Customer journeys in the travel industry
The effect of customer touchpoints with travel companies
P a g e 1 | 34
Customer journeys in the travel industry:
The effect of customer touchpoints with travel
companies on journey length and purchase probability
Master Thesis Marketing Management & Marketing Intelligence
Date: 14-1-2019
Author: Eline Jorinde Moorman
Address: Butjesstraat 4a 9712 EW Groningen
Email:
e.j.moorman@student.rug.nl
Phone number: +316-30588603
First supervisor: dr. P.S. van Eck
Second supervisor: dr. ir. M.J. Gijsenberg
Email:
p.s.van.eck@rug.nl
University of Groningen
Faculty of Economics & Business
Department of Marketing
P a g e 2 | 34
Summary
The travel industry has changed due to an ongoing trend of digitalization. It has provided opportunities for travel suppliers to target customers through intermediary agencies that make use of intelligent data gathering technology. During the last decade, different interfaces were provided through the internet. Online companies as Booking.com and TripAdvisor facilitate the customer with comparison-making, without the customers putting any effort in themselves. This research investigates how touchpoints with different travel companies’ websites effect the customer journey. It provides insights in the effects on journey length, the purchase probability of these journeys and how device usage moderates these effects. In order to contribute to the existing literature, data from GfK was analyzed. This research shows that in particular comparison websites (e.g., TripAdvisor) reduce the length of a customer journey, meaning that customers have fewer touchpoints when interact with this type of website. However, touchpoints with tour operator websites have the strongest positive effect on the purchase probability of a customer journey, whereas comparison websites reduce the purchase probability. Where a strong moderating effect of mobile device (versus fixed device) usage was expected, it appeared that fixed devices have a stronger effect in the customer journey.
P a g e 3 | 34
Preface
Dear reader,
This master thesis represents the final report of my Master Marketing Management & Marketing Intelligence. It illustrates my own interest in the online travel industry and the effect that the different travel companies have on a customer journey is therefore central. TV advertisements from for example Trivago made m wonder whether customers still prefer to purchase from an ‘original’ tour operator such as Corendon. Another question that came to mind was whether these websites also adhere to their promise of “saving time and money on your accommodation search” as Trivago states on their website. I therefore investigated how touchpoints with such travel companies, relative to the traditional tour operators, actually have an effect on customer journeys and whether these new comparison-making companies really disrupted the online travel industry.
Even though the process of writing a master thesis takes time and effort, in combination with perseverance, I am proud of the document that lies before you. I would like to thank my supervisor dr. Peter van Eck for his helpful feedback. My parents deserve a special acknowledgement, given their investment in me throughout my study in Groningen including their emotional support.
Enjoy reading,
P a g e 4 | 34
Table of Contents
1. Introduction ... 5
2. Theory ... 7
2.1. Current online travel industry ... 7
2.2. Expanding customer importance & customer experience ... 8
2.3. Conceptual model ... 8
2.4. Simplified journey – a shorter journey length ... 9
2.5. Purchase probability of a customer journey ... 9
2.6. Influence of device usage ... 10
3. Methodology ... 11
3.1. Research design ... 11
3.2. Data collection ... 11
3.3 Testing for effect on journey length – Poisson Model ... 11
3.4 Testing for effect on purchase probability – Logit Model ... 12
4. Results ... 13
4.1. Sample including only purchase journeys ... 13
4.2 Dataset including both purchase and no-purchase journeys ... 14
4.3. Data transformation ... 14
4.4. Estimation Negative Binomial Regression ... 14
4.4.1. Violations of Poisson Model ... 15
4.4.2. Estimation Negative Binomial Regression ... 15
4.5. Estimation Logit Model ... 17
4.5.1. Model selection ... 17
4.5.2. Model estimation – coefficient interpretation ... 18
4.5.3. Model estimation – odds ratio and marginal effects ... 18
5. Discussion ... 21
5.1. Research implications ... 21
5.2. Theoretical and managerial implications ... 22
5.3. Limitations and further research ... 22
6. References ... 24
7. Appendix ... 26
7.1. Codebook variables ... 26
7.2. Output R ... 27
P a g e 5 | 34
1. Introduction
The world of retailing has changed dramatically in the last two decades due to the advent of the online channel and ongoing digitalization (Verhoef, Kannan & Inman, 2015). For some markets, the online channel has even become a disruptive development. The travel industry was traditionally defined by customers going into a brick and mortar store in order for them to plan their travel trip, consisting of a flight and/or accommodation. However, the industry changed enormously with the rise of the internet. Customers are no longer restricted to offline stores, because the internet has enabled them to plan travels from their homes. The internet has brought consumers increased access to information to make their purchase decisions (Granados, Gupta & Kauffman, 2012).
During the last decade, different interfaces were provided through the internet. The increased usage of mobile phones, including mobile applications (i.e., apps), expanded the online possibilities for consumers. Nowadays, consumers can make use of mobile phones, laptops and tablets throughout their online journey. This evolution is seen as well in the online travel industry: consumers use multiple devices and search myriad websites to plan their trips (PhoCusWright, 2014). Wang, Park & Fesenmaier (2011) stated that there were indications of mobile technologies becoming the next wave of innovation that drives travel and tourism. One year after this article appeared, PhoCusWright (2012) showed a tremendous increase in mobile channel booking revenue, from $753 million to $1,368 million from 2011 to 2012 (Wang et al. 2016).
The internet has provided opportunities for the major travel suppliers (e.g., hotels, airlines) to target customers direct, thereby circumventing the traditional distribution channel through the travel agent. Furthermore, the online channel enabled new players to enter the field as well. Players as Booking.com, Expedia and TripAdvisor were disrupters for the travel industry (Verhoef, Kannan & Inman, 2015). These online companies enabled the customer to compare between different organizations within several clicks. Even though traditional retailers also introduced an online channel, these players unsettled the travel industry. The key feature of companies like TripAdvisor is that they facilitate the customer with comparison-making. TripAdvisor, for example, presents an overview of the available online accommodations between different suppliers. Booking.com does not make a comparison between different suppliers, but does display the different accommodations available online. They therefore act as intermediary companies. Customers are thus provided with the opportunity to compare between offerings, without putting in the effort themselves.
The process of digitalization led to a different customer journey. Customers now interact with firms through myriad touchpoints in multiple channels and media, resulting in more complex customer journeys (Lemon & Verhoef, 2016). With an explosion of mobile technologies and social media, multi-channel shopping has become a journey in which customers choose the route they take (Wolny & Charoensuksai, 2014). Each touchpoint leads to multiple new directions of which the customer can choose from. The complexity of the online customer journey lies in its length (multiple touchpoints across different channels) and diversity (multiple types of media). This complexity, however, is from a researcher’s point of view – the consumer enjoys a variety of options. The online world provides the customer with the opportunity to book online, but it also increased connectivity, communication, content consumption and content creation (Wang & Xiang, 2012: 308-309). Due to the fact that comparisons are already presented, the customer is enabled to leave out stages of the decision process. Therefore, these companies have the ability to create a simplified customer journey, meaning that the path to purchase is shorter.
P a g e 6 | 34
with these different companies is pivotal in order to gain insights in the relative effects within the current travel industry.
This study will contribute to the existing literature in a threefold way. First, this research will establish the effect of touchpoints with the different companies on the length of the journey, measured in total number of touchpoints. It will confirm whether customers have to put less effort in comparing the different companies and thus whether their customer journey is simplified once intermediary companies are visited. Second, this research will validate whether intermediary companies act as disrupters towards the online channels of the traditional tour operators. It will examine the effect of touchpoints with travel companies on the purchase probability of a customer journey. Third, this study investigates the role of device usage in a customer journey. Customers have an abundance of opportunities when it comes to appliances they can use in their journey. This study will investigate how device usage impacts the effect of touchpoints with travel companies on the customer journey. It will confirm whether media moderates the influence that such touchpoints have on journey length and purchase probability.
The main research question that will be answered in this research will therefore be the following:
How do customer touchpoints with travel companies effect the customer journey?
This main research question will be answered through a set of sub-questions: What is their effect on journey length?
What is their effect on purchase probability?
Does the type of device moderate the effects on the customer purchase journey?
The moderation will test for potential differences between media devices. Addressing this research question is important for managers in order to gain more insights in the customer journey of consumers that visit travel companies. The findings provide implications for the allocation of budget on advertising. Based on the effects of the different companies on purchase probability, the budget for advertising can be placed more strategically. Moreover, this attempt is essential for managers of the traditional tour operator to assess the effects of their competitors. The findings of this research can furthermore be related to value creation for the customer – instead of retrieving value from the customer, it is nowadays vital for a firm to generate value for the customer. Testing for a moderating effect of device usage can inform managers how to strategically advertise on different appliances. The results of this study have implications for firms why value creation and customer experience should be central in a company’s strategy.
P a g e 7 | 34
2. Theory
2.1. Current online travel industry
Developments in technology allowed customers to find new ways to research, buy and recommend products (Venkatesan et al. 2018: 55). Specifically in the online travel industry, customers benefit from an overwhelming flow of travel-related information from several sources that are all competing to gain their attention, custom and loyalty (Paraskevas & Kontoyiannis, 2005). However, this overwhelming flow is of little benefit to customers without an efficient way to screen all these alternatives effectively (Alba et al. 1997). Technology does not only provide customers with increased options in the online world, companies as well profit from a large pool of opportunities. To overcome the inefficiency for the customer, several online retailers have introduced intelligent agent technology that helps consumers sort the information available and make comparisons between the alternatives (Paraskevas & Kontoyiannis, 2005).
This research identifies three main travel companies. Multiple players in the online travel industry make use of comparison technology. Heerschap, Ortega, Priem & Offermans (2014) studied the analytics behind the online gathering of data. Small software programs, so-called robots or crawlers, trawl the internet collecting the desired information. This leads to the final page content, based on the search terms of the customer. Two types of robots are mentioned - the ‘dedicated’ and ‘general’ robot. In this article, both Booking.com and TripAdvisor are related to the first type of robot. Paraskevas & Kontoyiannis (2005: 487) define these comparison providers as “services that facilitate comparison of prices and product offerings among competing online travel retailers”. These comparison agencies exist because of such an intelligent data gathering technology – however certain differences in their concepts can be identified. The main distinction can be made based on the ability to make a purchase. These comparison-making companies create an efficient path to purchase for the customer, but not all corporations provide the possibility to purchase on the same page.
Paraskevas & Kontoyiannis (2005) highlight an important facet of comparison companies, namely that they differ from other online corporations in the way that they do not sell a product. They act as an intermediary company that presents the offering of several retailers. They arrange a comparison between the offers and direct the customer to the corresponding retailer after the customer clicked on an offer. Therefore, these comparisons are both price-based and offer-based. The comparison company provides the different prices of retailers per product offering, but also enables the customer to compare between the diverse set of offers. In the current online travel industry, companies as TripAdvisor and Zoover adhere to this definition. An example of TripAdvisor: when search*ed for hotels in Groningen, the company will show different hotels (e.g., Mercure Hotel and Apollo Hotel) that are offered by different retailers, such as Booking.com and Expedia.nl. Given the fact that this type of corporation only enables the customer to compare, the following of this research will refer to this first type of company as a
comparison company.
Corporations as Booking.com and Expedia.nl also make use of the comparison technology, but provide the customer with the opportunity to purchase on the same page as well. Thus, these companies arrange the offerings of several hotels and function as a purchase channel at the same time. This second type of corporation operates as an intermediary. It provides a platform where the offerings of hotels are presented and the accommodations can be purchased by the customer. Therefore this corporation will be referred to as an accommodation company throughout this study.
P a g e 8 | 34
operators as the benchmark in their relation to comparison companies. Following the digital evolution, these ‘traditional’ corporations set up an online channel where customers could buy their travel trip from their homes. This third type of corporation will be referred to as a tour operator company, implying that their offering consists of a specialized assortment. These offers are per tour operator different – thus they act on their own. How these three travel companies effect the customer journey will be investigated in this research.
2.2. Expanding customer importance & customer experience
An increased understanding of the customer journey in the online travel industry is important because of the fact that the customer plays a central part in both marketing research and practice. Over the years the importance of the customer has increased. Nowadays, not only the competences of the company are a determinant for the production of goods and services, so are the needs of the market. Producers are expected to benefit from an understanding of customer needs and preferences (Hoekstra et al. 1999). On the basis of segmentation, these needs and preferences are established. However, the market has changed in the last decades. There is growing customer diversity, the competition for the customer has become more intense, social media made its appearance and there is an increased level of customer-to-firm and customer-to-customer interactions (Kumar & Reinartz, 2012). The current leading concept, the customer concept, acknowledges that needs differ between people of the same segment and that accordingly heterogeneity of customer needs to be recognized. Thus, a proper understanding of the customer is key to a firm. It has become more important that the firm generates value for the customer than vice versa. Moreover, in the travel industry consumers demand greater value and service from the supplier of tourism products (Buhalis & Zoge, 2007). In order to gain an increased understanding of the customer and their online journeys individual data must therefore be analyzed.
The term customer experience is a buzzword in the world of marketing (Lemon & Verhoef, 2016). It has become the new differentiator in the 21st century. Delivering total customer experience goes beyond
mere customer satisfaction (Mascarenhas, Kesavan & Bernacchi, 2006). The experience covers the different levels of the consumption chain of the customer. Homburg, Jozić & Kuehnl (2017) recognize that the customer responds to the firm during the experience, namely by going through a journey of touchpoints along pre-purchase, purchase and post-purchase situations.
2.3. Conceptual model
This research investigates the effects of the three distinguished types of online travel companies. The conceptual model is presented in Figure 1. Each touchpoint is expected to influence the customer journey, which is measured on two levels, namely journey length and purchase probability. The moderating effect of device usage on the effect that these touchpoints have on the customer journey will be tested for as well. The visualization of the conceptual model can be seen in Figure 1 below.
P a g e 9 | 34
2.4. Simplified journey – a shorter journey length
An experience is built up through a collection of the so-called touchpoints in multiple phases of a customer’s decision process or purchase journey (Lemon & Verhoef, 2016). Customers move toward a purchase in a series of stages, including need recognition, information search, evaluation of alternatives, and ultimately, choice (De Haan, Wiesel & Pauwels, 2016). Customers engage with websites, apps and advertisements in pursuance of gaining information and these interactions are called touchpoints. This process is called the ‘customer decision journey’ or the ‘customer purchase journey’ (Lemon & Verhoef, 2016). Throughout this research, the definition provided by Lemon & Verhoef (2016: 71) will be used, namely that the customer journey can be referred to as the “process a customer goes through, across all stages and touchpoints, that makes up the customer experience”.
The customer journey has expanded due to changes in technology. It has allowed shoppers to go through a variety of touchpoints across online and offline media, channels, and devices on their paths to purchase (Kannan, Reinartz & Verhoef, 2016). It led to a customer journey that is non-linear. A customer may pass the stages of information search and evaluation of alternatives once they are not inclined to do research (Richardson, 2010). Industry experts as well state that the buying process is no longer linear (Venkatesan et al. 2018: 55). It changed the traditional view on the process of buying. Bonchek and France (2014) note that prospects do not come in at the top and out the bottom, but move through an ongoing set of touchpoints before, during and after a purchase. Baxendale et al. (2015) points out that the search process may iterate indefinitely while consumers revise brand/channel utilities, thus the passing of a phase is expected to have a large impact on the number of touchpoints in the customer journey.
The efficiency that comparison technology (used by accommodation and information/comparison corporations) provides, implies that the customer does not have to go through the traditional customer journey phases. Because this type of technology is expected to ease the path to purchase, fewer touchpoints are anticipated in customer journeys where such comparison-making companies are visited, causing the journey length to be shorter. Therefore, it is expected that visiting accommodation and comparison websites has a negative effect on the journey length. In view of the fact that the online content of tour operators does not provide this efficiency, more touchpoints are anticipated in journeys when these companies are visited, generating a longer journey length. This leads to the following set of hypotheses:
H1a: touchpoints with accommodation websites reduce the length of a customer journey H1b: touchpoints with comparison websites reduce the length of a customer journey H1c: touchpoints with tour operator websites increase the length of a customer journey
2.5. Purchase probability of a customer journey
Before internet, the travel industry consisted of three major components: suppliers (e.g., airlines, hotels), tour operators and the end-consumer (Buhalis & Zoge, 2007). Digitalization changed the sources of competitive advantage, since it affected the ability to differentiate. The role for each of three main players was specific and there was no overlap – suppliers used tour operators to reach the end-consumer (Buhalis & Zoge, 2007). However, the rise of the online world provided more opportunities for suppliers. New online companies became a new intermediary option. Travelers may buy more directly from suppliers, thus bypassing travel agencies, i.e. tour operators (Law, Leung & Wong, 2004). Therefore, the business competition has increased. The rise of the comparison corporations created an even larger platform for suppliers. Moreover, classic economic theory suggests that higher availability of information brings markets closer to perfect competition (Granados et al. 2012).
P a g e 10 | 34
(2014), the online market is dominated by accommodation companies as Booking.com. In 2013 they represented 38% of the global online market. Furthermore, Granados et al. (2012) state that price sensitive consumers will select a channel that offers easier comparison of product offerings and prices. For this group of consumers, comparison companies will thus be high on the list of choice. Concluding, tour operators fail in their advice-offering capability, based on the viewing of market shares.
Thus, the following is expected. Accommodation and comparison companies offer efficiency in comparison-making, causing the customer to move closer to a purchase. With the extensive choice set these corporations provide, the customer is more likely to make a purchase at the end of a journey.
Based on this literature, the following hypotheses can be formed:
H2a: touchpoints with accommodation websites increase the purchase probability
H2b: touchpoints with information/comparison websites increase the purchase probability H2c: touchpoints with tour operator websites decrease the purchase probability
2.6. Influence of device usage
The process of digitalization brought along an increase in media usage. The way customers can access the internet is rapidly changing. During the last decade, different types of media were added to the existing ‘traditional’ media. Nowadays, travelers use multiple devices, namely laptops, smartphones and tablets to plan their trips (PhoCusWright, 2014). Mobile technologies were expected to become the next wave of innovation that would drive travel and tourism (Wang et al. 2011). Currently, mobile is the fastest-growing method for travel booking (PhoCusWright, 2014). Customers mainly use their mobile phones – a distribution of media usage shows 52% taken by the mobile phone and 23% by personal computers (PhoCusWright, 2014). In this study, media is defined as the devices that a customer uses throughout the customer purchase journey. A distinction is made between mobile and fixed devices. A fixed device refers to a laptop or PC; a mobile device to a smartphone or tablet.
Smartphones have the capability to transform customers’ shopping experience (Persuad & Azhar, 2012). Customers use their mobile phones for both functional and entertainment reasons – they can browse unlimited and make use of social networking sites (Persuad & Azhar, 2012). It creates opportunities for marketers to influence the customer throughout their different actions and create an integrated marketing strategy. The smartphone is more than a connection tool – it is an extension of customer’s personality and individuality (Sultan & Rohm, 2005). It has become an indispensable instrument of an individual’s social and work life (Takao, Takahashi & Kitamura, 2009). Due to the centrality of the mobile phone in customers’ lives, it is expected that the usage of a mobile device has a stronger effect (compared to fixed devices) on the relation between travel companies and the number of touchpoints. Customers can take on every opportunity to go through the online world of the travel industry, through a smartphone that is an extension of their own individuality. Therefore, the following is expected:
H3: the usage of mobile devices (relative to fixed devices) strengthens the effect of the
touchpoints on the journey length.
Mobile marketing takes advantage of mobile devices’ pervasiveness and the very personal nature that is typical of communication when using those devices (Schierholtz, Kolbe & Brenner, 2007). The opportunities that marketers have to influence customers’ decision making, by sending them marketing messages and offers, can be perceived as intrusive (Persuad & Azhar, 2012). However, mobile devices are central in personal lives these days and they are the most used medium. The findings of PhoCusWright (2014) state that mobile is the fastest growing method for travel booking and that there has been a tremendous increase in mobile channel booking revenue. Thus, the following is expected:
H4: the usage of mobile devices (relative to fixed devices) strengthens the effect of the
P a g e 11 | 34
3. Methodology
3.1. Research design
This descriptive research is conducted in order to get a deeper understanding of the customer journey in the online travel industry. It involves quantitative analyses in order to present an answer to the research question. The final results can be used to describe the consumers in the online travel industry, as they form the population to be represented.
This paper uses two different models in order to assess the impact of the touchpoints with travel companies’ websites. The first model is a count model which describes how the three different types of touchpoints influence the journey length. This model therefore provides an answer to the first subpart of the research question. The second model involves a logit model, that answers the question how the three different types of touchpoints effect the purchase probability of an online journey. This model will provide an answer to the second subpart of the research question. The interaction effects between the type of touchpoint and the used device are taken into the models as well to assess the effect of device usage. Furthermore, both models include relevant variables, on the level of both a customer journey and socio-demographics, to assess for potential differences within the sample. The exact model specification including these variables is discussed more in depth in paragraph 3.3.
3.2. Data collection
The data was derived through a panel of GfK – a company that has “businesses around the globe and to make the best possible decisions every day, they need to really know what is going on, now and in the future” (GfK, 2018). In pursuance of this objective, GfK arranged a Crossmedia Link panel that measures consumers’ cross-media behavior. Rather than human observers, mechanical devices have made the observations. Therefore, the type of observation can be called a mechanical observation (Malhotra, 2010: 231). The respondents agreed to install a plug-in on their devices and data was collected once the person logged in on their device. The data was collected in the Netherlands.
The collected data is on two levels: customer journey observations are made on purchase-level. Socio-demographic data is gathered on user-level, thus based on the log-in plug of the respondent. A single user ID might have made different customer journeys that either led to a purchase or no purchase. The collection of the data took place between 1-6-2015 and 31-9-2016.
3.3 Testing for effect on journey length – Poisson Model
P a g e 12 | 34
The probability that the variable yi takes the value y for consumer i is denoted in equation (II) below:
Eq. (II) P( 𝑦𝑖 = 𝑦) = 𝜆i 𝑦 exp (−𝜆𝑖) 𝑦!
The Poisson distribution contains a mean and variance that are equal to λ. This parameter λ is related to the explanatory variables through a link function, shown in equation (III):
Eq. (III)
𝜆𝑖 = 𝑒𝑥𝑝 (𝑥𝑖′𝛽)
For each of the variables in the model the probability that the count variable yi takes the value y for consumer i depends on this exponentiated beta. If x1 changes with one unit (keeping all other variables in the model constant), the expected count (λ) is multiplied by exp(β1) – in case of an continuous
explanatory variable. For a binary explanatory variable, the following happens: if dummy x1 equals 1 rather than 0 (keeping all other variables in the model constant), the expected count (λ) is multiplied by exp(β1).
The initial Poisson model includes the three different touchpoints, namely accommodation, comparison and tour operator. This model answers the question how the journey length is affected by a touchpoint once these touchpoints occur in the journey. Thereafter, the model is adapted in a stepwise manner. Variables concerning characteristics of the touchpoints (duration of each touchpoint and device used to reach the touchpoint) and control variables (gender and age of the customer) are tested in the model while controlling for the model’s information criteria. The control variables consist socio-demographic consumer information, in order to account for individual differences between the consumers. Finally, the model tests for a moderating effect of device used.
3.4 Testing for effect on purchase probability – Logit Model
The second model applies a logistic regression. Given the binary nature of the dependent variable, ordinary least squares cannot be applied. Either a logistic regression or a probit model is applicable. The results of both techniques are relatively similar, however in a logistic regression the probabilities are easier to calculate and also the interpretation of the parameters is found to be more straightforward. Given the mathematical convenience, this paper applies a logistic regression to estimate the impact of the type of touchpoints on the purchase probability of a customer journey.
A logistic regression introduces a latent variable (yi*) that links the independent variables to the
dependent variable. The probability of a purchase depends on the value of the latent variable. The relationship between yi* and Yi (the observed dependent variable) is related as following: Yi = 1 if yi* >
0 and Yi = 0 if yi* ≤ 0. This leads to an estimation model that predicts the probability of a purchase, for
which the formula is shown in equation (IV):
Eq. (IV)
P[Yi = 1] = Ʌ(α + 𝑥𝑖′𝛽) =
exp(𝛼 + 𝑥𝑖′𝛽)
1 + exp(𝛼 + 𝑥𝑖′𝛽)
The specification of the logit model follows the same adaption as the poisson modeling. The initial model includes the three main touchpoints and is stepwise adapted by including journey specific variables (i.e. journey length, duration of the touchpoint and device used) and control variables (i.e.
gender and age of the consumer). This model will also test for the interaction effect between the
P a g e 13 | 34
4. Results
The two models are linked to two different datasets. Given the commercial nature of the travel industry, it is most important to provide insights in effect of touchpoints on the length of journeys that lead to a purchase. Therefore, the first set includes customer journeys that merely ended in a purchase, on which the Poisson model is applied. The second set of data contains all journeys, including no-purchase journeys, on the basis of which the purchase probability can be established via a logistic regression.
4.1. Sample including only purchase journeys
The first dataset includes 3,674 journeys in total. For the variable of journey length (i.e. the total number of touchpoints that appear per journey), several outliers were detected. Multiple values above 4000 can be seen as outliers (Figure 2 and 3 in the appendix). It is expected that these outliers influence the estimation of the model due to the fact that the dependent variable is the length of a customer journey. Therefore, the customer journeys consisting of more than 4,000 touchpoints are excluded from the dataset. The data on journey length is positively skewed (as can be seen in Figure 4 in the appendix). Moreover, 88.1% of the journeys falls into the category of 0-500 touchpoints, as Table 1 shows. The variable journey length has a maximum of 3,717 and an average of 227.3 touchpoints. The standard deviation for journey length is 369, indicating that the data points are spread out over a wider range of values. This reasonably high standard deviation might demonstrate overdispersion in this dataset. This will be accounted for in the final model estimation, described in paragraph 4.3.
Table 1. Distribution number of journeys per level of journey length
Amount of Touchpoints per Journey Frequency Percentage Cumulative Percentage
0 – 100 1822 50% 50% 101 – 200 701 19% 69% 201 – 300 369 10% 79% 301 – 400 208 5% 84% 401 – 500 143 4% 88% 501 – 1000 284 8% 96% 1001 – 2000 109 3% 99% > 2000 35 1% 100%
From the total amount of touchpoints in the different journeys 41.7% is with an accommodation website. 6.8% of the touchpoints is with a comparison website and 38.9% entails an interaction with a tour operator. Therefore, 87.4% of the touchpoints in this dataset are with either one of the focus touchpoints, consolidating the importance of these interactions relative to touchpoints with the app of the travel company or interactions with companies that provide only flight tickets.
P a g e 14 | 34
4.2 Dataset including both purchase and no-purchase journeys
The second dataset consists of 29,012 journeys that comprises both journeys leading to a purchase and no-purchase journeys. 3674 journeys resulted in a purchase, which is approximately 13% of the total number of journeys. All journeys combined lead to a total of 2,456,414 touchpoints, of which the majority was with an accommodation or tour operator website (respectively 36.8% and 36.6%). The remaining touchpoints concern an interaction with a comparison website (8%) and 18.6% other type of touchpoints (e.g., flight ticket websites, search engines or advertising-related).
The variable journey length contains a value of 64,503 touchpoints that is detected as outlier (Figure 5 and 6 in the appendix). This journey is therefore removed from the data. The maximum journey length is now 8,891 touchpoints. The variable has an average of 82.45 touchpoints per journey and a standard deviation of 247.5. For 138,786 touchpoints (5.8% of the total) the duration in seconds per touchpoint was not available. The missing values have been imputed with the mean of the overall duration. Thereafter, the duration variable has an average of 56 seconds per touchpoints and a standard deviation of 106 seconds. Most touchpoints were reached through a fixed device, namely 80%. The remaining touchpoints (20% of the total) were reached via a mobile device.
In order to retrieve an understanding of the control variables, the overall mean of the variable was used to impute the missing values for gender and age. The minimum age in this sample is 17 years old, whereas the maximum age is 94. The average age is 51 years old, with a standard deviation of 13.47 years. 57.7% of the sample is female; 42.3% is male.
4.3. Data transformation
A variable journeylength is created based on the number of touchpoints that were linked to a unique customer journey. This created a new variable that provided the total length of the journey of which a specific touchpoint was part of. Thereafter, three dummy variables were created to provide information whether a touchpoint was either with an accommodation, comparison or tour operator website. These dummy variables are included in the models to account for the effects of the three touchpoints on either journey length or purchase probability. The existing variable gender included values of 1 and 2 for respectively male and female. However this leads to an expected linear relationship between these two values. Therefore, the variable is transformed to a 0/1 level, where 1 represents a male consumer and 0 a female. Furthermore, a dummy variable is created for the character variable device used. For the variables duration, age as well as gender the sample mean was imputed to retrieve a thorough understanding of the effects of these variables.
4.4. Estimation Negative Binomial Regression
P a g e 15 | 34 Table 2. Model selection standard Poisson regression
Model df AIC BIC Log-Likelihood
Model 1 (only touchpoints) 4 561145163 561145209 -280572577 Model 2 (touchpoints + characteristics) 6 557954573 557954642 -278977280
Model 3 (touchpoints + characteristics + control) 8 555919615 555919708 -277959799
Nullmodel 1 576677637 576677648 -288338817
4.4.1. Violations of Poisson Model
There are however three violations of this Poisson model that need to be accounted for. These violations include that the mean is not equal to the variance, that zero events cannot be observed and whether there are more zeros than expected. Concerning the first violation, a Poisson distribution takes on the assumption that the mean is equal to the variance. In case the mean is larger than the variance, underdispersion is detected; if the mean is smaller than the variance, overdispersion is observed. A dispersion test is performed to check whether either one of these appear in the data. The test shows α = 0.858 (z-value = 497.5, p < 0), so the true alpha is significantly larger than zero. This implies that the mean and the variance are not equal and so a standard Poisson regression is not appropriate. In case of overdispersion a Negative Binomial Regression is more applicable to the data. The long tail of the skewed data can be modelled by the Poisson model only if the mean is the same as the variance. The Negative Binomial Regression accounts for this in a way that it can model the long tail by a low mean and a high variance. It therefore captures the heterogeneity in the data.
Concerning the second violation, in both the Poisson model and Negative Binomial Regression the probability to observe events that are zero is always positive. However, in this study zero events cannot be observed by definition. Due to the dependent variable of journey length the observations are all > 1. Therefore, the second violation is applicable to this dataset. Whereas both previous models assume that there are zeros in the distribution of the data, a Truncated Model does not take on this assumption. The third and last violation of Poisson models does not apply to this data, because zero events cannot be observed and thus there cannot be more zeros than expected. However, the Truncated Negative Binomial model cannot be performed by the program R and therefore, a negative binomial regression is applied to overcome the first violation of the poisson model. When interpreting the estimation of the model, it has to be taken into account that the model assumes a probability of zero counts.
4.4.2. Estimation Negative Binomial Regression
In order to find the best model fit for the Negative Binomial Regression different models are created following the same adjustments as is done for the standard Poisson regression. This leads to a best performing model including all variables, in which the journey length is estimated by the three different touchpoints, characteristics of those touchpoints (i.e. duration and device used) and the control variables (gender and age of the consumer). Table 3 shows the lowest AIC, BIC and log-likelihood for model 3.
Table 3. Model selection Negative Binomial Regression
Model df AIC BIC Log-Likelihood
Model 1 (touchpoints) 5 12850227 12850285 -6425108 Model 2 (touchpoints + characteristics) 7 12845864 12845946 -6422925
Model 3 (touchpoints + characteristics + control) 9 12843858 12843963 -6421920
P a g e 16 | 34
Before an interpretation can be made of the analysis that flows from this negative binomial regression, the independency of the variables in the model needs to be analyzed. The Variance Inflation Factor (VIF) scores for each explanatory variable are evaluated and each score is smaller than 3 (Table 11 in the appendix). It can therefore be concluded that the variables in the model are independent from one another. The initial Negative Binomial Regression however does not present an answer to the third sub-question of the general research sub-question: whether there appears an interaction effect between the effect of the touchpoints and the type of device that was used to reach that touchpoint. In the interest of this part of the research question the interaction between the touchpoint variables (accommodation,
comparison and tour operator) and device used are taken into the model as well. The estimation of both
regressions (ex- and including the interaction effect) is presented below in Table 4.
The two models shows significant effects for the included explanatory variables. The effect of tour
operator in the model excluding the interaction effect is slightly less significant than the other variables,
though still significant. High levels of significance are however common when large samples are used; effect size is therefore more interesting when interpreting the effects of the explanatory variables. The three touchpoint variables accommodation, comparison and tour operator have a negative effect on
journey length: if the variable equals 1 rather than 0 (keeping all other variables in the model constant),
the expected length of a journey is multiplied by exp(β). Hence the journey length decreases with respectively 26%, 5% and 1% in the first model and 31%, 42% and 31% in the second model. The two models differ in their effect sizes of the three touchpoints. Where accommodation has the largest negative effect in the first model (.74), the effect is less than either one of the three estimations in the second model (.69, .58 and .69). The variable device used also negatively effects the journey length: the first model estimates the effect of .81, whereas the second model shows a larger negative effect of .63. The second model thus predicts that the journey length decreases with 37% when the variable device
used equals 1 (representing a fixed device) rather than 0 (representing a mobile device). The expected
P a g e 17 | 34 Table 4. Model selection Negative Binomial Regression
Negative Binomial Regression Negative Binomial Regression
excluding interaction effect including interaction effect
Exp(β) P-value Exp(β) P-value
Intercept 1386.48 < 0 *** 1593.99 < 0 *** Accommodation 0.74 < 0 *** 0.69 < 0 *** Comparison 0.95 < 0 *** 0.58 < 0 *** Tour operator 0.99 < .05 * 0.69 < 0 *** Device used 0.81 < 0 *** 0.63 < 0 *** Duration 1.00 < 0 *** 1.00 < 0 *** Age 1.00 < 0 *** 1.00 < 0 *** Gender 0.94 < 0 *** 0.94 < 0 *** Acc:deviceused 1.18 < 0 *** Comp:deviceused . 1.85 < 0 *** Tour:deviceused 1.63 < 0 *** Observations = 834,262 Observations =834,262 Log-Likelihood = -6,421,921 Log-Likelihood = -6,419,867 Likelihood ratio test: Likelihood ratio test: chisq = 27809, p = 2.2e-16 chisq = 31916, p = 2.2e-16
AIC = 12,843,858 AIC = 12,839,757
4.5. Estimation Logit Model
4.5.1. Model selection
As an exploratory analysis on the effect of the touchpoints on the purchase probability of a journey, an ANOVA was applied. The results show a significant difference between the touchpoints and their effects (F = 3949, p < 0). Four different models are tested, which are adapted in a stepwise manner. The information criteria of these models and the null model can be found in Table 5. Model 4 shows the lowest AIC, BIC and log-likelihood and therefore shows the best model fit to the data. The model performs significantly better than a null model (chisq = 65650, p < 0). The third model (excluding the interaction effect between the touchpoints and the device used) also has a low AIC, BIC and log-likelihood, although slightly higher than model 4. This model also performs better than the null model (chisq = 67026, p < 0).
Table 5. Model selection Logistic Regression
Model df AIC BIC Log-Likelihood
Model 1 (only touchpoints) 4 3092034 3092085 -1546013 Model 2 (touchpoints + characteristics) 7 3051222 3051310 -1525604 Model 3 (touchpoints + characteristics + control) 9 3046446 3046560 -1523214
Model 4 (touchp. + char. + control + interaction) 12 3045075 3045228 -1522526
P a g e 18 | 34
4.5.2. Model estimation – coefficient interpretation
Before an estimation can be made, the VIF scores are checked to test whether the variables in the model are independent from each other. The variables in model 3 provide VIF scores < 3 (see Table 12 in the appendix). Model 4 shows high VIF scores, however these occur according to the inclusion of the interactions (see Table 13 in the appendix). The estimation of the two best performing models is shown in Table 6 in order to provide insights between their estimation. A first interpretation can be made from the coefficients of the variables. Both models show a significant negative effect for comparison and age, meaning that an increase in the variable leads to a decrease in the probability of observing Y = 1 (a customer making a purchase). The coefficients of accommodation, tour operator, journey length,
duration and gender are significantly positive and therefore an increase in these variables results in an
increase of observing Y =1. The interaction effect shows a significant positive interaction for each touchpoint with device used.
Table 6. Coefficient estimation of model 3 and 4
4.5.3. Model estimation – odds ratio and marginal effects
To interpret the actual number of the estimate, the odds ratio and marginal effects are generated. The odds ratio provides a number that presents the likelihood of happening versus non-happening, i.e. a customer making a purchase versus not making a purchase. The marginal effect is based on the derivative of the probability with respect to an explanatory variable, where the effect depends on xi. The estimated marginal effect provides information for the average observation (the logitmfx function from the package mfx is used in this study). Both assessments of the independent variables in model 3 and 4 can be found in Table 7.
Looking at the odds ratio’s in both models, the results for journey length and duration are notable because both models show a ratio of 1, meaning that both the length of a customer journey and the duration of a touchpoint are not related to the probability of a customer making a purchase. Furthermore, the models show a negative relation between the age of a consumer and the probability of them making a purchase. Whereas model 3 shows a positive relation between comparison and the purchase probability, model 4 finds a negative relation between the variables. For all other variables in the models (accommodation, tour operator, device used, gender, acc:device, comp:device and tour:device) a positive relation was found.
Model 3 Model 4
Estimate z-value Pr(>|z|) Estimate z-value Pr(>|z|)
P a g e 19 | 34
The marginal effects provide information on how much the probability of a purchase in a journey will increase or decrease following the nature of the explanatory variable. If the dummy variables
accommodation and tour operator increase from 0 to 1 (where 1 equals an interaction with the website),
the probability of a purchase increases with respectively .079 and .064 in model 3 and .019 and .046 in model 4. If the dummy variable comparison increases from 0 to 1, the probability of a purchase decreases with .011 in model 3 and .073 in model 4. The marginal effects for the continuous variables journey
length and duration show an effect of 0 in both models, implying that the probability of a purchase does
not change with an increase in these variables. If dummy variable device used increases from 0 to 1 (where 1 equals a usage of a fixed device), the probability of a purchase increases with .015 in model 3 and .012 in model 4. If the age of a customer increases with one unit, the purchase probability slightly decreases. Both models show a marginal effect of -.001. If the dummy variable gender increases from 0 to 1 (where 1 equals a male customer), the purchase probability increases with .041 in both models. The marginal effect on this estimate is the probability of a touchpoint (not with an accommodation website) reached through a mobile device leading to a purchase.
Table 7. Odds ratio and marginal effects of model 3 and 4
Model 3 Model 4
Odds ratio Marginal effect Odds ratio Marginal effect
Intercept .263 .294 Accommodation 1.541 .079 1.089 .019 Comparison 1.039 -.011 .715 -.073 Tour operator 1.391 .064 1.225 .046 Journey length 1.000 .000 1.000 .000 Duration 1.000 .000 1.000 .000 Device used 2.031 .015 1.738 .012 Age .995 -.001 .995 -.001 Gender 1.186 .041 1.196 .041 Acc:device 1.418 .081 Comp:device 1.464 .090 Tour:device 1.152 .033
The interpretation of the odds ratio for the interaction effect is somewhat different and can be best explained by the formula of the logit. In equation (V) accommodation and device used are placed in the model as an explanation:
Eq. (V)
P[Y=1] = β0 + β1acc + β2device + β3(acc*device)
Both explanatory variables are dummy variables with acc = 0 (the touchpoint is no interaction with an accommodation website) and device = 0 (a mobile device was used to reach the touchpoint). If both explanatory variables equal 0, the probability of Y =1 is identical to β0 (the intercept). Following this
P a g e 20 | 34
with another touchpoint (other than an accommodation website) and reaches this touchpoint through a fixed device, the odds of him or her making a purchase is 2.032 higher than not making a purchase. A negative relation is found for the observation of a customer not visiting an accommodation website through a mobile device, with an odds ratio of .294.
Table 8. Interpretation odds ratio and marginal effects for accommodation touchpoint
Accommodation Device P[Y=1] Odds ratio
1 Yes (1) Fixed (1) β0 + β1 + β2 + β3 .294 + 1.089 + 1.738 + 1.418 = 4.539
2 Yes (1) Mobile (0) β0 + β1 .294 + 1.089 = 1.383
3 No (0) Fixed (1) β0 + β2 .294 + 1.738 = 2.032
4 No (0) Mobile (0) β0 .294
In Table 9 the odds ratio’s for both the comparison and tour operator touchpoints are shown. These show the same relationship as the accommodation touchpoint does. The relation is strongest when a touchpoint is reached through a fixed device (both the focus touchpoints as any other touchpoint). There appears a positive relation when the touchpoint is reached through a mobile device, for both a comparison and tour operator touchpoint (respectively 1.009 and 1.519). The same negative relation was found for comparison and tour operator touchpoints as well.
Table 9. Interpretation odds ratio and marginal effects for comparison and tour operator touchpoints
Observation Odds
ratio
Observation Odds
ratio
1 Comparison touchpoint with fixed device 4.211 Tour operator touchpoint with fixed device 4.409
2 Comparison touchpoint with mobile device 1.009 Tour operator touchpoint with mobile device 1.519
3 Other touchpoint with fixed device 2.032 Other touchpoint with fixed device 2.032
4 Other touchpoint with mobile device .294 Other touchpoint with mobile device .294
Table 10 provides an overview of this research’s hypotheses and the corresponding findings. A further discussion of this paper’s results are discussed in the following chapter.
Table 10. Hypotheses and findings
Hypothesis Supported/Rejected
H1a Touchpoints with accommodation websites reduce the length of a customer journey Supported
H1b Touchpoints with comparison websites reduce the length of a customer journey Supported
H1c Touchpoints with tour operator websites increase the length of a customer journey Rejected
H2a Touchpoints with accommodation websites increase the purchase probability Supported
H2b Touchpoints with comparison websites increase the purchase probability Rejected
H2c Touchpoints with tour operator websites decrease the purchase probability Rejected
H3 The usage of mobile devices (relative to fixed devices) strengthens the effect Rejected
of the touchpoints on the journey length
H4 The usage of mobile devices (relative to fixed devices) strengthens the effect Rejected
P a g e 21 | 34
5. Discussion
This research has been executed in order to gain a deeper understanding of the customer journeys in the online travel industry, in pursuance of the main research question: “How do customer touchpoints with travel companies effect the customer journey?” This was investigated through multiple sub-questions, which included their effect on the journey length, the purchase probability of a customer journey and whether a moderating effect of device usage is at place. A Negative Binomial Regression and a Logit Model have been applied to examine the effects of touchpoints with accommodation, comparison and tour operator websites. Data from GfK was analyzed in order to get an answer to the research question.
5.1. Research implications
The Negative Binomial Regression provided an answer to the question whether the different touchpoints effected the journey length of a customer. Initially, a Poisson model would be applied, however this was adjusted to a Negative Binomial Regression. Given that the assumption of the mean being equal to the variance was violated, a standard Poisson regression could not be practiced. However, another violation was applicable to the data as well, namely that observations of zero could not be made due to the nature of the dependent variable. Following this second violation of a Poisson model a Truncated Negative Binomial Regression is most capable, because this type of model does not predict the probability for zero counts. Unfortunately, it has not been possible to retrieve a prediction from this model. Therefore, the interpretation of the results has to be adjusted according to the assumptions of the model.
Two models have been discussed to present an answer to the first and third sub-question of the main research question. Though these two models performed significantly better than the others, the Negative Binomial Regression including the interaction effect will provide an answer to the research question. This model had the lowest information criteria and foremost it includes the essential significant interaction effects. Where the research expected a positive effect of tour operator touchpoints on the journey length, a negative effect was found for all three touchpoints. The effect of comparison touchpoints was the largest: when a customer interacts with a comparison website, it reduces the length of a journey by 42%. This result confirms the non-linearity of customer journeys, as stated in literature. Comparison websites function as an intermediary channel, through which the total journey length is significantly reduced. The customer does not have to go through an extensive research him- or herself, because the overview is already presented by the comparison website. The extensive research therefore only takes the customer one click and the customer is provided with an overview of the possibilities in the travel industry. The effects of accommodation and tour operator websites were estimated to be similar to each other. Interactions with either one of these websites reduce the journey length by 31%, whereas the expectation was an increase in journey length when a customer interacts with tour operator websites. The analysis shows support for hypotheses 1a and 1b, however hypothesis 1c is rejected. The results show that the usage of fixed device strengthens the effect of touchpoints on journey length, whereas a stronger effect of mobile devices was expected. Therefore hypothesis 3 is rejected. The result can be explained by the large amount of touchpoints that were reached through fixed devices: almost 90% of the interactions were made through a fixed device. Potentially is the importance of fixed devices in online journeys greater than expected. Customers appear to have a preference for either a laptop or PC over mobile devices such as a smartphone or tablet.
P a g e 22 | 34
probability of a customer making a purchase increases by a factor of .019 for accommodation websites and .046 for tour operator websites. The effect of tour operator websites is thus larger, which can be seen as unexpected given the findings of PhoCusWright in 2014. This study stated that the online market is dominated by accommodation companies such as Booking.com and that 38% of the global market is represented by such companies. The findings of this study state that the tour operator websites are not dominated by these companies, in the way that touchpoints with tour operators lead to higher purchase probabilities. Rather than the expected positive effect of comparison websites, a negative relation with purchase probability was found. Therefore, hypothesis 2b is rejected. Whereas mobile devices (relative to fixed devices) were expected to have a stronger effect, the results of this research state the opposite. When a customer reached the touchpoints through a fixed device, the effect of the touchpoint on the purchase probability was stronger. A moderating effect of device usage is found, however not the expected effect for mobile device. Hypothesis 4 is therefore rejected.
5.2. Theoretical and managerial implications
As an addition to previous research, this paper exclusively looks at the different effects that touchpoints with accommodation, comparison and tour operator have on the customer journey and what role device usage plays in these journeys. Previous research has stated that mobile technologies are expected to become the next wave of innovation that would drive travel and tourism (Wang et al. 2011). The moderating effect that has been found for fixed devices implies that such devices (e.g., laptops and PC’s) play an important part in customer journeys, rather than mobile devices. It appears that the capabilities of such devices are not made fully advantage of. Smartphones have the capability to transform customers’ shopping experience, as stated by Persuad and Azhar (2012). This is where all travel companies can take advantage of and adapt their marketing strategy according to that. Responding to the centrality that a smartphone has in personal lives, many opportunities are open for marketers to influence the customer throughout their lives. Marketers should focus on reaching the customer without being pervasive and intrusive and create a customer experience that transcends the traditional customer journey. Creating value for the customer is vital for firms in order to retrieve value from the customers. The effects on purchase probabilities show that comparison websites do not dominate the online travel market. However, this might become the case in the near future. Therefore, this research shows that both tour operators and accommodation companies have to focus on their lead position in this at the moment. Given the commercial nature of the industry, a customer making a purchase is most vital for each company.
5.3. Limitations and further research
Even though this research has presented interesting findings, there are some limitations as well. The main shortcoming is the model that was applied to estimate the effects on journey length. Rather than a standard Negative Binomial Regression, a truncated version of this model would have been more preferred. The results of this research include the probability of zero counts, however zero counts are not observed and should therefore not be taken into the estimation of the model. Another limitation of this study is that it does not take into account where a purchase has been made. Further research can include that knowledge in order to provide a more in-depth answer to what the actual effects on purchase probability are. Finally, the research investigates the effects on journey length, however it is not investigated what the consequences are for these effects. This can be investigated in further research as well.
P a g e 23 | 34
P a g e 24 | 34
6. References
Alba, J., Lynch, J., Weitz, B., Janiszewski, C., Lutz, R., Sawyer, A., & Woods, S. (1997) Interactive home shopping: consumer, retailer, and manufacturer incentives to participate in electronic marketplaces. Journal of Marketing, 61(3), 38-53.
Baxendale, S., Macdonald, E.K., & Wilson, H.N. (2015). The impact of different touchpoints on brand consideration. Journal of Retailing, 91(2), 235-253.
Bonchek, M., & France, C. (2014). Marketing can no longer rely on the funnel. Retrieved from
https://hbr.org/2014/05/marketing-can-no-longer-rely-on-the-funnel
Buhalis, D., & Zoge, M. (2007). The strategic impact of the internet on the tourism industry. Information
and Communication Technologies in Tourism 2007, 481-492.
De Haan, E., Wiesel, T., & Pauwels, K. (2016). The effectiveness of different forms of online advertising for purchase conversion in a multiple-channel attribution framework. International Journal of Research
in Marketing, 33(3), 491-507.
GfK, 2018, About GfK, GfK, accessed on 4th of November 2018,
https://www.gfk.com/about-gfk/about-gfk/.
Granados, N., Gupta, A., & Kauffman, R.J. (2012). Online and Offline Demand of Price Elasticities: Evidence from the Air Travel Industry. Information Systems Research, 23(1), 164-181.
Heerschap, N., Ortega, S., Priem, A., & Offermans, M. (2014). Innovation of tourism statistics through
the use of new big data sources. Retrieved from
http://tsf2014prague.cz/assets/downloads/Paper%201.2_Nicolaes%20Heerschap_NL.pdf
Homburg, C., Jozić, D., & Kuehnl, C. (2017). Customer experience management: toward implementing an evolving marketing concept. Journal of the Academy of Marketing Science, 45(3), 377-401.
Hoekstra, J.C., Leeflang, P.S.H., & Wittink, D.R. (1999). The Customer Concept: The Basis for a New Marketing Paradigm. Journal of Market-Focused Management, 4(1), 43-76.
Kannan, P.K., Reinartz, W., & Verhoef, P.C. (2016). The path to purchase and attribution modeling: Introduction to special section. International Journal of Research in Marketing, 33(3), 449-456.
Kumar, V., Aksoy, L., Donkers, B., Venkatesan, R., Wiesel, T., & Tillmanns, S. (2010). Undervalued or Overvalued Customers: Capturing Total Customer Engagement Value. Journal of Service Research,
13(3), 297-310.
Kumar, V., & Reinartz, W. (2012). Customer Relationship Management. Concept, Strategy and Tools. Berlin Heidelberg: Springer.
Law, R., Leung, K., & Wong, J. (2004). The impact of the Internet on travel agencies. International
journal of Contemporary Hospitality Management, 16(2), 100-107.
Leeflang, P.S.H., Wieringa, J.E., Bijmolt, T.H.A., & Pauwels, K.H. (2015). Modeling Markets. New York: Springer.
Lemon, K.N., & Verhoef, P.C. (2016) Understanding Customer Experience Throughout the Customer Journey. Journal of Marketing: AMA/MSI Special Issue, 80(6), 69-96.
P a g e 25 | 34
Mascarenhas, O.A., Kesavan, R., & Bernacchi, M. (2006). Lasting customer loyalty: a total customer experience approach. Journal of Consumer Marketing, 23(7), 397-405.
Paraskevas, A., & Kontoyiannis, K. (2005). Travel comparison websites: an old friend with new clothes.
Information and Communication Technologies in Tourism 2005, 486-496.
Persuad, A., & Azhar, I. (2012). Innovative mobile marketing via smartphones: are consumers ready?
Marketing Intelligence & Planning, 30(4), 418-443.
PhoCusWright. (2012). U.S. mobile travel bookings to top $8 billion by 2013. Retrieved from
https://www.phocuswright.com/Travel-Research/Research-Updates/2012/U-S-Mobile-Travel-Bookings-to-Top-8-Billion-by-2013
PhoCusWright (2014). Online Travel Agencies: More Than a Distribution Channel. Retrieved from
https://discover.expediapartnercentral.com/wp-content/uploads/2016/12/OnlineTravelAgencies-MoreThanaDistributionChannel.pdf
Richardson, A. (2010). Using customer journey maps to improve customer experience. Harvard
Business Review, 15(1), 2-5.
Schierholtz, R., Kolbe, L.M., & Brenner. W. (2007). Mobilizing customer relationship management: A journey from strategy to system design. Business Process Management Journal, 13(6), 830-852. Sultan, F., & Rohm, A. (2005). The coming era of ‘brand in the hand’ marketing. MIT Sloan
Management Review, 83-89.
Takao, M., Takahasi, S., & Kitamura, M. (2009). Addictive personality and problematic mobile phone use. CyberPsychology & Behavior, 12(5), 501-507.
Venkatesan, R., Petersen, J.A., & Guissoni, L. (2018). Measuring and Managing Customer Engagement Value Through the Customer Journey. In R.W. Palmatier, V. Kumar, & C.M. Harmeling (Eds.), Customer Engagement Marketing (p. 53-74). Cham, Switzerland: Springer International Publishing AG.
Verhoef, P.C., Kannan, P.K., & Inman, J.J. (2015). From multi-channel retailing to omni-channel retailing: introduction to the special issue on multi-channel retailing. Journal of Retailing, 91(2), 174-181.
Wang, D., Park, S., & Fesenmaier, D. R. (2011). The role of smartphones in mediating the touristic experience. Journal of Travel Research, Accepted.
Wang, D., & Xiang, Z. (2012). The new landscape of travel: A comprehensive analysis of Smartphone Apps. In M. Fuchs, F. Ricci, & L. Cantoni (Eds.), Information and Communication Technologies in Tourism 2012 (p. 308–319). Vienna, Austria: Springer Vienna.
Wang, D., Xiang, Z., Law, R., & Tang, P.K. (2016). Assessing Hotel-Related Smartphone Apps Using Online Reviews. Journal of Hospitality Marketing & Management, 25(3), 291-313.
Wolny, J., & Charoensuksai, N. (2014). Mapping customer journeys in multichannel decision-making.
P a g e 26 | 34
7. Appendix
7.1. Codebook variables
Variable name R Description Coding
UserID Unique identification number for each user, based on the plug in of the user
Unique ID
PurchaseID Unique identification number for each journey, to identify which touchpoints are related to a certain journey
Unique ID
Type_touch Initial variable describing the different touchpoints in the dataset 1 = Accommodation Website 2 = Accommodation App 3 = Accommodation Search 4 = Comparison Website 5 = Comparison App 6 = Comparison Search
7 = Tour Operator Website Competitor 8 = Tour operator App Competitor 9 = Tour operator Search Competitor 10 = Tour operator Website Focus Brand 12 = Tour operator Search Focus Brand 13 = Flight tickets Website
14 = Flight tickets App 15 = Flight tickets Search 16 = generic search 18 = affiliates 19 = banner 20 = email 21 = prerolls 22 = retargeting Purchase_any Indication whether a purchase
journey is related to a booking with either the focus brand or a competitor
1 = purchase 0 = no purchase
Journeylength Discrete variable that indicates the sum of total number of touchpoints of a unique PurchaseID
Discrete number, ranging from 1 to 3717 (dataset purchase_any = 1) and 8891 touchpoints (dataset purchase_any = 0 and 1
Used_device Type of device that was used to reach the touchpoint
1 = fixed device (laptop or PC)
0 = mobile device (smartphone or tablet) Duration Duration of each touchpoint in
seconds
Capped at 720 seconds max
Type_touchpoint Addition to type_touch variable, where tour operator focus brand and tour operator competitor were merged
1 = touchpoint with accommodation website 2 = touchpoint with comparison website 3 = touchpoint with tour operator website 0 = other type of touchpoint
Accommodation Dummy variable that indicates whether touchpoint is with an accommodation website or not
