Valentin Ngobo, 2004). In this sense, cross-buying can be considered as a measure of relationship development or relationship extension (Verhoef & Donkers, 2005).

Exiting literature indicates that a broader relationship between the customer and the firm has a positive effect on relationship duration (Reinartz & Kumar, 2003), which reduces the customer’s switching probability. Additionally, it is expected that cross-buying increases the burden of customer switching behavior due to increased switching costs (Blattberg, Getz & Thomas, 2001). That is, a broader relationship can create a dependence of the customer on the firm, which increases the time, monetary and psychological cost of switching to a different service provider (Colgate & Lang, 2001).

Therefore, a broader customer-firm relationship makes the customer more dependent on the firm. Not only does the customer have a high familiarity with the firm and its offerings, also the firm possesses superior information to tailor these offerings to the customer’s needs. Both are expected to increase the switching burden and decrease the customer churn probability. Therefore, the following is hypothesized:

H1: The smaller the breadth of the customer-firm relationship, the higher the likelihood that the customer will switch to another service provider.

Relationship depth

The depth of the relationship between the customer and the firm could be reflected by the level of service usage (Bolton et al., 2004). It is assumed that service usage and retention are interconnected processes, given that a customer has to renew the contract with a provider to have continued access to the associated service. A higher level of service usage reflects a deeper relationship between the customer and the firm (Bolton et al., 2004) and is expected to reflect the customer’s intentions to continue this relationship (Reinartz & Kumar, 2003).

Furthermore, the higher the service usage, the better able the customer is to form realistic performance expectations, which decreases disconfirmation (Anderson & Sullivan, 1993). Additionally, a higher usage level increases the number of exchanges between the customer and the firm, which is expected to increase a customer’s identification with the firm (Bolton et al., 2004). Furthermore, customers that have a deeper relationship with the firm develop a strong and generally positive attitude towards it (Keaveney & Parthasarathy, 2001), which lowers the negative impact of an incidental service failure on customer satisfaction (Keaveney & Parthasarathy, 2001).

Moreover, the more experience a customer acquires through interactions with the firm, the lower the uncertainty of the exchanges. Increased usage of the firm’s service provides customers with a greater knowledge of the company and of the requirements to use its services satisfactorily (Alba & Hutchinson, 1987). Therefore, it is expected that a customer’s perceived cost of switching suppliers increases when service usage is higher, which leads to a decrease in customer churn. This is reflected in the following hypotheses:

Relationship length

It is expected that customers that have a longer relationship with the firm are less likely to switch to a different service provider. This because a longer relation between the customer and the firm makes the customer more familiar with the firm and its offerings. Consequently, a higher level of familiarity is expected to increase switching costs. Also, existing literature identifies that the length of the customer-firm relationship provides an indication of the base level of customer satisfaction (Bhattacharya, 1998; Bolton, 1998). Furthermore, brand loyalty increases with time and thus will be higher for customers that have a longer customer-firm relationship (DuWors & Haines, 1990).

Also, research identifies that customers that have a long relationship with the firm weigh prior cumulative satisfaction more heavily and new information relatively less heavily (Bolton, 1998). Therefore, it is expected that long-term customers are less sensitive to failures on the side of the firm. Further, a longer relationship duration between the customer and the firm reflects past behavioral loyalty, which often translates into future loyalty (Verhoef, 2003). Additionally, research by Reinartz and Kumar (2003) identifies that longer customer relationships indicate greater customer satisfaction. This is reflected into the following hypotheses:

H3: The shorter the duration of the customer-firm relationship, the higher the likelihood that the customer will switch to another service provider.

2.3.2 Customer characteristics Age

Previous research has indicated that age is important to consider when it comes to customer churn and channel choice. More specifically, previous research indicates that younger customers are more likely to churn (e.g. Colgate & Lang, 2001; Pablo Maicas Lopez, Polo Redondo & Sese Olivan, 2006) and more likely to engage in multichannel shopping (Nesar & Sabir, 2016).

Furthermore, it has been identified that the cues on which people rely in the purchasing process differ with age. Older consumers are more likely to rely on their own experience, whereas younger people tend to rely more on advice from others (Homburg & Giering, 2001). Therefore, the following is hypothesized:

H4: The younger the customer, the higher the likelihood that a customer will switch to a different service provider.

Gender

Gender has attracted research interest in assessing behavioral differences between men and women in the purchasing process. For example, Homburg and Giering (2001) identify that for men repurchase decisions are influenced by the quality and functionality of the product or service itself, whereas women’s loyalty is determined by personal interaction processes. Thus, for women the satisfaction with the sales process and interaction with salespeople is of greater importance than for men.

Further, prior research by Melnyk, Van Osselaer and Bijmolt (2009) identifies that women are generally more loyal to individuals, whereas men tend to be more loyal to groups of people. In line with this reasoning, it is expected that men are more loyal to companies, given that these are represented by multiple people. Therefore, the following is hypothesized:

H5: The likelihood that the customer will switch to another service provider is higher for females than for males.

Past switching behavior

It is expected that a customer’s past switching behavior is impacting current switching behavior. The more experienced the customer is in switching, the lower the perceived effort associated with changing service providers (Jones et al., 2000). Given this, it is more likely that customers will engage in switching again.

Additionally, customers that engaged in switching are expected to have a shorter current relation with the firm, increasing the likelihood of switching (Pablo Maicas Lopez et al., 2006). The following hypothesis is tested in this research:

2.3.3 Marketing instruments Price

Previous research identifies the important role that pricing plays in explaining variance in customer (re)purchase decisions. As identified by Keaveney (1995) dissatisfaction with the price paid is a critical factor for consumers to churn from their current service provider. Also, price is used as a cue in evaluating the customer’s experience with a product or service and in shaping the attitude toward a service provider (Bolton & Lemon, 1999).

Expected is that customers are more likely to switch if they perceive that the prices of their current service provider are (too) high. For example, customers might believe that competing firms offer a product with a better price-value trade-off and might churn for that reason. Theory suggests that that price perception and price sensitivity differs significantly between customers that churn and customers that stay at the firm (Keaveney, 1995), in line with the notion that long-term customers have a lower price sensitivity. Therefore, the following is hypothesized:

H7: The higher the price paid, the higher the likelihood that the customer will switch to a different service provider.

Channel usage

When predicting churn for customers in the telecommunications market, it is important to assess how customers behave in a multichannel shopping environment. If the firm knows how customers behave during the churning process, it is possible to anticipate customer activities that are antecedents of churning behavior (Mau, Cvijikj & Wagner, 2015). In this manner, the firms are able to allocate marketing resources more effectively across channels. Research by Chang and Zhang (2016) stresses the importance of considering the different channels that customers have access to, since multichannel marketing tools are increasingly important in increasing customer retention and customer value and decreasing churn rates.

In a multichannel shopping environment, the firm and its customers get in touch via various advertising and communication channels. Previous research identifies that the selection of a channel in the search and purchase phase is based on the perceived benefits and need fulfillment offered by the specific channel (Verhoef, Neslin & Vroomen, 2007).

Facebook page. Therefore, customers in the telecommunications market have several options and different channels that can be used to gather information about the offerings in the market.

Online channels

The online channel increases the amount of information available to customers and increases the accessibility of this information. In this way, the online channel increases the awareness of switching opportunities and increases the market transparency, which influences customer churn (Holtrop et al., 2017). Additionally, the online channel offers low transactional cost and a high level of convenience (Verhoef et al., 2007). Therefore, the online channel supports customers in their process of accessing and gathering information in an easy and convenient way.

Most importantly, pricing information is widely available online. Hence, the online channel enables consumers to easily compare prices and access competitive information, which has accelerated the trend of buying on price. Previous research by Verhoef and Donkers (2005) concluded that channels that focus heavily on price create less customer loyalty, which applies to the online channel.

Based on the above-mentioned features that the online channel possesses; it is expected that it increases the awareness about available alternatives. Moreover, the online channel allows customers to compare offers based on price, which is likely to increase customer churn.

Offline channels

The offline channel enables customer to physically examine the product before purchasing and to make use of sales staff assistance in the shopping process (Arora et al., 2017). Additionally, prior research identifies that offline channels lead to customer excitement about the retailer experience, which is helpful in building a relationship between the customer and the firm (Verhoef et al., 2007).

Given the experience that the offline channel offers, it is expected that customers using the offline channel attach more value to building and maintaining the relationship with the firm. Given the different characteristics of the online and offline channel, and the customers that use a respective channel, channel choice is expected to moderate the relationship between the different churn drivers and customer churn.

The following moderation hypotheses are tested in this research:

Relationship Breadth. For customers that have a broad relation with the firm, the online channel offers

opportunities for further exploration of what the firm offers in an easy and convenient manner. Also, many firms have a broader online assortment and offer only a part of it at the offline channels (Verhoef & Donkers, 2005). Therefore, it is expected that for customers in the online channel, the relationship between the breadth of the customer-firm relation and customer churn is strengthened. This is reflected in the following hypothesis:

H8: Using the online channel strengthens the relationship between relationship breadth and customer churn.

Relationship Depth. Customers that have a deeper relationship with the firm are more familiar with the

firm and its offerings. Therefore, these customers are likely to have a better sense of what they are looking for and might appreciate the accessibility of the information that they are looking for online. Also, these customers benefit from the low transactional costs and speed and convenience of the online channel. Therefore, the following is hypothesized:

H9: Using the online channel strengthens the relationship between relationship depth and customer churn.

Relationship Length. The longer the relationship between the customer and the firm, the less likely the

customer is to switch to another service provider due to high prior satisfaction. Existing literature identifies that offline channels are suitable for building relationships with customers, whereas online channels are well-suited for maintaining the relationship once built. The following relationship between the length of the customer-firm relationship and customer churn is hypothesized:

Age. Prior research identifies that younger customers are more likely to change brands and firms due to a

preference for innovativeness and a lower risk aversion. On the other hand, impaired information-processing abilities and an increasing risk aversion often prevent older customers from switching. Given that the online channel possesses certain features that makes customer switching easier and alternative options more easily accessible, the following is hypothesized:

H11: Using the online channel strengthens the relationship between age and customer churn.

Gender. As mentioned previously, women attach a lot of value to the interaction with a salesperson in the

purchasing process. Given that the interaction between the customer and a salesperson is rather limited in the online channel, it is expected that the gender-related effects on churn probability are larger for online customers. Therefore, the following is hypothesized:

H12: Using the online channel strengthens the relationship between gender and customer churn.

Prior Churn. Given that the online channel is likely to decrease the burden to switch, it is expected that it

will strengthen the relationship between prior churn and future churn. More specifically, for customers that have prior churn experience it is hypothesized that they have a higher future churn likelihood given that their familiarity with switching is likely to decrease the switching burden. The features of the online channel, such as the immediate access to competitive offers and the option to immediately switch online, are expected to decrease this burden even further. This leads to the following hypothesis:

H13: Using the online channel strengthens the relationship between prior churn and customer churn.

Price. Given the accessibility of pricing information online, the online channel is expected to emphasize

pricing information more than the offline channel does. As argued by Verhoef and Donkers (2005), channels that focus heavily on price are creating less customer loyalty, since these channels attract more price sensitive customers. Therefore, the following relation between channel choice and price is hypothesized:

2.4 Conceptual Model

Integrating all the expected relationships that are discussed above, results in the conceptual model as shown in Figure 6. It provides an overview of the hypotheses that are tested in the current research. In the next chapter, additional information is provided about which variables are used to represent the concepts as presented here.

3. Data Collection

This chapter explains the data collection procedure that is followed in the current research. First, it specifies the specific setting in which the current study is applied. Then, it presents the outcomes of the preliminary qualitative research that is performed before formal data collection. Afterwards, it discusses the data collection and operationalization of the concepts that are discussed in the previous chapter.

3.1 Telecommunications Market in the Netherlands

The current research focuses on customer churn in the Dutch telecommunications industry, which is getting more saturated and shows stagnating mobile penetration rates (Verbeke et al., 2012). In this market, the number of telephone subscriptions exceeds the number of inhabitants (ITU, 2019; CBS, 2019). As the market is getting more mature, the topic of churn is of greater concern to the firms and managers operating in the market (Kim et al., 2004).

There are different mobile communication services’ companies active in the Dutch telecommunications market, that could be divided into mobile network operators and virtual operators (Calvo-Porral & Lévy-Mangin, 2015). In the Netherlands, there are four mobile network operators, being KPN, T-Mobile, Tele2 and Vodafone, that possess their own network. Furthermore, there are several virtual operators that rent a mobile network from one of the mobile network operators. More specifically, Simyo and Telfort pay rent to use KPN’s network, hollandsnieuwe uses the network of Vodafone and Ben uses the T-Mobile network.

Subscriptions from all above-mentioned providers are offered by the focus firm of this thesis and information is available about customer decisions related to subscription choice. The focus firm of this research is not the only retailer that offers customers the opportunity to start or renew a mobile subscription. There are several other on- and offline retailers active in the telecommunication market that directly compete with the focal firm. Moreover, most providers enable customers to start or renew their mobile subscription via their own channels. Therefore, the focus firm of this thesis is competing with several retailers and providers in the market for telecommunication subscriptions in the Netherlands. Given the competitiveness of the market, customer churn is of great concern for the focus firm of this research.

3.2 Outcomes Qualitative Research

Qualitative data is collected using telephone interviews in which open-ended questions are used to ask consumers about their experiences with the focus firm and telecommunications products in general. A total of 78 telephone interviews is conducted between the 28th of February and the 14th of March. The answers provided to the survey and the existing churn literature are both used as starting points in constructing a definitive conceptual framework.

The telephone interviews indicate that the current retention strategy of the focal firm is perceived as rather intrusive given the high number of messages sent. Furthermore, the timing of the messages is not optimal, since other firms start their retention efforts earlier. That is, the focus firm’s retention messages might not be relevant anymore, given that many customers already extended their contract or started a new one.

For telephone choice, the interviews indicate that price is the most important consideration in selecting where to buy a telephone. Furthermore, the reputation of the brand and the reviews of peers and other customers are important as well. Also, fast (next day) delivery is of high importance for many customers. Additionally, many customers attach value to physically examining the product before purchase. Interestingly, many women rely on their partner in deciding which phone to buy and are not involved in the final decision themselves.

When it comes to subscription choice, price is the most frequently mentioned determinant of the final choice. When buying a subscription, the presence of a physical store is less relevant for many customers. Most importantly, it should be easy and convenient to start or renew the subscription, for example via just one mouse click.

Additionally, the option to combine a mobile telephone subscription with TV and Internet at home is interesting for many customers. Furthermore, the network quality and coverage are important considerations as well, and customer reviews are examined to determine this. It turns out that many customers have a high deal sensitivity, which means they are often attracted by an attractive offer shown in a TV commercial or an online banner.

3.3 Alterations to Conceptual Model

Therefore, a promotion-related variable discount is added to the conceptual model. This construct reflects whether the customer bought the specific subscription based on a discount or promotion. That is, many customers indicate that their decisions related to telecommunication products are based on attractive acquisition offers made by telephone providers or retailers. Therefore, the influence of giving a discount as a part of the acquisition strategy is considered an important marketing instrument for firms.

Expected is that if a customer is acquired via an offer or promotion, this customer is more likely to fall for an attractive competing offer as well, reflecting a high deal sensitivity (Verhoef & Donkers, 2005). Therefore, the following is hypothesized:

H15: If a customer is acquired via a promotion, the likelihood that the customer will switch to another service provider is higher.

H16: Using the online channel strengthens the relationship between promotion and customer churn.

The resulting conceptual framework is presented in Figure 7 below.

Figure 7. Final conceptual framework

3.4 Data Preparation

3.4.1 Data selection

respective years are selected, since it is the most recent 2-year timeframe from which complete data is available. The available data includes information on 204,503 customers in 2016 and contains 79 variables. The 2018 dataset contains data on 325,095 customers reflected in 140 variables.

The current research does not have the necessary resources to collect additional data related to customer perceptions, even though the qualitative research suggests that there are reasons to assume that these are influencing the customer churn decision.

Verhoef (2003) suggests a solution to this problem by arguing that relationship characteristics can be used to proxy customer perceptions. These perceptions (e.g. satisfaction, commitment, involvement, and trust) are considered important in explaining churn but often hard and costly to assess for firms. Given that the current research does not have the resources to measure these perceptions directly, the approach as suggested by Verhoef (2003) is followed and perceptions are proxied using the different relational characteristics (i.e. depth, breadth, length) that are described in section 2.3.

3.4.2 Data cleaning

First, the raw dataset is cleaned by removing personal and privacy-sensitive customer information. That is, variables related to contact details such as phone number, address or email address are removed from the dataset before analyzing the data. Moreover, all information that is sensitive and potentially interesting for competitors of the focal firm is removed. This leads to removal of all variables related to margins and costs as well as information related to the contracts between the focal firm and different providers.

Additionally, all orders that are not labeled as ‘completed and sent’ are removed from the dataset. This because it happens that a customer places multiple orders and cancels all but one or that a customer is rejected by a certain provider. Furthermore, all orders that have a start year that is in the past or too far in the future are removed from the dataset, since these are assumed to be in the dataset by accident. For example, it is not possible that someone takes a subscription in 2016 with a start date in 2018, since getting a subscription is only possible 4 months before the subscription starts. Similarly, a customer cannot sign a contract in 2016 when the subscription has already started before (e.g. 2015).

3.4.3 Variable description

This section provides an overview of the different variables that are used in this research. More specifically, it identifies how the theoretical constructs that are presented in the conceptual model in Figure 7 are operationalized. An overview of all variables is included at the end of the section in Table 1.

Customer churn. The dataset provides information about the current and future provider of each customer.

If these two providers are not the same, it can be concluded that the customer churns. Therefore, a variable

Churn is created to indicate whether the customer churns or not. Given that churn is a binomial choice, the

variable can either take a value of 1 to indicate that the customer churns or a value of 0 if the customer does not churn.

Relationship breadth. The following variables in the dataset are used to identify the relationship breadth

(i.e. cross-buying behavior) of the customer: insurance and number of accessories.

More specifically, the variable insurance indicates whether the customer purchased an insurance, that is added as an add-on option when purchasing a subscription in combination with a telephone. If the field is empty, no insurance was purchased. Otherwise, the type of insurance is indicated. The binary variable

Insured is created to reflect whether a customer has an insurance (Insured = 1) or not (Insured = 0).

Further, the variable number of accessories reflects the number of accessories that the customer purchased in combination with the subscription. Examples of accessories that could be purchased are cell phone cases or protective screens. If a field is empty, this implies that there were no accessories purchased and that it equals 0. The values of the number of accessories variable need to be corrected by dividing them all by 1000. A new variable Accessories is created in which all NA’s are set to 0 and the values of the number of

accessories variable are corrected.

Relationship depth. The variable bundel1 is used to reflect the relationship depth between the customer

Additionally, the content of the subscription is considered important in reflecting the relationship depth between the customer and the firm. Therefore, the variables minutes per month, SMS per month and MB

per month are included to reflect the relationship depth between the customer and the firm as well. In case

a customer has a shared bundle (e.g. a shared total of 100 for minutes and texts), the maximum number is assigned to all categories within the shared bundle. That is, in this example both minutes and texts is set equal to 100. Different categories for the usage variables are created to reflect the content of the subscription. The selection of the categories is based on how providers categorize subscription content on their websites. This results in inclusion of the variables Minutes, SMS, and MB with the following four usage categories: 0-500, 501-1000, 1001-1500 and 1500+.

Relationship length. The variable duration of contract reflects the duration of the current subscription.

Additionally, the transactional dataset contains information about whether the customer renewed its previous subscription in the variable Prior_Churn. Also, it is known which future action the customer takes, which is reflected in the variable Churn.

Based on these, the minimum total relationship duration can be determined, which is reflected in the variable rel_length. In doing so, the minimum subscription duration at a specific provider is used in calculations. For most providers, this minimum duration equals 12 months, while four providers offer the option to get a subscription that can be cancelled on a monthly basis. The total relationship length equals the duration of the current subscription and the minimum duration of the prior or future contract in case of renewal. If a customer did not renew its previous subscription and does not renew the current one, the relationship length is set equal to the duration of the current contract.

Age. The dataset includes demographic information on the customers in the dataset. More specifically, the

variable birth date is used to determine the age of the customer. The age of the customer is calculated in years and stored in the variable age.

Gender. The variable gender reflects the gender of the customer. It is used to create a binary variable Male

either equal to 1 for males or 0 for females.

Prior churn. The variable retention indicates whether the customer renewed its previous subscription.

A variable Prior_Churn is created that either has a value of 1 (customer has churned before) or a value of 0 (customer has not churned before), reflecting the choice’s binomial nature. The dataset does not allow for identifying past churning before the previous subscription, so the variable Prior_Churn is used as a proxy for the customer’s previous switching behavior.

Price. The variable Maandprijs reflects the monthly fee that a customer pays for the mobile subscription,

and the variable Bundle_1_price reflects the additional monthly fee for any bundle that is added to the subscription. Further, the variable additional_payment represents the upfront payment that a customer pays at the start of the subscription. All these price-related variables are included in the remainder of this research.

Channel choice. The dataset provides information on the channel via which the purchase was made in the

variable sales location. A distinction is made between an online purchase and an offline purchase in one of the physical stores of the focal retailer. Based on the sales location variable, a dummy variable for online purchases is created, that is called online. It is a binary variable that takes a value of 1 when a purchase is done online and a value of 0 when a purchase is made in either one of the offline stores.

Promotion. The variable abo ppm1 reflects the discounted price that the customer pays per month. This

variable can be used to identify whether the customer was acquired via a promotion or discount. A variable

discount is created that either takes a value of 1 (customer is acquired via a promotion) or 0 (customer is

Table 1. Variable description

Variable Type Description Notation

Churn DV Whether a customer churns in period t Binary (1 = Yes; 0 = No)

Online IV Whether a purchase is made via an online

channel

Binary (1 = Yes; 0 = No)

Maandprijs IV The monthly fee (in euros) that a customer pays

for the subscription

Numeric

Bundle_1_price IV The monthly fee (in euros) that a customer pays

for any add-on bundle

Numeric

Additional_payment IV The upfront payment (in euros) at the start of the subscription

Numeric

Insured IV Whether a customer buys an insurance Binary (1 = Yes; 0 = No)

Accessories IV Average number of accessories that a customer

purchases

Numeric

Bundle IV Whether any add-on bundle is added to the

subscription

Binary (1 = Yes; 0 = No)

Minutes IV The number of minutes a customer can call per

month

Categorical (0-500, 501-1000, 1001-1500, 1500+)

SMS IV The number of text messages a customer can

send per month

Categorical (0-500, 501-1000, 1001-1500, 1500+)

MB IV The amount of data a customer can use per

month

Categorical (0-500, 501-1000, 1001-1500, 1500+)

Rel_length IV The minimum total relation length between the

customer and the provider in months

Numeric

Age IV The age of the customer in years Numeric

Male IV The gender of the customer Binary (1 = Male; 0 =

Female)

Prior_Churn IV Whether the customer has churned at the end of

its previous subscription

Binary (1 = Yes; 0 = No)

Discount IV Whether the customer is acquired via a

promotion

3.5 Data Exploration

This section further explores the data that is used in this paper. It focuses on how data imperfections, such as missing data and outliers, are treated. Afterwards, descriptive statistics of the included variables are provided.

3.5.1 Missing data

In order to examine the proposed hypotheses correctly, a missing data check was performed. For some of the variables in the dataset a missing value reflects a certain decision made by the customer. Therefore, for certain variables in the dataset a missing value is essentially equal to a value of zero.

For the variables Brand and additional_payment there are an equal number of 19,469 missing values identified. This because customers have the possibility to sign a contract with a provider without purchasing a hardware product such as a phone. In case the customer decides to take a sim only subscription, there is no information about the brand or purchase price of a phone, since the customer does not purchase any. The dummy variable Simonly reflects the type of subscription the customer purchased. This variable confirms that there is no missing data for the respective variables for customers that purchased a subscription in combination with a hardware product. Therefore, the NA values are essentially equal to zero.

The variable nbServiceProvider shows 75,681 missing datapoints. This because it is not compulsory to use the same phone number after signing a new subscription. Customers can decide to take a new phone number and not make use of the number portability options that providers offer. If a customer decides to take a new phone number, there is no information about the provider from which they took their phone number. Using the variable nummerbehoud (i.e. ‘number portability’) it is confirmed that many customers decide to take a new phone number rather than keeping their existing one, which results in missing datapoints for the

nbServiceProvider variable in the dataset. Therefore, the missing datapoints here are essentially zero values

as well.

After adjusting the above-mentioned variables by setting their NA’s equal to zero, a missing data check is performed on the remaining variables.