• No results found

The path to churn:

N/A
N/A
Protected

Academic year: 2021

Share "The path to churn:"

Copied!
90
0
0

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

Hele tekst

(1)

The path to churn:

Measuring the influence of customer touchpoints

on the initial decision to churn in the Dutch

telecom industry

By: Leon Wilbrink

(2)

The path to churn:

Measuring the influence of customer touchpoints

on the initial decision to churn in the Dutch

telecom industry

Leon Wilbrink June 2017 S1919377 Delflandplein 96, 1062HT Amsterdam l.w.b.wilbrink@student.rug.nl 0657786854

MSc. Thesis Marketing Management & Marketing Intelligence University of Groningen

Faculty of Economics & Business Department of Marketing PO Box 800, 9700 AV Groningen

Second supervisor:

Dr. Jelle T. Bouma

j.t.bouma@rug.nl

First supervisor:

(3)

Managerial Summary

Current customer journey research focus often on the path to purchase. Contrary, this research focuses on the path to churn. This research forms a better understanding of the customer journey of a mobile customer and its path to churn, by measuring the contribution of the different touchpoints on the initial decision to churn as opposed to renew or sleep. The industry of application will be the Dutch mobile industry, which due to confidentiality reasoning is mentioned as ‘the company’.

Existing literature shows that firms in the telecommunications industry nowadays face an approximate annual churn rate of approximately 30%. Therefore this industry can be considered as one of the top sectors on the list of those suffering from customer churn, and it underlines the importance of good churn management within this industry. In terms of academic purposes, this research can fill up several gaps in the literature by giving fundamental insights in influences which were not investigated in this specific conditions with respect to the relationship between touchpoints and the initial decision to churn as opposed to renew or sleep. Next to that this research will give new insights in the moderating effects of the aspects of timing, contact initiation, channel of contact, and type of communication used on the relation between the touchpoints and the initial decision to churn as opposed to renew or sleep, as applied in the Dutch telecom industry. For managerial purposes this research will give insights in how to improve the budget allocation per channel and/or touchpoint, as well as to reduce more churn more effectively and efficiently by focusing on the touchpoints which contribute the most towards the initial decision to churn as opposed to renew or sleep.

The research is conducted by applying the method of attribution modeling, in which the estimates of the different independent variables show its contribution towards the outcome variable. Attribution modeling is applied by making use of a binary logistic regression, in which the two outcome states represent the initial decision to churn versus renewing or doing nothing. Several models are created, by separately modeling channel variables on different levels - from a highly aggregated level until specified on a more detailed level via subcategorization. Finally a stepwise selection method for logistic regression selects the most relevant variables for the final model, on which the conclusions are based.

The results of the research confirm that a significant relationship exists between customer touchpoints and the initial decision to churn as opposed to renew or sleep. The results also confirm the moderating effect of timing, contact initiation, the channel of contact, and the type of communication used on the mentioned relationship. When the contact between customer and firms is customer initiated, the contribution on the initial decision to churn is higher than when the contact is initiated by the firm. Next to that, touchpoints experienced in the last period of the customer journey are proven to have a higher moderating effect than touchpoints experienced in the first or middle period of the customer journey. Also proven is that online and offline touchpoints differ in their contribution to the customers’ initial decision to churn as opposed to renew or sleep. At last, when used rich vocal communication during the contact, the contribution of touchpoints on the initial decision to churn is higher than when using standardized written communication during the contact.

(4)

When a firm wants to reduce churn in the Dutch telecom industry, based on this research it is recommended to focus on the most extreme contributors towards an initial churn decision as opposed to renew or sleep. These most extreme contributors are churn calls to the call center in the middle and last period of a customer journey, as well as a store visit concerning a change order in the last period of the customer journey. These are all offline, customer initiated contacts which make use of rich vocal communication. Where the current focus of the company is mostly on outbound contacts, the research proofs that the inbound contacts matter the most with respect to the churn decision. Next to this, it is recommended to develop a model which recognizes customer behavior patterns, so firms like the company can act on this behavior before the initial churn decision has been made.

Also recommended is to provide every employee which has direct customer contact with individual customer churn probabilities plus next best actions when churn probabilities are high, with the goal to prevent churn. Currently this occurs only on limited scale, but recommended is to expand this data as quick as possible to a bigger scale. The last recommendation with respect to churn management concerns the data input of the company, in which I recommend to upgrade this data input for better analysis purposes. When also being able to link user data within the app to individual customers, just as banner data, the customer journey can be measured on a more complete level.

(5)

Preface

Five months ago I decided to move from Groningen to Amsterdam, to write my MSc Marketing Management and Intelligence thesis at the Data & Analytics department of the company. After an internship in online marketing, my interest in data analysis began to grow and in my first conversations with my (new) colleagues of the company we discussed several thesis subjects, among which the subject of attribution modeling. After some days thinking about it I chose that subject to be my master thesis subject, because of the nice link with online marketing. In that working field I saw the possibilities of good data analysis in a world of big data. But it also got me thinking how it was possible that some large companies still struggle to value the contribution of the different marketing channels correctly within a customer journey, with all the data available nowadays. This thought increased my interest for my master thesis subject of attribution modeling and my motivation to generate some interesting insights.

In my days at the company I had the possibility to work three days a week on my thesis. The other two days I worked for the company on other projects, which was a really fun combination. The fun increased when my attribution modeling project became a part of a bigger project of the company. In this way I felt that the importance of a good analysis increased, which had the same effect on my motivation.

Looking back at the past five months I can state that in the process of my thesis there were ups and downs. The first three months passed by fairly well, but one crashed laptop and several computer issues later my stress level increased significantly. I guess it is all part of the process of writing a master thesis and luckily everything sorted out well, ultimately resulting in this research.

(6)

Table of Contents

Managerial Summary ... 3

Preface ... 5

Table of Contents ... 6

1. Introduction ... 1

1.1 Need for measurement ... 1

1.2 Path to churn ... 1

1.3 Importance of churn management ... 1

1.4 Heuristic approach ... 1

1.5 Customer journey analytics ... 2

1.6 Goal of the research ... 2

1.7 Research questions... 2 1.8 Scoping ... 3 1.9 Relevance ... 3 1.10 Overview ... 4 2 Theory ... 4 2.1 Churn ... 4 2.2 Customer experience ... 5

2.3 Customer journey analysis ... 5

2.4 Different stages ... 6 2.4.1 Pre-purchase stage ... 6 2.4.2 Purchase stage ... 6 2.4.3 Post-purchase ... 7 2.5 Timing ... 7 2.5.1 Heuristic method ... 7 2.6 Contact initiation ... 8 2.6.1 CIC ... 8 2.6.2 FIC ... 8 2.7 Channel of contact ... 9

2.8 Type of communication used ... 10

2.9 Conceptual model ... 11

3 Design ... 13

3.1 Data population ... 13

(7)

3.2.1 Data preparation ... 14 3.2.2 Data extraction ... 14 3.2.3 Multicollinearity ... 15 3.2.4 Outliers ... 15 3.3 General effects ... 15 3.4 Variable selection ... 16 3.4.1 Literature influences... 16 3.4.2 Availability ... 16 3.4.3 Modeling process ... 17 3.4.4 Variable categorizing ... 17

3.5 Covariates and clustered effects ... 18

3.5.1 Timing ... 18

3.5.2 Initiation of contact ... 18

3.5.3 Channel of contact ... 18

3.5.4 Type of communication used ... 18

3.6 Method of analysis ... 19

3.6.1 Model assumptions ... 19

3.6.2 Stepwise logistic regression ... 20

3.7 Data period ... 20

3.8 Model fit criteria ... 20

3.8.1 Face validity ... 20

3.8.2 Statistical validity ... 20

3.8.3 Goodness of fit ... 21

3.9 Descriptive statistics ... 22

4 Results ... 25

4.1 Modeling development process ... 25

4.2 Model fit ... 25

4.2.1 Variable effects ... 26

4.2.2 Model 1 – Only channels ... 26

4.2.3 Model 2 – Including timing ... 26

4.2.4 Model 3 – First channel subcategorization ... 27

4.2.5 Model 4 – Extensive channel categorization, personal webpage added ... 27

4.2.6 Model 5 - Category split on variables not concerning mobile subjects ... 27

(8)

4.2.8 Model 7 – Stepwise model ... 28

4.3 Final model decision ... 28

4.4 Further interpretation of the final model ... 28

4.5 Nonsignificant effects ... 30

4.6 Clustered effects ... 30

4.6.1 Standardization of coefficients ... 30

4.6.2 Timing ... 31

4.6.3 Initiation of the contact ... 31

4.6.4 Channel of contact ... 32

4.6.5 Type of communication ... 32

4.7 Covariates ... 32

4.8 Nonsignificant effects ... 32

5 Conclusions ... 34

5.1 Goal of the research ... 34

5.2 Research questions... 34

5.3 Overall effect ... 35

5.4 Timing ... 35

5.5 CIC versus FIC... 35

5.6 Online versus offline ... 36

5.7 Rich vocal communication versus standardized written ... 37

5.8 Covariate effects ... 38

5.9 Nonsignificant effects ... 38

5.10 Hypothesis acceptation/rejection ... 39

5.11 Final conclusion ... 39

6 Recommendations and Future research ... 41

6.1 Timing ... 41

6.2 Contact initiation ... 41

6.3 Channel of contact ... 42

6.4 Type of communication used ... 42

6.5 Covariate effects ... 42

6.6 Nonsignificant results ... 42

6.7 Overall recommendation ... 42

6.7.1 Churn management ... 43

(9)

6.8 Limitations and suggestions for future research ... 44

7 References ... 45

8 Appendix A – Research overview ... 50

Appendix B – Full variable list ... 51

Appendix C – Multicollinearity test ... 54

Appendix D - Model 1: Only channels ... 56

Confusion Matrix and Statistics – Model 1 ... 56

Lift curve – Model 1 ... 57

ROC chart – Model 1 ... 57

Appendix E - Model 2: Including timing ... 58

Confusion Matrix and Statistics ... 59

Lift curve – Model 2 ... 59

ROC Chart – Model 2 ... 59

Appendix F - Model 3: First channel categorization ... 60

Confusion Matrix and Statistics ... 61

Lift curve – Model 3 ... 62

ROC Chart – Model 3 ... 62

Appendix G - Model 4: Extensive channel categorization, personal webpage added ... 63

Confusion Matrix and Statistics ... 65

Lift curve – Model 4 ... 65

ROC Chart – Model 4 ... 66

Appendix H -Model 5: Category split on ‘other’ variable ... 67

Confusion Matrix and Statistics ... 69

Lift curve – Model 5 ... 70

ROC Chart – Model 5 ... 70

Appendix I - Model 6: Category split on ‘website_sales’ variable into two shops ... 71

Table 23: Output – Model 6 ... 72

Confusion Matrix and Statistics ... 72

Lift curve – Model 6 ... 73

ROC Chart – Model 6 ... 73

Appendix J - Model 7: Final model/Stepwise selection ... 74

Confusion Matrix and Statistics ... 74

Lift curve – Final model (Model 7)... 75

(10)

Appendix L - Average odds ratios final model ... 76

Appendix M - Average odds ratios: Clustered effects ... 77

Timing ... 77

Contact initiation ... 78

Channel of contact ... 79

(11)

Page 1/90

1. Introduction

The world around us gets more digital every day. In this world, companies are constantly on the move to collect and store data, because the data provides them with valuable insights about (potential) customers and their behavior like conversions or churn. And a better understanding allows business units to set better goals and to make intelligent, informed changes to the way the firm interacts with the customers (Bagley, 2016).

1.1 Need for measurement

Looking from a business perspective, marketing managers are being required to demonstrate the profitability of their marketing actions down to the level of their individual customer and on an ongoing basis. At the same time, customers expect firms to increasingly customize their products and services to meet their demands. Thus, increasing profit pressures, customer demand heterogeneity, and advances in technology all suggest that firms need to develop an orientation that is appropriate for survival and success in increasingly interactive market environments (Ramani, 2008). Interactions help firms refine their knowledge about customer tastes and preferences (Srinivasan, Anderson & Ponnavolu, 2002). The effective and efficient management of interactions and the interfaces at which these interactions occur are increasingly being recognized as sources of lasting competitive advantage (Rayport & Jaworski, 2005), more efficient marketing campaigns and improved customer satisfaction (Verhoef, Kooge & Walk, 2016).

1.2 Path to churn

In customer journey research several articles already have investigated the path to purchase (De Haan, Wiesel & Pauwels, 2016) (Kannan, Reinartz & Verhoef, 2016). Though, less often the focus of the research is on the path to churn. Churn is a very important factor, because it is known that the cost of acquiring a new customer can substantially exceed the cost of retaining an existing customer (Siber, 1997). From a business intelligence perspective, churn management process under the customer relationship management (CRM) framework consists of two major analytical modeling efforts: predicting those customers who are about to churn and assessing the most influential drivers on which an operator can (re-)act in terms of retention (Jahromi et al., 2010). Where the first is more focused on predicting who churns by identifying people, this research focuses on the latter and so is more focused on the contribution of the different touchpoints in the customer’s decision journey. In this research a touchpoint corresponds to a contact moment of a firm and a customer.

1.3 Importance of churn management

With an approximate annual churn rate of 30% the telecommunications industry can be considered as one of the top sectors on the list of those suffering from customer churn (Jahromi et al., 2010). As a result, companies are increasingly managing customer retention proactively by identifying valuable customers who are likely to churn and taking appropriate action to retain them (Ascarza, Iyengar & Schleicher, 2016). By identifying the importance of drivers of churn, firms will be better able to allocate their management and budgets towards certain channels and touchpoints. Customers are able to contact the firms via a big amount of touchpoints. There are the traditional offline touchpoints like television, radio, mail, call center (inbound and outbound), or in-store. Though, firms are investing more and more nowadays in online communication channels (Li & Kannan, 2014), like e-mail, apps, and websites. Attributing value to all those types of touchpoints can therefore be a complex job.

1.4 Heuristic approach

(12)

Page 2/90 channel which should lead to the desired goal? Often the heuristic attribution techniques, namely, first and last click attribution, are applied when measuring marketing effectiveness. Though, prior research (e.g. Abhishek, Fader & Hosanagar, 2016; Anderl, Becker, Von Wangenheim, Schumann, 2016; Li & Kannan, 2014; Xu, Duan & Whinston, 2014) indicate that the techniques of attribution conversion to the very first or last click can produce incorrect conclusions. These researches show that first or last click can overestimate specific channels in the conversion attribution, while other assisted channels values stay underrated in the path to conversion.

1.5 Customer journey analytics

Now that more and more data gets available and measureable, academics are able to research the path of customers detailing their interactions with different touchpoints in their purchase funnel. In the field of customer journey analytics, this leads to a heightened academic and practitioner interest in attribution modeling (Kannan, Reinartz & Verhoef, 2016). In attribution modeling online and offline channels in a customer’s purchase funnel are attributed the appropriate credit for the conversion related outcomes, taking into account the carry over effects within and spillover effects across channels (Kannan, Reinartz & Verhoef, 2016). But customer journey analytics have mainly developed quantitatively in the online environment by considering the attribution of different touchpoints to purchase and sales (e.g. Li & Kannan, 2014). Though, as Lemon & Verhoef (2016) state, this work should be extended by not only measuring online elements, but also taking offline elements into account. The same holds for the different phases of the customer journey the customer experiences. In this research via the use of customer journey analytics and attribution modeling, the contribution of the most important touchpoints of churn as opposed to renew or do nothing will be measured and its values will be interpreted. These touchpoints will be as well dividable into time periods, customer initiated and firm initiated, online and offline, and dividable in touchpoints which make use of standardized written communication and rich vocal communication.

1.6 Goal of the research

This research can form an answer to several problems firms confront regarding the attribution of churn drivers, by providing insights via customer journey analytics. In this research I map the customer journey of churners versus non-churners. Non-churners are customers who renew or who sleep, in which the latter is defined as doing nothing after their contract has expired. The goal of the mapping is to gain insights in the importance of the different events or touchpoints, subsequently to include them into the models. When these variables are proven to be significantly relevant by the model, firms can act on these insights in terms of allocating budget and marketing actions. After defining the relevant variables, the value of the variables can be generated. Via the use of customer journey analytics and attribution modeling, the contribution of the most important drivers of churn will be measured. The attribution model will provides scores for every element within the customer journey, after which its importance can be derived and managerial implications can be formed.

1.7 Research questions

To summarize, the main research question answered in this paper is:

What is the importance of the different touchpoints within the customer journey towards the initial decision to churn as opposed to renew or do nothing?

Sub questions answered in this research will be:

1. Do customer touchpoints have an effect on the initial decision to churn as opposed to renew or sleep?

(13)

Page 3/90 3. Do touchpoints experienced in the last period of the customer journey have a higher contribution towards the initial decision to churn as opposed to renew or sleep than touchpoints experienced in the first or middle period?

4. Does the effect of the different touchpoints on the initial decision to churn depend on the initiation of contact?

5. Do customer-initiated contacts have a higher contribution towards the initial decision to churn as opposed to renew or sleep than firm-initiated contacts?

6. Does the effect of the different touchpoints on the initial decision to churn depend on the channel of contact?

7. Do online and offline touchpoints differ significantly in their contribution to the customers’ initial decision to churn as opposed to renew or sleep?

8. Does the effect of the different touchpoints on the initial decision to churn depend on the type of communication used during the contact?

9. Do rich modes of communication have a negative contribution on the initial decision to churn as opposed to renew or sleep?

10. Do standardized written modes of communication have a positive contribution on the initial decision to churn as opposed to renew or sleep?

11. Do covariates contribute on the initial decision to churn as opposed to sleep or renew?

1.8 Scoping

This research will be conducted at the company in the Netherlands. The company is the largest telecom and IT service provider in the Netherlands. It provides telecom and IT services for both the commercial as the business market. In this research, I will focus on the mobile market. To be specific, in this research the sample group will consist of customers who already have a mobile subscription of the company, and who passed their end-of-contract moment during the data collection period. Within the business of the company, this end-of-contract moment is defined as the day the customer is no longer restricted to its contract with the company and all contractual agreements be expired. All these customers have had the same possibility of staying or churning. By taking only customers who passed their end of contract moment in the dataset, I am able to compare churners versus non-churners in this research. Next to that, there is already written a lot about the acquisition of customers. By focusing on the phases among the decision to churn or stay, I hope to generate new insights. But to do so, first insights are needed in the customer experience and the customer journey.

1.9 Relevance

This research can be beneficial for both academic and managerial purposes. In terms of academic purposes, this research can fill up several gaps in the literature, by giving fundamental insights in influences which were not investigated in these specific conditions. This research will give insights in the relationship between the initial decision to churn and touchpoints. Regarding these touchpoints, highlighted are aspects like online or offline touchpoints, firm-initiated or customer-initiated touchpoints, the timing of touchpoints, and the influence of different channels in which the events took place. All this is applied within the Dutch telecom industry. The combination of measuring the different touchpoints aspects and the appliance of the research in the Dutch telecom industry fills up gaps in the literature and makes this research relevant in academic perspective.

(14)

Page 4/90

1.10 Overview

This research is divided into six chapters. Chapter one forms the introduction. After this, chapter two will provide the overview of existing scientific theories that are relevant for this research. In chapter three the design of the research, including its approach and some first descriptive statistics, will be discussed. Chapter four includes the results and will sum up all relevant outcomes from the analyses. Chapter five will provide the conclusions that can be based on the results, after which chapter six will point out the most important recommendations and specific future research ideas for other researchers.

2 Theory

In the field of churn and it’s drivers many subjects have yet been researched. Some of these subjects will be discussed here, to provide a good overview of what is relevant for this research.

2.1 Churn

Managing churn is a key challenge in customer relationship management (e.g. Blattberg, Kim & Neslin, 2008, Reinartz & Kumar, 2000, Reinartz, Krafft & Hoyer, 2004). In industries such as telecom the importance of customer retention cannot be overstated (Ascarza, Iyengar, Schleicher, 2016). Firms in the telecommunications industry nowadays face an approximate annual churn rate of approximately 30% and can therefore be considered as one of the top sectors on the list of those suffering from customer churn (Jahromi, 2010). As a result, companies are increasingly managing customer retention proactively by identifying valuable customers who are likely to churn and taking appropriate action to retain them (Ascarza, Iyengar, Schleicher, 2016). Firms with a high level of churn, coupled with an increasing cost of new customer acquisition, can have severe long-term financial consequences. Risselada, Verhoef & Bijmolt (2010) state that churn management, being a part of Customer Relationship Management (CRM), is of utmost importance for firms, since they strive for establishing long-term relationships and maximizing the value of their customer base (Bolton, Lemon & Verhoef, 2004); (Rust, Siong, 2006). They state that losing a customer negatively affects a company in a number of ways. It leads to an immediate decrease in sales revenue and given that a company will have to attract more new customers when churn rates are higher, it will lead to an increase in acquisition costs (e.g. Athanassopoulos, 2000; Rust & Zahorik, 1993). Moreover, in the case of services that are sold on a contractual basis, losing a customer is not just a product less sold, but in fact the well-defined termination of a relationship (Risselada, 2010).

It has been shown in the literature that preventing churn, or customer retention, is profitable to a company for several reasons:

(1) Acquiring new clients costs five to six times more than retaining existing customers (Bhattacharya, 1998; Rasmusson, 1999; Colgate, Stewart, Kinsella, 1996; Athanassopoulos, 2000);

(2) Long-term customers generate higher profits, tend to be less sensitive to competitive marketing activities, become less costly to serve, and may provide new referrals through positive word-of-mouth, while dissatisfied customers might spread negative word-of mouth (Mizerski, 1982; Stum & Thiry, 1991; Reicheld, 1996; Zeithaml, Berry & Parasuraman, 1996; Paulin et al., 1998; Ganesh, Arnold & Reynolds, 2000)

(3) Losing customers leads to opportunity costs because of reduced sales (Rust & Zahorik, 1993). A small improvement in customer retention can therefore lead to a significant increase in profit (Lariviere & Van Den Poel, 2005).

(15)

Page 5/90 discussed are the aspects of timing, the initiation of contact, online versus offline, and two types of communication.

2.2 Customer experience

The importance of preventing churn highlights the urge of good customer management. Consequently, creating a strong customer experience has become a leading management objective for many firms (Lemon & Verhoef, 2016). The Marketing Science Institute view customer experience as one of its most important research challenges in the coming years (Research priorities 2016). Likely because of the increasing number and complexity of customer touchpoints, but also through the belief that creating strong, positive experiences within the customer journey will result in improvements to the bottom line, by improving performance in the customer journey at multiple touchpoints (i.e., higher conversion rates) and through improved customer loyalty and word of mouth (Court et al., 2009; Edelman & Singer, 2015; Homburg, Jozic & Kuehnl, 2015).

In their paper, Lemon & Verhoef (2016) conceptualize customer experience as a customer’s “journey” with a firm over time during the purchase cycle across multiple touchpoints. They also conceptualize the total customer experience as a dynamic process. The customer experience process flows from pre-purchase (including search) to purchase to post-purchase; it is iterative and dynamic. In each stage the customer experiences touchpoints, in which only some of them are in control of the firm. This process may function as a guide to empirically examining customer experiences over time during the customer journey, as well as to empirically modeling the effects of different touchpoints on the customer’s experience (Lemon & Verhoef, 2016). To this extend, the first hypothesis which will be tested is:

H1: Customer touchpoints have an effect on the initial decision to churn as opposed to renew or sleep.

2.3 Customer journey analysis

A major consideration when studying customer experience is an understanding of the customer journey (Lemon, 2016). Already a lot has been said and written about the subject of customer journey and customer journey analysis (e.g. Anderl et al., 2016; Court et al., 2009; De Haan, Wiesel & Pauwels, 2016; Lemon & Verhoef, 2016). In a customer journey analysis, firms focus on how customer interact with multiple touchpoints, moving from consideration, search, and purchase to post-purchase, consumption, and future engagement or repurchase (Lemon & Verhoef, 2016). Businesses interact with their users and potential clients across multiple channels and touchpoints, including television, telephones, the internet, mobile phones, and retail stores. Customer journey analytics allow businesses to take data from all these sources and where possible make sense of it. This allows them to better understand customer journeys and the effects of touchpoints on purchase behavior (Verhoef, Kooge & Walk, 2016). But although most of the theory mentioned is focused on purchase behavior, the aspects of the theory show many similarities which are applicable for churn behavior.

(16)

Page 6/90 Figure 1: The customer journey including the loyalty loop (Court et al., 2009)

The visual displays that, in an ideal situation, the customers will not consider being in an ending funnel. They should be captured in a loop, making the decision-making process a more circular journey. This extension or reshaping of the funnel suggests that during the post-purchase stage, a trigger may occur that either leads to customer loyalty (through repurchase and further engagement) or begins the process anew, with the customer reentering the pre-purchase phase and considering alternatives (Court et al., 2009). Customers in the “loyalty loop” will stay loyal and take the decision to repurchase again. When customers reach a decision at the moment of purchase, the marketer’s work has not come to an end: the post-purchase experience shapes their opinion for every subsequent decision in the category, so the journey will be an ongoing cycle (Court et al., 2009). In this research, the gross of the customers had made their personal decision to re-enter the loop, or to exit and churn. Only the sleepers form an exception, because they do not make a decision and therefore extend their post-purchase experience. Though, by measuring with a maximum time frame, the sleepers did get an opportunity to get to a decision moment.

2.4 Different stages

The decision to renew or churn depends on the phase the customer is in. The well-known theory of Court (2009) can be linked to the study of Lemon & Verhoef (2016). This theory is originally designed for describing the three stages of the purchase funnel, but these stages form many similarities with the stages the customers go through in this research.

2.4.1 Pre-purchase stage

The first stage – pre-purchase – encompasses all aspects of the customer’s interaction with the brand, category, and environment before a purchase transaction. Traditional marketing literature has characterized pre-purchase as behaviors such as need recognition, search, and consideration (Lemon, 2016). The first two stages of Court (2009) can thus be translated into the pre-purchase phase of Lemon & Verhoef (2016).

2.4.2 Purchase stage

(17)

Page 7/90 of this research will experience. The purchase stage covers all customer interactions with the brand and its environment during the purchase event itself. It is characterized by behaviors such as choice, ordering, and payment (Lemon & Verhoef, 2016). The purchase phase can be linked to the third stage of Court (2009), the moment of purchase.

2.4.3 Post-purchase

The last stage of Lemon & Verhoef (2016) is the purchase stage, which is similar to the post-purchase phase in Court et al. (2009). This stage encompasses customer interactions with the brand and its environment following the actual purchase, only that this process is extended to include the “loyalty loop” as a part of the overall customer decision journey (Court et al., 2009; Edelman & Singer, 2015). For consistency matters, I continue with the phases of Lemon & Verhoef (2016) during this research, though including the loyalty loop supported by Court et al. (2009) and Edelman & Singer (2015).

The customers in this research all go through a likewise process as mentioned by Court et al. (2009), although there is one important difference. This research focuses on customers in their decision journey to churn or stay loyal, so the research does not focus on the choice to purchase. The customers have all passed the three stages at least once because they are or were the owner of a mobile subscription of the company. Though this research does not focus on this first conversion, but on another form of conversion; the decision to churn or not. All customers enter a new decision process and so a new customer journey, because they face the decision to churn or not at their end-of-contract moment. The customers can re-walk the purchase funnel via the loyalty loop, and decide to loyal and repurchase again, or exit the funnel and churn.

2.5 Timing

Because the stages matter for the customers regarding their decision moment, it is of importance to take into account when the touchpoint is experienced. Assuming the stages of the customer’s journey are parallel to the stages within the model of Court et al. (2009), I link the experienced touchpoints to the aspect of timing, because touchpoints experienced a long time before a decision moment can have a different effect from very recently experienced touchpoints. To check for this effect, the theory of advertising ad stock will be discussed. Much research is done about the effects of advertisement on purchase decisions. These advertisements all form touchpoints in the customer journey. When looking at the timing aspect and theory about advertising ad stock, the research of Abhishek, Fader & Hosanagar (2016) and Hu, Du & Damangir (2014) are relevant. The underlying theory of ad stock, as explained by Joseph (assessed at April 2017), describes that exposure to advertising builds awareness in consumer markets, resulting in sales. Each new exposure to advertising increases awareness to a new level and this awareness will be higher if there have been recent exposures and lower if there have not been. This is the decay effect of ad stock and this decay eventually reduces awareness to its base level, unless or until this decay is reduced by new exposures (assessed at April 2017). Abhishek, Fader & Hosanagar (2016) show via a Nerlove and Arrow model that ad stock decays exponentially over time. According to their model, the effect of past advertising decays over time and is replenished by more advertising. Hu, Du & Damangir (2014) adopt a similar approach and assume that the effect of an ad decreases exponentially over time. Therefore, timing seems to have an effect on the decision making of consumers.

2.5.1 Heuristic method

(18)

Page 8/90 simplistic and do not reflect the consumer’s behavior. Other models divide credit equally among all touchpoints with which the consumer engages. Those models also do not accurately weight the impact of each channel (Lamont, 2014). For example when applying the “last click” approach, all prior channels get no credit, while it can be assumed that they had a contribution in the customer journey. Last-click methods track only through which activity a customer came (directly) to the point of conversion, neglecting all activities that came before (Abhishek, Fader & Hosanagar, 2016). In other words, the last-click metric significantly underestimate the contribution of prior clicks within the customer journey. Following the latter statement, hypothesized is that:

H2a: The effect of the different touchpoints on the initial decision to churn as opposed to renew or sleep depends on the timing of the touchpoints.

H2b: Touchpoints experienced in the last period of the customer journey have a higher contribution towards the initial decision to churn as opposed to renew or sleep than touchpoints experienced in the first or middle period.

2.6 Contact initiation

There are several types of touchpoints in which customers can have contact with firms. One separation in which touchpoints can differ in value is in terms of the initiative of the contact. In this research I distinct the effects of customer-initiated contacts (CICs) and firm-initiated contacts (FICs).

2.6.1 CIC

Customer-initiated contacts are initiated by (the behavior of) customers or prospective customers (Li, 2014), so the contact between the firm and the customer is established as a consequence of the customers actions. The distinction of Li & Kannan (2014) can be linked to the work of Lemon & Verhoef (2016). In their research, Lemon & Verhoef (2016) have made the distinction between brand-owned and customer-owned touchpoints. Also in this, the origin of the contact is an important differentiator. With customer-initiated contacts the consumers seek out information on their own initiative, in a channel chosen by themselves. The propensity to consider a customer-initiated channel might evolve over a long time horizon (Valentini, Montaguti & Neslin, 2011). From their awareness, experience, and expectations about these channels, customers may make these channel consideration decisions in advance and store them in memory for use when the appropriate occasion arises. That is, consumers evaluate each channel they are aware of with regard to the benefit it provides versus the incurred search costs and arrive at a smaller set of channels that they would consider for future information search when a purchase intention arises (Hauser & Wernerfelt, 1990; Mehta, Rajiv & Srinivasan, 2003). The customer-initiated contacts as discussed can be linked to Lemon & Verhoef (2016) their customer-owned touchpoints. These types of touchpoints are part of the overall customer experience but that the firm, its partners, or others do not influence or control. CICs are initiated by (the behavior of) the customers or prospective customers (e.g. Li & Kannan, 2014; Shankar & Malthouse, 2007); Wiesel, Pauwels & Arts, 2011), include search, price comparison sites, referrals and retargeting (De Haan, Wiesels & Pauwels, 2016).

Customer-owned touchpoints are the most critical and prevalent post-purchase, when individual consumption and usage take center stage (Lemon & Verhoef, 2016). When looking at the loyalty loop of Court (2009), this is the phase before the customers (should) re-enter the loyalty loop, and therefore a crucial phase towards the decision to churn or not.

2.6.2 FIC

(19)

Page 9/90 only at the time of encounter. Thus, the firm-initiated channels enter into customers’ consideration sets only when customers encounter them as a result of a firm’s targeting (Li & Kannan, 2014). (Lemon, 2016) Lemon & Verhoef (2016) categorizes FIC’s under brand-owned touchpoints, which they define as customer interactions during the experience that are designed and managed by the firm and under the firm’s control. They include all brand-owned media (e.g. advertising, websites, and loyalty programs) and any brand-controlled elements of the marketing mix (e.g. attributes of product, packaging, service, price, convenience, sales force).

As (De Haan, 2016) state, advertising effectiveness differs systematically along the dimensions of CIC versus FIC. Via an SVAR model they analyze, amongst other effects, the effect of the mentioned contact types. In this research, they also found that customer-initiated channels are more effective than firm-initiated channels, and that CICs exhibit higher sales elasticities than FIC, likely because they appear less intrusive and more relevant (De Haan, Wiesels & Pauwels 2016; Wiesels, Pauwels & Arts, 2011). Taking that into mind, along with the theory of Lemon & Verhoef (2016) stating that CIC touchpoints are most critical and prevalent post-purchase, it is hypothesized that:

H3a: The effect of the different touchpoints on the initial decision to churn as opposed to renew or sleep depends on the initiation of contact.

H3b: Customer-initiated contacts have a higher contribution towards the initial decision to churn as opposed to renew or sleep than firm-initiated contacts.

2.7 Channel of contact

Businesses have started using the Internet at the beginning of its development during the early nineties of the 20th century. Since then, the process of Internet adoption has had a significant growth for businesses, especially as a new communication and distribution channel. The trend of focusing on digital promotion techniques and slowly abandoning traditional media has started with the growth of popularity of the Internet at the end of the last century and has continued progressively (Bilos, Rulic & Kelic, 2014). But does this mean that offline touchpoints lose importance to online touchpoints? In literature there is not yet a unified view on whether online or offline touchpoints have a greater impact on the decision to churn or not. In many cases this impact is depending on factorial circumstances like industry, and a customers’ phase in his/her customer journey.

This is in line with the theory of Lemon & Verhoef (2016), who also state that channels differ in benefits and costs, often making one channel more useful for a specific stage in the purchase funnel than their channels. Though they state that these differences are, however, shrinking due to technological developments and diffusion of new channels. Customers differ in their preference and usage of channels across different purchase phases, and specific multichannel segments can be identified that differ in terms of consumer characteristics. Channel choices in the purchase funnel are affected by one another because of lock-in effects, channel inertia, and cross-channel synergies (Lemon & Verhoef, 2016). This makes it hard to generalize the effects of online versus offline touchpoints on the contribution in a decision making process.

Chang & Zhang (2016) uncovers two latent relationship states that customers migrate to and from — an active state and an inactive state characterized by different levels of purchase frequency, responsiveness to marketing, and profitability. Chang & Zhang (2016) also found that offline retail channels can be used to migrate customers from an inactive to an active state, effectively serving an “educational” or “revival” purpose, whereas online channels have the biggest impact of keeping currently active customers active.

(20)

Page 10/90 allocation per communication channel (online and offline) and conversion to communication objectives while determining the cost effectiveness of every communication channel used. In their experiment, Bilos, Rulic & Kelic (2014) show that offline communication techniques generate a higher absolute effect in achieving communication objectives in the observed population than online communication techniques. The experiment was set up at a shopping center, in which both communication techniques were used to generate a conversion. All participants entered a decision journey, in which the offline touchpoints showed to be more effective in reaching its goal (Bilos, Rulic & Kelic, 2014). The proof for a higher effectiveness of offline communication versus online communication techniques is supported by Kotler & Armstrong (2010), which show that the most effective advertising channels are all traditional media: catalogs, television, and direct mail.

Concluding on these different perspectives of looking at the theory and purposes of different channels, it can be said that there is probably no generally ‘best channel’ which represents the decision to churn the most. Though, the theory shows that there are differences, so hypothesized is that:

H4a: The effect of the different touchpoints on the initial decision to churn as opposed to renew or sleep depends on the channel of contact.

H4b: Online and offline channel touchpoints differ significantly in their contributing effect towards the customers’ initial decision to churn as opposed to renew or sleep.

2.8 Type of communication used

The next aspect which can be distinguished in this research is based on literature of (Venkatesan, 2004). They found academic support for channel communication to be antecedents of purchase frequency and contribution margin. In this research I frame two types of communication nodes, which are based on and supported by the Customer Lifetime Framework of Venkatesan & Kumar (2004):

 Rich modes (e.g. face-to-face, trading event meetings)  Standardized modes (e.g. direct mail, telephone, website)

Face-to-face communications and trading event meetings are the richest and most direct mode of communication possible among channel members (Mohr & Nevin, 1990). Relational customers tend to have high commitment and trust with their suppliers, which results in less uncertainty, more cooperation, and less complexity in their relationships than in those of transactional customers (Morgan & Hunt, 1994). Rich modes of communication are preferred to standardized modes when issues in the channel structure are complex and when there is a high degree of uncertainty in the relationship (Venkatesan & Kumar, 2004). Rich modes of communication are also effective in converting transactional customers to relational ones (Shephard, 2001).

Direct mail, telephone and web-based communication are the most standardized and cost-effective modes of individual-level communication available to an organization. Standardized modes are also the most cost-effective method for identifying customers who are interested in an organization’s current promotion (Shephard, 2001). For transactional customers, direct mail can be used in combination with telephone sales to generate interest in products while simultaneously improving the return on investment (Nash, 1993). For relational customers, direct mail serves to maintain commitment and trust by communicating relationship benefits (Morgan & Hunt, 1994) and to inform the best customers about new product offerings.

(21)

Page 11/90 H5a: Rich modes of communication have an negative contribution on the initial decision to churn as opposed to renew or sleep.

H5b: Standardized written modes of communication have an positive contribution on the initial decision to churn as opposed to renew or sleep.

2.9 Conceptual model

As mentioned this research will explain the contribution of different customer touchpoints on the initial decision to churn, as opposed to the decision to renew or sleep. When capturing the research in a conceptual model, it is important to know where to focus on when looking at the model. In this research I am focusing on the relationship between the independent variables and a known outcome (dependent) variable, and mainly the independent variables’ contribution shares. The research will investigate the effect of the different touchpoints on the outcome variable, while moderately affected by effects like timing, contact initiation, channel of contact, and type of communication used during the contact. At last also covariate effects for age, gender, the amount of years a customer is a client, the amount of mobile subscriptions within the customer’s household, and the fact whether a customer possesses a subscription including handset or only a Sim Only subscription taken into account. The interpretation of the touchpoints will be elaborated further in the next chapter.

Figure 2: Conceptual model

Initial decision to

churn versus

renew or sleep

Touchpoint X1 Touchpoint X2 Touchpoint X3 Touchpoint X4 Touchpoint Xn Customer journey of customeri - Timing - Initiation of contact - Channel

- Type of communication used during contact

Geclustered effects

Covariate effecten

- Age - Gender

- Years mobile customer

(22)

Page 12/90 All customers can experience their own customer journey, meaning that they experience different touchpoints, in a different order, at a different frequency, at different times. Below in figure X an example of a possible customer journey as experienced by a Dutch mobile customer is visualized. The horizontal axe represents a time line, going from the first of October until the last of April. The vertical axis represent all the possible channels where touchpoints can be experienced. Note that the subjects or subcategories of the touchpoints are not included, as well as the frequency.

(23)

Page 13/90

3 Design

The data used for this research is collected via all kind of sources. In this chapter I will discuss which data this is, how this data is collected and which decisions will be made regarding this data collection, in order to set up the model. Further information about the data and its possible value is presented in appendix A.

3.1 Data population

The research will be held at a Dutch telecom company in the Netherlands. With a wide range of products and services, this company provides her different services and brands to a large number of diverse client groups - at home and abroad. Among these services the company provides connection to landlines, internet, television, mobile subscriptions and other services all traditional telco’s provide.

The customers of the company are very diverse and unique. Therefore they all have their own unique way of behavior in for example terms of conversion, consumption and product or service processing. In this research I focus on active customers who already had a subscription the company of research during the period of October 2016 and April 2017. All these customer have their end-of-contract moment in April 2017. This group of customers is, among other types of customers within the company, being aware of their upcoming decision moment (end-of-contract), and therefore actively concerned in their customer journey.

All customers in the dataset can ultimately be divided into one of the three mentioned categories of churners, extenders or sleepers. When taking the mentioned prerequisites into account for data selection and looking at the population suitable for this analysis, this totals an amount of XXXXX customers. All these customers go through their own journey and will ultimately be divided into one of the three mentioned categories of churners, extenders or sleepers. These categories are defined as follows:

Churners: Customers who had the initial intention to churn on their mobile subscription between October 2016 and April 2017 (n=XXXXX). This includes customers who’s churn request is blocked on the company’s initiative or cancelled on their own request, and after this blocking or cancellation went sleeping or renewing.

Extenders: Customers who initially requested a renewal on their subscription, no matter if handset or sim only, between October 2016 and April 2017 (n = XXXXX). This also includes customers who’s churn request is blocked on the company’s initiative or cancelled on their own request, and after this blocking or cancellation went sleeping or renewing.

Sleepers: Customers who did not do any churn or renewal request between the 1st of November in 2016 and the 30th of April in 2017, but were included the dataset because of their end-of-contract moment in April 2017 (n = XXXXX).

(24)

Page 14/90 Total Churners Renewers Sleepers

XXXXX XXXXX XXXXX XXXXX

Table 1: Division of population

3.2 Data type

The research done in this paper is of quantitative, explorative nature. In this research, different type’s data will be analyzed. The data consists of numerical values, categorical values, text values, and is partly of structured and partly unstructured of nature. This data is collected via the the company data warehouse system, which consists of several databases with all kinds of unique information about various topics. Every data source collects and stores its information within their relevant databases, which is ultimately brought together into one data warehouse. For example the call center employees log their calls and activities in an internal system. This internal system links the logged information into a database with purely call information. The call database is then linked to the data warehouse, so it can be stored. The same holds for store employees, who log their sales or other conversions in a system, in which this information ultimately gets stored in the data warehouse. All web-data and all outbound marketing activities are provided with tags, so via the same storage process and data warehouse system it is possible to do analysis on this data on customer level.

3.2.1 Data preparation

To be able to extract the right data needed for the analysis, conditions are being stated. Using these conditions, a new event log table is created, which can function as the pool of data on which all the analysis can be done.

To be selected as a customer for the analysis don in this research, the following conditions are required:

- Mobile customer of the company - End of contract moment in April 2017

3.2.2 Data extraction

After the data preparation, the new event log table can be created. The following figure will provide a good picture of this process during the research, including the order of the steps indicated between the brackets:

Figure 4: Data extraction process Customer

information Event database

(25)

Page 15/90 When looking at the overall steps taken, there were three key steps taken to create a new event log for this research. The data used for this research was extracted from various databases, but in short the process looked like the figure above. At first, all the necessary customer information was selected and joined to the event database. This event database contained all events the customers experienced, except for the webpage events. These webpage events are stored in another database, the web event database. Therefore, step two was a separate join of all individual customers to the web events they experienced. The last big step taken was to union these two tables together, so one big event log was created, containing one record per event. Later on, this event log was transformed into a regression matrix, by grouping all events to the individual customer level and making all experienced events a continuous variable.

3.2.3 Multicollinearity

A problem which could lead to unreliable parameter estimates in the model is multicollinearity. With multicollinearity there are relations between the predictor variables, so they tend to be correlated (Leeflang et al., 2015). To check for this effect factor scores called the variance inflation factor (VIF) are calculated. A VIF greater than 5 is often taken to signal that collinearity is a problem (De Vaus, 2013). In appendix C the outcome table of the VIF scores is presented and although variables approach the score of 5, no variables show a score higher than this boundary value and no resolution actions have to be taken.

3.2.4 Outliers

To secure the reliability of the insights better, the data is cleaned for outliers. Mentioned earlier where the cutoff rules of calls to the call center > 300 per half year and website visits > 300 per half year. In this way for example an ID which daily showed non-stop records of phone calls to the call center was excluded of the dataset. Although this ID was defined by the system as a mobile customer, this turned out to be test calls from the call center for training purposes. In table 2 all the maximum amounts of events experienced by an individual customer are presented. These cutoff rules are based on the company standards.

Event Cutoff value

Total 320

Website visits 300

Call center 60

Mobile campaign 20

Store visit (Company owned and external)

20

Non-mobile campaign 20

Mechanic visit 5

Table 2: cutoff values per event

3.3 General effects

(26)

Page 16/90

Subject Explanation

Outcome variables Churn Renew Sleep Independent variables Web page

Telemarketing (outbound calls) Call center (inbound calls) Email

Direct mail Text message Mechanic visit Store visit

External store visit App banner

Covariates Age

Gender (base: male)

Amount of years mobile customer

Amount of mobile subscriptions within household Handset versus Sim Only (base: handset)

Table 3: Variable overview

The table presents all the variables which will be included in the model. These touchpoint can differ in terms of the places where the touchpoint is experienced, so online versus offline, the initiation of the contact (firm-initiated versus customer-initiated), and timing of the touchpoint. The mentioned effects will be measured as well in the model by categorizing the channel variables into these effects. The experienced touchpoints all contribute to one moment of decision. In this moment of decision, there are three options for the customer; the customer can churn from the company, renew their subscription contract, or do nothing and become a ‘sleeper’. In the last case, the current subscription will continue as it was, with a monthly terminable option. In appendix B a more extensive table is presented, in which all variables which are used in the modeling process are mentioned and explained.

3.4 Variable selection

The selection of variables chosen for this analysis is driven by three elements; literature influences, availability and frequency of occurrence of touchpoints. In the previous chapter several hypotheses are stated, mentioning effects of touchpoints on the decision to churn. These effects are taken into account in the variable selection and categorization processes.

3.4.1 Literature influences

Previous literature on relevant topics for this research formed a first influencer when selecting and categorizing the variables. In this respect the elements of timing, channels, contact initiation and communication types are being taken into account in this research. Furthermore, the literature influences the variable selection in such a way that I want to measure an as complete as possible customer journey, with as many types of touchpoints. The literature showed a gap on for example the combination between online and offline data, as well as data generated from a telecom industry.

3.4.2 Availability

(27)

Page 17/90 unable to collect online banner data, as well as detailed user data of the app. Because the company is not yet able to connect data from these channels to individual customers, it was not possible to use that type of data for this research. Regarding the app data does this only holds for user data which was customer initiated. There is some app data available which regards expressions initiated by the company.

3.4.3 Modeling process

Although this data limitation, the available data still makes it possible to map robust customer journeys on a high level of completeness. Because of the high level of variety of variables on the other aspects, I have chosen to start with a basic model which measures the customer journey on its highest level of abstractness, so touchpoints will be clustered together. From there, I want to look at the contribution-level of each clustered group of touchpoint in the output data of that corresponding model. When a group of touchpoints gets attributed a high significant level of value on the decision to churn as opposed to renew or sleep, I will zoom in on that variable and split it up into new, more divided variables. This process will continue until I found the right balance between splitting a variable and it’s attribution-level. After creating a very detailed model, the model fit will be assessed. At last a stepwise regression technique will be applied to the logistic regression, to cut back the detailed model to a model which is more parsimonious and therefore will show a better fit. This latter technique will be discussed later on.

3.4.4 Variable categorizing

To find structure in the unstructured values available in the company databases and to get a first indication of which variables are important to measure, I’ve made use of the frequency of occurrence of these variables and its subjects within current journeys. In this regard, I made use of a Python script, which scraped all the subjects of the event types from the relevant databases of the company, and clustered them into homogeneous groups. In combination with every subject the script showed the frequency of occurrence of the subject, so the relevancy can be indicated. By selecting the most frequently occurring subjects, I got a good first indication of which touchpoint types were relevant to measure the most complete customer journey. By combining this method with the variables indicated by the literature, I made the variable selection for the first attribution model in the way as can be extracted from appendix B.

During the data extraction all touchpoints get labeled with a subject by a text matching technique, which subtracts certain text elements from a long concatenation of subjects. After this subject-labeling, the second layer of clustering categorizes the variable by exact matching on for example channel or other conditions, specified per category type. The categorizing is visualized into the figure below (Figure X), to get a good picture of the process:

Figure 5: Structuration process

The flow above displays that the text mining categorizes the values into subjects, which specifies the touchpoint. This text mining is applied on a column of concatenated subjects of the events. On this point, the channel of experiencing the event is not yet taken into consideration. Therefore, after this first division of the variables into categories, another more exact technique of subtraction will be applied. This channel consideration is taken place in the next step, by exactly linking the event to a

(28)

Page 18/90 channel. Although these subjects were very unstructured and many fields contained unique values, the clustering method showed that the categorization as showed in appendix B was possible for web events. To give an overview of the elements and corresponding type of variables used in this research, table 11 can be consulted in appendix B.

The variables presented in appendix B are a result of the first round of clustering. As mentioned, when a variable shows a significant and great amount of impact, it can be that it will be split up and tested in parts in the next model.

3.5 Covariates and clustered effects

As presented in the conceptual model and mentioned in the previous chapters, there are certain covariates which can have an influence on the decision to churn and should therefore be taken into account, to make the research as complete as possible. These covariates are the following:

- Age - Gender

- Amount of years being a mobile customer - Number of mobile subscriptions in the household - Handset (0) or Sim Only customer (1)

3.5.1 Timing

Ultimately, to measure the timing aspect of the touchpoints within a journey, the events will be labeled depending on the timing of the occurrence. The time frame of the research is defined into three static periods. All touchpoints will be classified into one of these three labels, depending on the timeframe in which the touchpoint was experienced;

- First period: touchpoint was experienced between 01-11-2016 and 31-12-2016 - Mid period: touchpoint was experienced between 01-01-2017 and 28-02-2017 - Last period: touchpoint was experienced between 01-03-2017 and 30-04-2017

In this way the attribution of touchpoints experienced in a certain time period can be compared to another time period.

3.5.2 Initiation of contact

The type of contact can be divided into customer initiated (CIC) and firm initiated contacts (FIC). When looking at the channels of the company, the division of these types of touchpoints is made as follows;

- Customer initiated contacts: Website visits, call center calls, and store visits (Company owned stores and external stores)

- Firm initiated contacts: Emails, TM, DM, text messages, mechanic visits and app expressions.

3.5.3 Channel of contact

When looking at online versus offline channels, the following division is applied in the research; - Online channels: Website, email and the app

- Offline channels: TM, DM, text message, mechanic, call center, and stores (Company owned and external).

3.5.4 Type of communication used

To measure the effects of the type of communication, I divided the touchpoints into rich and standardized written communication types. The division looks as follows;

(29)

Page 19/90 - Standardized written communication: Webpage, EM, DM, text message and app.

3.6 Method of analysis

The method of analysis used for this research is an attribution model. Because the customer has three options in his decision moment, the dependent variable can show more than two values. This ultimately means that a multiclass model has to be set up. In multiclass attribution modeling with a nominal dependent variable and no varying regressors over the explanatory variable, the type of method which should be used is a multinomial logistic regression. With a binomial logistic regression function, the probability to meet one state as a function of the attributes of the alternative state available (churn as opposed to renew or sleep) will be predicted from a large sample of paths, to understand how much each touchpoint contributes to the ultimate decision.

Logistic regression is a method for fitting a regression curve, y = f(x), when y consists of proportions or probabilities, or binary coded (0,1-failure, success) data. When the response is a binary (dichotomous) variable, and x is numeric, logistic regression fits a logistic curve to the relationship between X and Y (King, 2016).

The logistic function is: Y

=

1+exp (ß0+ß1∗X)exp (ß0+ß1∗X) The logistic regression fits ß0 and ß1, and the logistic curve is nonlinear

(S-shaped) as can be seen in figure X. Therefore a logit transformation has to be done to make it linear:

logit(Y) = ß0 + ß1 ∗ Xt

Hence, logistic regression is linear regression on the logit transform of Y, where Y is the proportion (or probability) of success at each value of X. The outcomes will be logarithmic probabilities or log odds, which is the log of the odds ratio and should be interpreted in terms of relative risk or odds:

𝑝 = exp((x ′ß)) (1 + exp(x′ß)) p 1 − p= exp(x ′ß) 𝑙𝑛 p 1 − p= x ′ß

Here p/(1-p) measures the probability that y = 1 relative to the probability that y = 0 (Cameron & Trivedi, 2009), named the relative risk or the odds ratio. The anti-log transformation makes the logarithm undone and transforms the S-shaped slope into an odds ratio. This odds ratio can be interpreted as the ratio or probability that one belongs to one state, compared to the other state (King).

3.6.1 Model assumptions

Unlike discriminant function analysis, logistic regression does not assume that predictor variables are distributed as a multivariate normal distribution with equal covariance matrix. Instead, it assumes that the binomial distribution describes the distribution of the errors that equal the actual Y minus the predicted Y. The binomial distribution is also the assumed distribution for the conditional mean of the dichotomous outcome. This assumption implies that the same probability is maintained across the range of predictor values (Peng, Lee & Ingersoll, 2002).

Referenties

GERELATEERDE DOCUMENTEN

21 inclusion leads to more outreach to poorer borrowers by MFIs, looking at the difference between non-profit and for-profit MFIs, the rest of the fixed

The moderating effect of an individual’s personal career orientation on the relationship between objective career success and work engagement is mediated by

Figure 9: The moderating effect of legal origin on the relationship between ERM adoption and firm performance, of the single Regression Model II.. The statistical

In this three-way interaction model, the independent variable is resource scarcity, the dependent variables are green consumption and product choice and the moderators are

The effect of price on the relation between resource scarcity and green consumption reverses in a public shopping setting.. shopping setting (public

However, when the touchpoint were measured individually – instead of all customer initiated touchpoints combined and all firm initiated touchpoints combined – customer initiated

›  H4: Average product price positively influences the effect of the amount of opens on customer churn.. ›  H5: Average product price positively influences the effect of the amount

Disruptive technologies and the presence of the online channel resulted not only in increasingly connected consumers and enriched shopping experiences but also in the