The effect of acquisition device type on the expected customer
retention in the online tours and activities industry
University of Amsterdam
AbstractThis thesis examines the relationship between the adoption of mobile or desktop as an acquisition device and the expected retention of a customer in the online tours and activities industry. The BG/NBD model was calibrated using a dataset of Tiqets, a Dutch company representative for the online tours and activities industry, to derive the expected customer retention of a cohort of Dutch customers from Tiqets. The results of the hierarchical regression model show that there is no significant relationship between device type and expected customer retention and therefore the model inhibits no predictive power. The monetary value of the customer’s first transaction seems to be a better predictor of expected customer retention as compared to device type, although this relationship is statistically insignificant. Another finding is that the monetary value of the first transaction is lower for customers adopting the mobile device, compared to adopters of the desktop device. These findings imply that managers in the online tours and activities industry should not allocate higher budgets to acquire customers on mobile devices, in pursuit of higher expected customer retention. When seeking higher monetary transactions values, managers should allocate greater resources to desktop device adopters. The findings deepen the theoretical understanding of the role that device type plays in predicting consumer behaviour.
Master’s Thesis
Master’s in Business Administration (track: Marketing) Student: P.L.A. Nijssen | 10387013
Date of submission: 24-06-2016 Supervisor: dr. E. Korkmaz
Statement of Originality
This document is written by Patrick Nijssen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
1. Introduction ... 3
2. Literature review ... 5
2.1 Customer-base analysis: different contractual settings ... 6
2.2 Probability models for customer-base analysis ... 7
2.3 Customer retention ... 9
2.4 Device types ... 10
2.5 Hypotheses ... 13
3. Data and method ... 14
3.1 Data collection ... 15
3.2 Variables ... 15
3.3 Independent, dependent and control variables ... 16
3.4 The BG/NBD model: parameter estimation ... 16
4. Results ... 19
4.1 Distribution of the data ... 19
4.2 Descriptive statistics ... 19
4.3 Hypotheses testing – hierarchical and simple linear regression ... 21
5. Discussion ... 23
5.1 Generalizability of results ... 25
5.2 Theoretical and managerial implications ... 26
5.3 Limitations and directions for future research ... 27
6. Conclusion ... 29 References ... 30 Appendix ... 33
1. Introduction
Mobile technology has infiltrated virtually every aspect of people’s lives (Varnali & Toker, 2010). Consumers and companies can now communicate anytime and anywhere by using mobile devices. In addition, companies have incredibly powerful technologies at their disposal to communicate with customers and to collect information about them (Rust et al., 2010). By the means of these technologies, companies now have access to rich customer data. The ability to precisely understand the behaviour of customers by using this data has enabled professionals to link customer metrics to specific marketing actions. This process by which data from customer behaviour are used to make business decisions is referred to as customer-base analysis. Ample literature is in place on statistical models for customer-base analysis (Fader & Hardie, 2009; Batislam, Denizel & Filiztekin, 2007). The models use data on customers’ past transactions with a company to make predictions about their future behaviour, both on an individual customer level, and on an aggregate level. These predictions help practitioners to determine which customers are likely to defect, which customers are likely to be profitable and how to allocate resources between customer acquisition and retention (Gupta, 2009). The models therefore attempt to assist managers in increasing the understanding of their customers.
An industry that has been lacking customer understanding and that is increasingly relying on mobile technology is the online tours and activities industry. This industry comprises companies that sell entrance tickets to major touristic attractions and museums worldwide. It is a growing industry where consumers already spend somewhere between $100 and $200 billion annually, while at the same time it is said to be one of the greatest untapped markets of digital travel (Kressmann, 2016). Consequently, leading global travel companies have shown their interest in the industry. In 2014, TripAdivsor acquired one of the largest online tours and activities companies, Viator, for approximately $200 million (O’Neill, 2014).
Furthermore, in 2015, Expedia started investing significantly in the industry (Schaal, 2015). On the other hand, as well as established companies, start up companies are joining the industry. Some of them are becoming serious players, having received several million dollars of funding (Gasdia, Kremer & Quinby, 2015).
Despite this, for all players in the industry one truth applies: the trouble of understanding the behaviour of their customers. Illustrative is the fact that consumers do not think of ‘tours and activities’ as a group or travel product. The online tours and activities industry is home to more than thirty distinct categories (among which: museums, amusement parks, musicals, comedy, sports) and consumers see each of them differently (Gasdia et al., 2015). Since there is a clear need for developing consumer insights, it is very interesting to conduct a study into consumer behaviour, considering that companies in the online tours and activities industry can benefit from increased knowledge on drivers of specific consumer behaviour. By the use of customer-base analysis, this thesis addresses an area of research that is of interest to the online tours and activities industry, and which can be relevant for other online businesses. The acquisition device of a customer, the device that is adopted by a customer for their first transaction at a company, will be researched in the light of the expected retention of this customer. The thesis’ pursuit is to discover what the effect is of the adoption of mobile or desktop as an acquisition device, on the expected retention of a customer. Therefore, I formulate the following research question:
What is the relationship between the type of acquisition device adopted by a customer and the expected customer retention in the online tours and activities industry?
This research attempts to contribute to the scientific literature, while being relevant from a managerial perspective. The outcome of the research may have implications on
marketing budget allocation decisions for online tours and activities companies and other online businesses. As the average cost of acquiring a customer online is regarded as higher than offline (Jain & Singh, 2002), profitability can only be obtained if a customer will make repeat purchases in the future. Understanding which customers are expected to purchase more frequently in the future is therefore relevant from a business perspective. Moreover Priceline Group, the umbrella company of Booking.com, reported it had spent $2.8 billion on online advertising in 2015 (May, 2016). This highlights the magnitude of the online advertising industry, and adds relevancy to the research in question.
This thesis intends to make its contribution by using database analysis to look into actual consumer behaviour at a representative company in the online tours and activities industry. Therefore, relevant knowhow will be acquired for practitioners in this industry.
The thesis is structured as follows. The next section discusses different probability models for customer-base analysis and it focuses on the antecedents of customer retention. Furthermore, the theoretical link between customer retention and device type is established. In the third section the data collection method is discussed and there is elaborated on the BG/NBD model. The fourth section consists of the results section, where there is reported on the analyses and the results. The last sections represents the discussion and conclusion in which implications, limitations and directions for future research are considered.
2. Literature review
Customers are the lifeblood of any company. Without customers, a company has no revenue, no profits and therefore no market value (Gupta & Zeithaml, 2006). Rust, Moorman, and Bhalla (2010) set the stage for a customer-cultivating organization, where they discuss reshaping the marketing department to a ‘customer department’. Their notion of the emergence of a CCO, the Chief Customer Officer, highlights the fact that companies are
becoming increasingly customer-oriented.
2.1 Customer-base analysis: different contractual settings
Companies with a strategic focus on creating long-term customer relationships build databases to identify their customers, track customer buying behaviour, and predict changes in customer purchase patterns at an individual level through customer-base analysis (Batislam et al., 2007). Customer-base analysis is concerned with distinguishing ‘active’ customers from ‘inactive’ ones, and predicting their future level of purchases considering their observed past purchase behavior (Schmittlein, Morrison & Colombo, 1987, in Batislam et al., 2007). The distinction between active and non-active customers is of vital importance for the development of models that adequately predict customer buying behaviour. In order to reach for acceptable forecasts, it is important to distinguish between contractual and noncontractual settings. In a contractual setting, the time at which a customer ends its relationship with a company is observed. This buyer-seller relationship is governed by a contract, which often predetermines the length of the relationship (Wübben & Wangenheim, 2008). In a noncontractual setting, it goes unobserved when the customer becomes inactive (Reinartz & Kumar, 2000). Customers do not notify the company when they stop being a customer. Therefore, any firm that has a noncontractual relationship with its customers can never know exactly how many customers it has at any point in time (Fader & Hardie, 2009). The challenge that is faced here is how to differentiate between those customers who have ended their relationship with the company versus those who are simply in the middle of a long gap between transactions (Fader, Hardie & Shang, 2010). The distinction between a contractual and noncontractual setting is fundamental. It is inappropriate to apply a model developed for a contractual setting in a noncontractual setting, and vice versa (Fader & Hardie, 2009). The main objective of the probability models developed for noncontractual settings are 1) to
predict which customers are most likely to be alive at a given point in time and 2) to make predictions about a customers’ transaction frequency, amount and timing. This thesis will make use of such a model.
2.2 Probability models for customer-base analysis
A variety of models are available for the purpose of predicting customer buying behaviour in a noncontractual setting. A highly regarded model in the literature addressing the challenges of a noncontractual setting is the Pareto/Negative Binomial Distribution (NBD) model proposed by Schmittlein et al. (1987). This model assumes that customers buy at a steady rate for a period of time and then become inactive. Since the dropout time of a customer is not directly observed, the only evidence that a customer may have become inactive is a suspiciously long period of time without any transaction after the last observed purchase, which is modeled in the Pareto/NBD model. The Pareto/NBD is a powerful model for customer-base analysis, but its empirical application can be challenging. Perhaps it is therefore that only a few studies have actually documented its empirical validation (Fader et al., 2005). Jerath, Fader and Hardie (2011) present a generalization of the Pareto/NBD model, which offers new insights about the customer ‘death’ process. Their framework, called the periodic death opportunity (PDO) model, assumes that death can only occur at discrete points in calendar time, independent of the transaction time.
As an alternative to the computational complex Pareto/NBD model, Fader et al. (2005) developed a variant of the Pareto/NBD model that they call the beta-geometric/NBD (BG/NBD). The modeling of the dropout process, the moment when customers become inactive, is the major difference between the Pareto/NBD and BG/NBD model. Contrasting the Pareto/NBD model, the BG/NBD model assumes that a dropout can only occur immediately after a purchase. The computational burden is therefore significantly reduced in
the BG/NBD model so that it becomes possible to implement the model even in a spreadsheet environment (Batislam et al., 2007). This is reflected in the five assumptions upon which the BG/NBD model is based (Fader et al., 2005):
1. While active, the number of transactions made by a customer follows a Poisson process with transaction rate λ. This is equivalent to assuming that the time between transactions is distributed exponential with transaction rate λ, i.e.,
f (tj| tj−1;λ) = λe−λ (tj−tj−1),t
j> tj−1≥ 0.
2. Heterogeneity in λ follows a gamma distributionwith pdf
f (λ | r,α) = αrλr−1e−λα
Γ(r) ,λ > 0.
3. After any transaction, a customer becomes inactive with probability p. Therefore the point at which the customer ‘drops out’ is distributed across transactions according to a (shifted) geometric distributionwith pmf
P (inactive immediately after jth transaction) = p(1− p)j−1, j =1,2,3,....
4. Heterogeneity in p follows a beta distribution with pdf
f ( p | a,b) = p
a−1(1− p)b−1
B(a,b) ,0 ≤ p ≤1,
where B(a,b) is the beta function, which can be expressed in terms of gamma functions: B(a,b) = Γ(a)Γ(b) / Γ(a + b).
5. The transaction rate λ and the dropout probability p vary independently across customers.
Fader et al. (2005) show that the Pareto/NBD model and the BG/NBD model yield similar results without a substantial loss in the model’s fit, leading them to conclude that the BG/NBD model should be viewed as an attractive alternative to the Pareto/NBD model. It is
therefore that there will be made use of the BG/NBD model in this thesis.
2.3 Customer retention
Research in the area of customer-base analysis has looked at building better models for the main components of customer lifetime value: customer acquisition and customer retention (Gupta, 2009). As modern companies become predominantly customer-oriented, they increasingly seek revenue from the creation and sustenance of long-term relationships with their customers. In such an environment, marketing serves the purpose of understanding the drivers of customer lifetime value (Gupta et al., 2006; Rust et al., 2010). Customer lifetime value is generally defined as ‘the present value of all future profits obtained from a customer over his or her life of relationship with a firm’ (Gupta et al., 2006). The emergence of online business has increased the importance of customer lifetime value models. Numerous online businesses do not have highly valued material assets. Such companies can therefore only be valued correctly when the value of their intangible assets is brought into the equation. As the value of the customer base is the most important intangible asset of these online businesses (Schulze, Skiera & Wiesel, 2011), understanding the lifetime value of the customers of these companies gives a more accurate picture of their potential (Jain & Singh, 2002).
As stated previously, online businesses can benefit from a thorough understanding of various factors that drive customer lifetime value. Considering that customer retention is one of the main components of customer lifetime value, it is critical that we understand what drives customer retention and why it is regarded as such an important element. For several reasons, customer retention is seen as a valuable component within the context of customer lifetime value. Not only do existing customers tend to purchase more than new customers (Bolton, 1998), it also has been shown that increases in retention rates can have a positive effect on the profits of a company. The main reason for this effect is that the cost of retaining
an existing customer is lower than the cost of acquiring a new customer (Kamakura, Ramaswami & Srivastava, 1991; Bolton, 1998; Weinstein, 2002). Weinstein (2002) suggests that it costs at least five times more to acquire a new customer than to retain an existing one.
Ranaweera and Prabhu (2003) define customer retention as ‘the future propensity of a customer to stay with a particular company’. The current literature on customer retention focuses primarily on unobservable constructs like customer attitudes and intentions (Gupta & Zeithaml, 2006) and it broadly suggests three predictors of customer retention: overall customer satisfaction, affective commitment and calculative commitment (Gustafsson, Johnson & Roos, 2005). Customer satisfaction is defined as the degree to which customer expectations of a product or service are met against their perceived performance (Murali, Pugazhendhi & Muralidharan, 2016) and it has traditionally been regarded as the fundamental determinant of customer retention (Ranaweera & Prabhu, 2003; Tamuliene & Gabryte, 2014). Whereas affective commitment is seen as the trust and reciprocity in a relationship, calculative commitment refers to the existence of switching costs or lack of viable alternatives (Gustafsson et al., 2005). Furthermore, Ranaweera and Prablu (2003) observe several interaction effects of satisfaction and trust, which suggest that in absence of trust, satisfaction will have less impact on retention. Also, they find that in an industry with switching costs, companies are likely to be able to retain even those customers who are less satisfied. This evidence is supported by Tamuliene and Gabryte (2014) who find that switching costs have, after satisfaction, the greatest impact on customer retention.
2.4 Device types
In contrast to the aforementioned constructs, there is limited research conducted on observable constructs that have an effect on customer retention. However, improvements in information technology have made it easier for companies to collect enormous amounts of
observable data, like customer transactions data. This allows companies to use data on revealed preferences rather than attitudes and intentions (Gupta et al., 2006). An example of observable data is the device that is adopted by the customer for their transactions. Especially for online businesses, research into device adoption mode of customers can lead to new insights. Although the academic literature on mobile device adoption is growing, the topic is still under development and research is in its early stages (Varnali & Toker, 2010; Wang, Matlhouse & Krishnamurthi, 2015), which makes it an interesting research area.
To complete an online transaction, two broadly defined device types can be adopted: mobile (smartphone or tablet) or desktops (personal computer or laptop). As computer usage has shifted from desktops to mobiles, interfaces have shifted from computer mice to touchscreens (Brasel & Gips, 2014). There is a true explosion worldwide in the use of handheld electronic communication devices, such as mobiles. According to eMarketer (2014), there will be more than two billion smartphone users in 2016. The huge number of adopters of these devices and the related services indicate a growing mass audience for mobile communication, an emerging digital lifestyle and a mass market for executing mobile transactions (Shankar, Venkatesh, Hofacker & Naik, 2010). As mouse-driven desktops pave the way for touchscreen devices, the role of touch in online consumer behaviour has become increasingly important. Brasel and Gips (2014) show across two studies that touchscreens create stronger psychological ownership over products in online shopping scenarios when compared to mice. This in turn increases the endowment effect, the effect that causes consumers to overvalue items that they perceive they own. The endowment effect raises the switching barriers perceived by customers, which might therefore increase customer retention for adopters of the mobile device.
Furthermore, the mobile device inhibits two main characteristics that make it unique in relation to other devices. The first unique characteristic is ubiquity, which refers to the ability
of users to receive information and perform transactions wherever they are and whenever they want (Smutkupt, Krairit & Esichalikul, 2010; Okazaki & Mendez, 2013). The mobile device is ultra-portable (Shankar & Balasubramanian, 2009) and is switched on almost all of the time. People do not leave their homes without their mobile device and usually do not leave them unattended. This relates to the second characteristic: personalization. The mobile device is a constant companion of its owner, is rarely used by anyone except its owner (Smutkupt et al., 2010) and it is treated as a personal accessory (Shankar et al., 2010). Since the mobile is such a personal device it is not unlikely that the personality of an individual can predict mobile device usage (Butt & Philips, 2008). Research conducted by Turner et al. (2008) and Butt and Philips (2008) show that people who are high on extraversion are more likely to report carrying their mobile device with them at all times. Moreover they are more likely to use their device in public and spend more time on their mobile device. Mooradian and Olver (1997) show that extraversion is a predictor of positive affect, which in turn positively influences customer satisfaction. And as the fundamental driver of customer retention, higher levels of customer satisfaction are known to increase customer retention.
Even though mobile devices’ screen size and functionalities are limited compared to desktops, given their temporal and spatial flexibility (Okazaki & Mendez, 2013), they suffice when customers want to achieve specific goals that require little search efforts. This mobile convenience reinforces customers’ experiential state of being in a relationship with the firm and therefore leads to repurchase intentions. Moreover, customers who interact with a company via mobile devices integrate its services into their daily routines, because mobile devices are an integral part of customers’ routines (Wang et al., 2015). This integration is reflected in customers’ recurring mobile activities, which lead to repeated purchases.
In contrast to Wang et al. (2015), Huan, Lu and Ba (2015) suggest that the limited screen size of the mobile device reduces the richness of the information presented, as
compared to the desktop device. Furthermore, they state that the mobile device is less effective for multitasking compared to desktops. This in turn increases the information search costs, as consumers have to search product information separately. These higher information search cost influence the product type that consumers purchase from the mobile channel; thus consumers may depend on desktops to purchase products that need heavy information search efforts. Moreover, higher information search costs lead to higher price sensitivity (Hoque & Lohse, 1999). This indicates that customers using the mobile device would purchase products of lower monetary value. Furthermore, research in the area of multichannel shopping show a clear distinction in the device selection decision, focusing on the information gathering prior to a purchase and the actual purchase itself (Kollmann, Kuckertz, & Kayser, 2012). As a result of its accessibility and convenience in achieving specific goals that require little search efforts, the mobile device is the preferred device for the stage of information gathering prior to a purchase. Additionally, a transaction on any device can only be completed if the payment of the purchase is successful. The desktop device is regarded as a more reliable device for secure and safe payment than the mobile device (Mallat, 2007), which indicates that customers prefer the desktop device for the handling of the payment of the actual purchase.
2.5 Hypotheses
While considering previous research in the field of mobile marketing and interpreting the relationship of the device type with customer retention, the conclusion is formed that customers adopting the mobile device as an acquisition device have a higher expected retention as compared to customers adopting the desktop device. Therefore, I hypothesize that:
H1: Customers adopting mobile as an acquisition device have a higher expected retention compared to customers adopting the desktop device.
Moreover, due to the unique usage situations and characteristics of the mobile device, customers adopting the mobile device incur higher information search costs as compared to customers adopting the desktop device. In turn, this leads to higher price sensitivity for adopters of the mobile device. Therefore, I hypothesize that:
H2: Customers adopting mobile as an acquisition device have a lower monetary value of their first transaction compared to customers adopting the desktop device.
3. Data and method
A quantitative method of data analysis is used in this study. By applying well-established standards, like the BG/NBD model, this research can be replicated and subsequently analyzed and compared to similar studies.
Data from Tiqets, a Dutch tours and activities company, are used in this study. Since the start of their business in December 2013, Tiqets has shown significant growth, which has been acknowledged by the closing of a funding round of $4 million at the end of February 2016 (O’Neill, 2016). Tiqets has become one of the leading online companies in the business of selling tickets for tours and activities in Europe and is therefore chosen as a representative company to use as a source of data. Tiqets sells their products, entrance tickets, through their website as well as in partnership with resellers like hotels. For the purpose of this thesis, data covering customers that have bought tickets through the Tiqets website are exclusively used. Data from reseller transactions are therefore disregarded. Furthermore, Tiqets has not yet developed a mobile application. All transactions, either from mobile or desktop, are therefore
executed on the same platform, the Tiqets website.
3.1 Data collection
The complete dataset was collected with the assistance of a senior backend developer at Tiqets. Tiqets has a custom build database in place. Data regarding the recency, frequency and total observed time of the customer buying behaviour (used as input variables for the BG/NBD model) were retrieved from this database. Data with regard to the monetary value of the first transaction of a customer was also retrieved from the Tiqets database. Data about the acquisition device adopted by the customer was retrieved from Google Analytics. All data from the Tiqets database and Google Analytics were merged accordingly, and were subsequently imported to SPSS for analyses.
3.2 Variables
The acquisition device, either mobile or desktop, is the device adopted by a customer for their first transaction at Tiqets. The first transaction of a customer at a company is generally considered to be one of the most important ones (Verhoef & Donkers, 2005). Therefore the device adopted for this first transaction, the acquisition device, is used in this study.
Expected customer retention is operationalized in this thesis as the expected future transactions of a customer in the dataset, or phrased differently: the amount of transactions a certain customer is expected to make in a predefined period in the future. The dataset contains customers who conducted one or more purchases at Tiqets. Therefore, every expected future transaction relates to the retention of this customer. This can be viewed as an arguable stance, which is elaborated upon in the limitations section.
paid for their first purchase at Tiqets. As previously stated, the first transaction of a customer at a company is generally considered to be one of the most important ones (Verhoef & Donkers, 2005). Therefore the monetary value of the first transaction is used in this study.
3.3 Independent, dependent and control variables
To test the first hypothesis, device type, is used as the independent variable. The expected future transactions are the dependent variable. The monetary value of the first transaction of a customer in the dataset is used as a controlling variable. The expected future transactions and the monetary value of the first transaction are both numerical data. Device type is categorical data and is therefore coded as a dummy variable to allow for more elaborate statistical analyses. ‘0’ reflects the adoption of a mobile device, and ‘1’ relates to the desktop device.
To test the second hypothesis in this study, device type is again used as the independent variable. Furthermore, the monetary value of the first transaction is used as the dependent variable. The expected future transactions are not taken into account when testing the second hypothesis.
3.4 The BG/NBD model: parameter estimation
The expected future transactions are calculated on an individual customer level, using the BG/NBD model as proposed by Fader et al. (2005). The BG/NBD model is chosen for its appropriateness in noncontractual settings and its relative easy use and applicability in business environments. To calculate the expected future transactions per customer, three pieces of information about each customer’s past buying behaviour are required: his ‘recency’ (when his last transaction occurred), ‘frequency’ (how many transactions he made in a specified time period), and the length of time over which his purchasing behaviour is
observed (Fader et al., 2005).
Table 1 summarizes the descriptive statistics of the dataset, following a similar structure for calculations as Korkmaz, Fok and Fuik (2013). The full dataset focuses on a single cohort of Dutch customers who conducted their first purchase at Tiqets during the period of July 1, 2015 until September 30, 2015. The dataset covers their initial and repeat purchase occasions for the period July 1, 2015 through March 31, 2016. Out of the 3372 customers who made their first purchase at Tiqets during the period of July 1, 2015 until September 30, 2015, 690 customers made one or more repeated purchases. The dataset covering only Dutch customers was chosen for its appropriateness in most accurately calculating the expected future transactions on an individual customer level. Dutch customers showed to be most heterogenetic in their purchasing behaviour, as compared to other subsets of the entire Tiqets customer database. In other words, Dutch customers showed the most variance in their repurchase behaviour, while they also repurchased more often. This is explained by the fact that Tiqets is a Dutch company. Because of its steep growth, Tiqets has been mentioned many times in renowned newspapers and websites in the Netherlands. Therefore, Tiqets is better known in the Dutch market. Furthermore, Tiqets’ business started with focusing on the Dutch market, which has led to more repurchase behaviour among Dutch customers as compared to customers from other countries.
Table 1. Descriptive statistics of Tiqets dataset Tiqets dataset
Number of customers 3372
Available time frame 274 days
Zero repeaters (fraction) 2682 (0.79)
Number of purchases (all) 4184
Average number of purchases per customer 1.24
The BG/NBD model was calibrated on the Tiqets dataset using Microsoft Excel and the add-in Solver function. The model holds four model parameters (r, alpha, a, b), where (r, alpha) determine the shape and scale of the purchase rate gamma distribution and (a, b) represent shape parameters of a beta distribution that determines the distribution of the dropout probabilities across individuals of the Tiqets dataset. A maximum likelihood approach was used, following the guidelines by Fader et al. (2005). Furthermore, a sales forecast of repeat purchasing by the cohort of 3372 customers was created. The final step consisted of predicting a particular customer’s future purchasing, given information about his past behaviour and the parameter estimates. The results of the parameter estimation are shown in table 2 and interpreted according to Korkmaz et al. (2013).
Table 2. Results of BG/NBD Maximum Likelihood Estimates Tiqets dataset r 0.1103 α 31.7581 r/α 0.0035 a 9.6746 b 2.5136 a/(a+b) 0.7938 log-likelihood -6123.1152
According to the BG/NBD model an average customer in the dataset makes 0.0035 transactions per day, while active. The shape parameter (r=0.1103) indicates a low to moderate level of heterogeneity in purchase rates across customers (Schmittlein et al., 1987). The probability that an average customer defects the next purchase is estimated at 0.7938 (a/(a+b)). Furthermore, the large value of a suggests that there is a low dispersion in defection rates.
4. Results
4.1 Distribution of the data
Customers without an assigned acquisition device were excluded from the dataset. Furthermore, skewness, kurtosis and normality tests were performed. The items expected future transactions and monetary value of the first transaction are not normally distributed. For expected future transactions skewness is 8.8 and the kurtosis is 179.4, which indicates that the distribution is negatively skewed and has many scores in the tails and is extremely pointy (leptokurtic). For the item monetary value of the first transaction skewness is 4.3 and kurtosis is 47.6, which indicates that the distribution is positively skewed and has many scores in the tails and is pointy (leptokurtic). Appendix 1 and 2 show an overview of the distribution of the variables and corresponding histograms.
The items expected future transactions and monetary value of the first transaction were transformed with a log10 transformation to normalize the distributions. Unfortunately, the log10 transformation did not decrease the skewness and kurtosis of both items significantly, and therefore did not make the data more normally distributed.
The choice was made to test the hypotheses while acknowledging its asymmetrical distribution.
4.2 Descriptive statistics
Table 3 contains the descriptive statistics for the variables under investigation in this thesis. The total number of observations is 3372, which reflect 3372 different customers. The average expected future transactions are .0777. This means that on average, a customer from the dataset is expected to make .0777 transactions at Tiqets in the period of 1 April 2016 until 31 December 2016. Furthermore, the average monetary value of the first transaction of a customer in the dataset is €45,03. In this dataset, 1760 customers were recorded as adopting
mobile as their acquisition device and 1612 customers were recorded as adopting desktop as their acquisition device. This explains the mean of .48 for the dummy variable device type, considering that 48% of all observations are located in the upper category. However, since the distribution is skewed, the median is a better indicator of central tendency (Field, 2013). The skewness of the distribution has as a consequence that the mean loses its ability to provide the best central location for the data. Therefore, the median values are also presented, as these are not as strongly influenced by skewed values as the mean. The median for expected future transactions is .0806 and for monetary value of the first transaction the median is 38.99. The median for expected future transaction is slightly larger than its mean, as the distribution of this variable is negatively skewed. The median for monetary value of the first transaction is smaller than its mean, as the distribution of this variable is positively skewed.
Table 3. Descriptive statistics
Variables N Min Max Median Mean SD
Device type 3372 0 1 .00 .48 .500
Expected future transactions 3372 .0008 .6999 .0806 .0777 .0278 Monetary value first transaction 3372 8.50 499.94 38.9900 45.0301 25.2327
Table 4 represents the correlation matrix. In this matrix, the Pearson correlation coefficient and its significance is depicted. There is no statistically significant correlation between device type and expected future transactions. The variable monetary value of the first transactions shows a stronger negative correlation with expected future transaction as compared to device type, although this relationship is not statistically significant. However, monetary value of the first transaction is significantly (p<.05) positively correlated with device type.
Table 4. Pearson correlations
Variables 1 2 3
Device type -
Expected future transactions -.008 -
Monetary value first transaction .038* -.017 - *Correlation is significant at the 0.05 level (2-tailed)
4.3 Hypotheses testing – Hierarchical and simple linear regression The formulas for the regression are as follows:
Formula 1: E(Y (t) | X = x,tx,T ) =α+β(D) +ε Formula 2: MV ( X (t)) = α + β(D) +ε
Where E(Y (t) | X = x,tx,T ) represents the expected future transactions, MV ( X (t)) represents the monetary value of the first transaction and D relates to device type.
Table 5 presents the hierarchical regression results. Hierarchical regression was performed to research the ability of device type to predict expected future transactions, after controlling for the monetary value of the first transaction. In the first model of hierarchical regression, one predictor was entered: monetary value of the first transaction. This model was statistically insignificant F (1, 3370) = .8931; p = .335 and explained 0.00% of variance in expected future transactions. After entry of the device type in model 2 the total variance explained by the model remained at 0.00% F (1, 3369) = .168; p = .682. The introduction of device type explained zero additional variance in expected future transactions, after controlling for the monetary value of the first transaction.
The hierarchical regression model estimates that when moving from mobile to desktop, the expected future transactions rise with .000 (p-value: .682). When controlling for the monetary value of the first transaction the coefficient remains constant but becomes even more insignificant. It seems therefore that there is no statistically significant difference
monetary value of the first transaction does seem to make a difference, however this is insignificant (p-value: .343). The model estimates that a one standard deviation increase in monetary value of the first transaction will lead to a .016 standard deviation decrease in the number of expected future transactions. The corresponding standardized coefficient for device type is: -.007. It can be argued therefore that the effect of the monetary value of first transaction is twice as large as the effect of acquisition device type.
Table 5. Hierarchical regression model for expected future transactions
Statistical significance: *p<.05
Table 6 presents the simple linear regression results. Simple linear regression analysis was used to examine the relationship between the device type and the monetary value of the first transaction. This model was statistically significant F (1, 3370) = 4.8923; p = .028. The model estimates that moving from mobile to the desktop will lead to a .038 standard deviation increase in the monetary value of the first transaction.
Table 6. Simple linear regression model for monetary value of the first transaction
Statistical significance: *p<.05
Variables R R2
R2
Change B SE β t
Model 1 .17 .000 .000
Monetary value first transaction .000 .000 -.017 -.965
Model 2 .18 .000 .000
Monetary value first transaction .000 .000 -.016 -.948
Device type .000 .001 -.007 -.410 Variables R R2 R2 Change B SE β t Model 1 .038 .001 .001 Device type 1.909 .869 .038* .028
The first hypothesis is rejected. The hierarchical regression model showed no statistically significant relationship between device type and expected future transactions and therefore the model inhibits no predictive power. The monetary value of the first transaction even seems to be a better predictor of the expected future transactions as compared to device type, although this relationship is also insignificant. The second model showed a statistically significant relationship between device type and the monetary value of the first transaction and therefore the second hypothesis is accepted.
5. Discussion
The global adoption of the mobile device and mobile services is growing rapidly. Consequently, mobile marketing has been an emerging topic of interests for academics as well as practitioners (Shankar & Balasubramanian, 2009). By implementing the BG/NBD model on a dataset from Tiqets, I have attempted to discover if the adoption of mobile or desktop as an acquisition device has an effect on the expected retention of a customer in the online tours and activities industry. The results show that there is no credible evidence that the acquisition device type of a customer can predict the expected retention of a customer. Although it was proposed that the adoption of the mobile device as the acquisition device, as compared to desktop, would result in higher levels of expected customer retention, no such relationship was found to be significant.
Gasdia et al. (2015) demonstrate that there is a clear need to develop new customer insights in the online tours and activities industry, an industry that is increasingly relying on mobile technology. As companies in this industry have trouble understanding the behaviour of their customers, they can benefit from increased knowledge on drivers of specific consumer behaviour. This study has shown that the adoption of the mobile device as the acquisition device does not drive higher expected customer retention. Therefore, it seems that
the unique usage situations and characteristics of the mobile device, like ubiquity (Smutkupt et al., 2010) and personalization (Shankar et al., 2010) do not affect the expected customer retention as was predicted in this study. However, the specific effects of these constructs were not measured in this study. Therefore, interferences about their exact relationship with expected customer retention cannot be made. Research conducted by Turner et al. (2008) and Butt and Philips (2008) that showed a significant positive relationship between extraversion and mobile device adoption, was interpreted as an argument for predicting higher expected customer retention for mobile device adopters. The findings of the current study do not align with this argument. Nonetheless, the findings of Turner et al. (2008) and Butt and Philips (2008) are not said to be contrasted by the findings of this study, since the relationship of extraversion with mobile device adoption was not under investigation in this study. I can therefore only reason upon the role that these different constructs play in the relationship between device adoption mode and expected customer retention.
Another finding of this study is that the monetary value of the first transaction seems be an even better predictor of expected customer retention as compared to acquisition device type. Although this relationship is not significant, the stronger ability of monetary value of the first transaction to predict expected customer retention could possibly be due to the notion that customers that have higher monetary values of their first transaction might be customers with higher purchasing power. This higher purchasing power in turn enables these customers to purchase more frequently in the future and they therefore inhibit higher levels of expected retention.
Additionally, there is a statistically significant relationship of an increase in monetary value of the first transaction when moving from mobile to desktop. This indicates that customers purchase products of lower monetary value on the mobile device, as compared to the desktop device. This confirms earlier research findings of Huan et al. (2015), who show
that the limited screen size of the mobile device increases the information search costs. This in turn raises the price sensitivity of adopters of the mobile device (Hoque & Lohse, 1999), leading them to purchase products of lower monetary value as compared to adopters of the desktop device. This is how I interpret the anticipated relationship between device type and monetary value of the first transaction. However, since the constructs of information search costs and price sensitivity were not investigated in this study, I cannot make interferences about their exact role in the relationship between device adoption mode and monetary value of the first transaction.
5.1 Generalizability of results
Tiqets can be considered as a representative company for the online tours and activities industry. Tiqets has rapidly grown to become one of the biggest companies in this industry. Other companies in this industry include Viator, GetYourGuide, Musement, and TicketBar. The combining efforts of all these companies, including Tiqets, are estimated to cover almost the entire online tours and activities industry (Gasdia et al., 2015). Therefore, Tiqets can be seen as a representative company for this industry. However, a couple remarks should be made. Although Tiqets has rapidly grown its market share in the online tours and activities industry, it can be argued that they are not completely representative for the industry, since they have only become a relevant player the past years, and have not been market leader over a longer timespan. Furthermore, with regard to customer retention, there is an important notion to be made regarding the Tiqets checkout flow on the website. The Tiqets website does not inhibit a basket, or a shopping card, where customers can purchase different products at the same time and pay them all at once. If a customer wants to buy two different products on the Tiqets website, they need to make two separate purchases, which is recorded as repurchasing behaviour. This is a unique feature of the Tiqets website, which other websites
of companies in the online tours and activities industry do not inhibit. It could therefore be argued that the Tiqets website design already inherently leads to more repurchase behaviour. In turn, this has an effect on the input variables of the BG/NBD model, and therefore have its effects on the calculated expected future transactions. In this research, there has not been controlled or corrected for this matter.
Although this study focuses on a cohort of Dutch customers, Tiqets has acquired customers from all continents and over eighty countries. Moreover, their business model strongly coincides with the business models of other companies in the online tours and activities industry (Gasdia et al, 2015). All in all I therefore argue that the results form this study can be seen as representative for the current online tours and activities industry, with regard to the Dutch customer segment.
5.2 Theoretical and managerial implications
The main acquisition channel in the online advertising environment is Google Adwords, the online advertising platform of Google. Google reports on $75 billion of revenue in 2015, most of which is earned through their advertising platforms (“Alphabet Announces”, 2016). As can be concluded, this is an enormous business. In consideration of this magnitude, it is very valuable to have superior knowledge on different drivers of customer acquisition, and their implications for other customer metrics. In a situation where companies know that a customer acquired via the mobile device is likely to purchase more frequently in the future, strategies can be designed to act on this. Likely steps that can be undertaken are the allocation of more budget to mobile customers as well as optimizing the website for mobile transactions.
Analyses of the researched data do not indicate a predictive effect of acquisition device on expected customer retention. Therefore, it shows that customers who adopt the mobile device as the device for their first transaction at a company do not necessarily appear to have
increased repurchase behaviour in the future, as compared to desktop adopters. This implies that managers in the online tours and activities industry should not allocate more budget towards customers that are acquired via the mobile device, when seeking higher expected customer retention. Furthermore, the effect of acquisition device type on monetary value of the first transaction indicates that managers can allocate more resources to adopters of the desktop device, in pursuit of higher transaction values of newly acquired customers.
As customer retention is considered to be one of the main components of customer lifetime value (Gupta, 2009) it is critical for academics, as well as practitioners, to understand what drives customer retention and which elements can possibly predict different levels of customer retention. The findings of this research show that the device adoption mode of a customer is not a predictor of the expected retention of this customer. Therefore, it is relevant to know for academics that because this research does not find a significant effect of observable constructs, like device type, on expected customer retention, unobservable constructs, like customer attitudes and intentions, might still be regarded as the main driver of customer retention (Gupta & Zeithaml, 2006). Moreover, the significant effect of acquisition device type on monetary value of the first transaction deepens the theoretical understanding of the role that device type plays in predicting consumer behaviour.
5.3 Limitations and directions for future research
In this thesis the definition ‘expected customer retention’ is used for defining expected repurchases of an existing customer. However, the probability model used in the thesis, the BG/NBD model, assumes that customers can only drop out immediately after their purchase. Therefore, it can be argued that ‘expected future transactions’ are not equivalent to ‘expected customer retention’. I acknowledge that there can be a difference in interpretation, but I choose to use the definition of expected customer retention as equivalent to expected future
transactions for the sake of consistency in this thesis.
Limitations of this research were primarily due to data availability issues. One of the limitations of this thesis is that it is based on a relatively sparse dataset. A common feature of start up companies is that customer retention is fairly low. Likewise, this has been the case for Tiqets. Although this argument played a role in the decision to use the BG/NBD model as a probability model for customer-base analysis, the dataset used in this thesis was less informative then would be ideal. This resulted in a sub-optimal calculation of the model parameters in the BG/NBD model and therefore in the calculation of the expected future transactions on an individual customer level. Moreover, as Tiqets is a company that started its operations only three years ago, this study makes use of a relatively small timespan of observations and does not include a validation period. Extending the current study, while using a more informative dataset, a longer timespan of observations and including a validation period is therefore a suggestion for future research.
Furthermore, this research was based on a dataset of Dutch customers from a Dutch company. The Dutch customer cohort was chosen in this study for its appropriateness in most accurately calculating the expected future transactions, as compared to customer cohorts from other countries. However, data of customers from different countries and different companies might lead to other results. Additionally, the acquisition channel of a customer has showed to affect customer retention (Verhoef & Donkers, 2005), and the adoption of a certain device is strongly related to the channel on which the customer is acquired. Therefore, there might be explanatory power in the relationship between device type and expected customer retention, when channels are taken into account. Extending the current study, by using data from customers from different countries, as well as including the acquisition channel of these customers are fruitful directions for future research.
found to be significant. I argue that this is possibly due to the limited screen size of the mobile device, which increases the information search costs (Huan et al., 2015), which in turn raises the price sensitivity of mobile device adopters (Hoque & Lohse, 1999) as compared to desktop adopters. Future research might focus on determining if information search costs and price sensitivity truly function as mediators of the relationship between device type and monetary value of the first transaction.
Moreover, this research does not correct for seasonality influences because of the computational burden it convenes. Gasdia et al. (2015) mention that the online tours and activities industry, as well as the entire tourism industry, is subject to seasonality influences. Future research may therefore attempt to predict customer buying behaviour, while correcting for seasonality influences. This attempt may be guided by previous research of Zitzlsperger, Robbert and Roth (2009).
6. Conclusion
This thesis examined the relationship between the adoption of mobile or desktop as an acquisition device and the expected retention of a customer in the online tours and activities industry. The BG/NBD model was calibrated on the dataset of Tiqets, to derive the expected customer retention of a cohort of Dutch customers from Tiqets. The results of the hierarchical regression model show that there is no significant relationship between device type and expected customer retention and therefore the model inhibits no predictive power. The results of the simple linear regression model show that customers adopting the mobile device have a lower monetary value of their first transaction, compared to adopters of the desktop device. The findings of this thesis have implications for budget allocations of managers in the online tours and activities industry. Furthermore, it deepens the theoretical understanding of the role that device type plays in predicting consumer behaviour.
References
Alphabet Announces Fourth Quarter and Fiscal Year 2015 Results (2016, February 1). Retrieved from: https://abc.xyz/investor/news/earnings/2015/Q4_google_earnings/index.html Batislam, E. P., Denizel, M., & Filiztekin, A. (2007). Empirical validation and comparison of models for customer base analysis. International Journal of Research in Marketing, 24(3), 201-209.
Brasel, S. A., & Gips, J. (2014). Tablets, touchscreens, and touchpads: how varying touch interfaces trigger psychological ownership and endowment. Journal of Consumer Psychology,
24(2), 226-233.
Butt, S., & Phillips, J. G. (2008). Personality and self reported mobile phone use. Computers
in Human Behavior, 24(2), 346-360.
Fader, P. S., & Hardie, B. G. (2009). Probability models for customer-base analysis. Journal
of interactive marketing, 23(1), 61-69.
Fader, P. S., Hardie, B. G., & Lee, K. L. (2005). “Counting your customers” the easy way: An alternative to the Pareto/NBD model. Marketing science, 24(2), 275-284.
Fader, P. S., Hardie, B. G., & Lee, K. L. (2005). Implementing the BG/NBD model for Customer Base Analysis in Excell. Retrieved from: http://www.brucehardie.com
Fader, P. S., Hardie, B. G., & Shang, J. (2010). Customer-base analysis in a discrete-time noncontractual setting. Marketing Science, 29(6), 1086-1108.
Field, A. (2013). Discovering statistics using IBM SPSS statistics. Sage.
Gasdia, M., Kremer, A., & Quinby, D., (2015). The In-Destination Experience - Shopping, Dining Activities & Tours. Phocusright Inc. Retrieved from: http://www.phocuswright.com Gupta, S., & Zeithaml, V. (2006). Customer metrics and their impact on financial
performance. Marketing Science, 25(6), 718-739.
Gupta, S., Hanssens, D., Hardie, B., Kahn, W., Kumar, V., Lin, N. & Sriram, S. (2006). Modeling customer lifetime value. Journal of service research, 9(2), 139-155.
Gupta, S. (2009). Customer-based valuation. Journal of Interactive Marketing, 23(2), 169-178.
Gustafsson, A., Johnson, M. D., & Roos, I. (2005). The effects of customer satisfaction, relationship commitment dimensions, and triggers on customer retention. Journal of
marketing, 69(4), 210-218.
Hoque, A. Y., & Lohse, G. L. (1999). An information search cost perspective for designing interfaces for electronic commerce. Journal of Marketing Research, 387-394.
Jain, D., & Singh, S. S. (2002). Customer lifetime value research in marketing: A review and future directions. Journal of interactive marketing, 16(2), 34-46.
Jerath, K., Fader, P. S., & Hardie, B. G. (2011). New perspectives on customer “death” using a generalization of the Pareto/NBD model. Marketing Science, 30(5), 866-880.
Korkmaz, E., Kuik, R., & Fok, D. (2013). " Counting Your Customers": When will they buy next? An empirical validation of probabilistic customer base analysis models based on purchase timing. ERIM Report Series Research in Management, (ERS-2013-001-LIS). Kollmann, T., Kuckertz, A., & Kayser, I. (2012). Cannibalization or synergy? Consumers' channel selection in online–offline multichannel systems. Journal of Retailing and Consumer
Services, 19(2), 186-194.
Kressmann, J. (2016). Future of Tours and Activities: Tech and Marketing. Skift. Retrevied from: http://skift.com
Mallat, N. (2007). Exploring consumer adoption of mobile payments–A qualitative study. The
Journal of Strategic Information Systems, 16(4), 413-432.
May, K. (2016, Febrary 18). Priceline Group spent $2.8 billion on online advertising in 2015. Retrieved from: http://www.tnooz.com/article/priceline-group-spent-2-8-billion-on-online-advertising-in-2015/
Mooradian, T. A., & Olver, J. M. (1997). “I can't get no satisfaction:” The impact of personality and emotion on postpurchase processes. Psychology and Marketing, 14(4), 379-393.
Murali, S., Pugazhendhi, S., & Muralidharan, C. (2016). Modelling and Investigating the relationship of after sales service quality with customer satisfaction, retention and loyalty–A case study of home appliances business. Journal of Retailing and Consumer Services, 30, 67-83.
Okazaki, S., & Mendez, F. (2013). Perceived ubiquity in mobile services. Journal of
Interactive Marketing, 27(2), 98-111.
O’Neill, S. (2015, July 24). TripAdvisor acquires Viator, the tours and activities agency, for $200M. Retrieved from: http://www.tnooz.com/article/viator-gets-acquired/
O’Neill, S. (2016, March 7). Tiqets, the attractions platform, nets $4 million in funding. Retrieved from: https://www.tnooz.com/article/tiqets-funding/
Ranaweera, C., & Prabhu, J. (2003). The influence of satisfaction, trust and switching barriers on customer retention in a continuous purchasing setting. International journal of service
industry management, 14(4), 374-395.
Reinartz, W. J., & Kumar, V. (2000). On the profitability of long-life customers in a
noncontractual setting: An empirical investigation and implications for marketing. Journal of
Rust, R. T., Moorman, C., & Bhalla, G. (2010). Rethinking marketing. Harvard business review, 88(1/2), 94-101.
Schaal, D. (2015, October 22). Expedia makes a $6.4 million bet on tours and activities in new TV advertising. Retrieved from: https://skift.com/2015/10/22/expedia-makes-a-6-4-million-bet-on-tours-and-activities-in-new-tv-advertising/
Schulze, C., Skiera, B., & Wiesel, T. (2012). Linking customer and financial metrics to shareholder value: The leverage effect in customer-based valuation. Journal of Marketing,
76(2), 17-32.
Shankar, V., Venkatesh, A., Hofacker, C., & Naik, P. (2010). Mobile marketing in the retailing environment: current insights and future research avenues. Journal of interactive
marketing, 24(2), 111-120.
Shankar, V., & Balasubramanian, S. (2009). Mobile marketing: a synthesis and prognosis.
Journal of Interactive Marketing, 23(2), 118-129.
Smutkupt, P., Krairit, D., & Esichaikul, V. (2010). Mobile marketing: Implications for marketing strategies. International Journal of Mobile Marketing, 5(2), 126-139.
Tamuliene, V., & Gabryte, I. (2014). Factors Influencing Customer Retention: Case Study of Lithuanian Mobile Operators. Procedia-Social and Behavioral Sciences, 156, 447-451. Turner, M., Love, S., & Howell, M. (2008). Understanding emotions experienced when using a mobile phone in public: The social usability of mobile (cellular) telephones. Telematics and
Informatics, 25(3), 201-215.
Varnali, K., & Toker, A. (2010). Mobile marketing research: The-state-of-the-art.
International Journal of Information Management, 30(2), 144-151.
Verhoef, P. C., & Donkers, B. (2005). The effect of acquisition channels on customer loyalty and cross-buying. Journal of Interactive Marketing, 19(2), 31-43.
Wang, R. J. H., Malthouse, E. C., & Krishnamurthi, L. (2015). On the go: how mobile shopping affects customer purchase behavior. Journal of Retailing, 91(2), 217-234. Wübben, M., & Wangenheim, F. V. (2008). Instant customer base analysis: Managerial heuristics often “get it right”. Journal of Marketing, 72(3), 82-93.
Zitzlsperger, D. F. S., Robbert, T., & Roth, S. (2009). Forecasting Customer Buying Behaviour: Controlling for Seasonality, 1-7.
2 Billion Consumer to get Smart(phones) by 2016. (2014, December 14). Retrieved from: http://www.emarketer.com/Article/2-Billion-Consumers-Worldwide-Smartphones-by-2016/1011694
Appendix 1. Distribution of variables Variables N Skewness statistic Skewness Std. Error Kurtosis Kurtosis Std. Error Device type 3372 .088 .042 -1.1993 .084
Expected future transactions 3372 8.814 .042 179.388 .084 Monetary value first
transaction
3372 4.301 .042 47.616 .084
2. Histograms Device type
Expected future transactions
