The effectiveness of different forms of touchpoints
across mobile and fixed devices for online
purchases in the travel industry
Predicting the path to successful conversionsBy Ilse Mein
The effectiveness of different forms of touchpoints
across mobile and fixed devices for online
purchases in the travel industry
Predicting the path to successful conversionsBy Ilse Mein
S540545
Pelmolenlaan 31
9679GA, Scheemda
06-50484696
I.Mein@student.rug.nl
University of Groningen
Faculty of Economics and Business
MSc Marketing
Supervisor:
P.S. van Eck
Management summary
Nowadays, online customer journeys become increasingly complex since people encounter more touchpoints in multiple channels across different devices. During the path-to-purchase, people can easily switch between firm-initiated touchpoints and customer-initiated
touchpoints and between using fixed devices and mobile devices. Since the effect of different touchpoints on a purchase decision differ across different devices, it is important to
understand how people engage with touchpoints and devices during their path-to-purchase. Understanding the effectiveness can clarify opportunities for budget allocations and identify wasted marketing spend, but most marketers are still unable to quantify the impact of digital advertising. To understand the effect of each touchpoint across different devices on the path-to-purchase in the travel industry, the study addresses the following research question: To what extent do branded customer-initiated touchpoints, generic customer-initiated
touchpoints and firm-initiated touchpoints influence an online purchase decision and do these effects differ between using mobile or fixed devices? The travel industry is of particular importance since the traditional way of intermediating travel has changed with the rise of online travel agencies. To address the research question, a binary logistic with interaction effects between all forms of touchpoints and devices is estimated, resulting in stable
interpretations of the effect of different touchpoints on a purchase decision and whether these effects are strengthened or weakened by the type of device that is used. The results show that the effect of an online purchase decision is strongest among customer-initiated touchpoints, subdivided into branded customer-initiated touchpoints and generic customer-initiated
Preface
This thesis is written to fulfill the graduation requirements of the MSc Marketing Program at the University of Groningen. I was engaged in researching and writing this thesis from September 2019 to January 2020. Despite some setbacks, I learned a lot about academic writing, coding, and analyzing during this process.
Table of contents
Management summary ... 3 Preface ... 4 1. Introduction ... 6 2. Theoretical background ... 8 3. Research design ... 15 3.1 Data collection ... 15 3.2 Data preparation ... 16 3.3 Methodology ... 16 4. Results ... 19 4.1 Data exploration ... 19 4.2 Model specification ... 20 4.2.1 Specification Model 1 ... 204.2.2 Evolutionary process Model 1 ... 21
4.2.3 Specification Model 2 ... 22 4.3 Model estimation ... 23 4.3.1 Estimation Model 1 ... 23 4.3.2 Estimation Model 2 ... 26 4.4 Model validation ... 27 4.4.1 Validation Model 1 ... 27 4.4.2 Validation Model 2 ... 28
4.5. Logistic regression assumptions ... 28
5. Conclusion ... 30
6. Limitations and further research ... 34
References ... 35
Appendix 1 – Description of touchpoints ... 38
Appendix 2 – Outliers ... 39
Appendix 3a – Correlations Model 1 ... 40
Appendix 3b – Correlations Model 2 ... 41
Appendix 4 – Number of occurrences ... 42
Appendix 5a – Lift curve Model 1 ... 44
Appendix 5b – Lift curve Model 2 ... 45
Appendix 6a – Linearity assumption Model 1 ... 46
Appendix 6b – Linearity assumption Model 2 ... 47
Appendix 7a – Influential values Model 1 ... 48
Appendix 7b – Influential values Model 2 ... 49
1. Introduction
In the last ten years, the advertising landscape has completely changed. Digital advertising has grown enormously and is becoming increasingly important as almost half of the world’s population uses the Internet and people of all age groups are connected to the Internet (Lath, 2019). Given the growing importance of the Internet, retailers are forced to use multiple online marketing channels, such as search engine advertising, display advertising, affiliate advertising or email, to reach potential customers on the Internet (Anderl et al., 2016a). As a result, digital advertising spending in the U.S. will surpass traditional advertising for the first time in 2019 (eMarketer, 2019).
In the online buying process, people encounter more touchpoints in multiple channels.
However, only a few of these touchpoints are under the firm’s control, leading to increasingly complex online customer journeys (Boomtown, 2019). Subsequently, the rapid increase in the use of mobile devices increases the complexity of online customer journeys even more as people can easily switch between mobile and fixed devices in the online path-to-purchase (de Haan et al., 2018). Since the effect of different touchpoints on a purchase decision differ across different devices, it is important to understand how people engage with touchpoints and devices during their path-to-purchase (Kaatz et al., 2019). A better understanding of these effects can clarify opportunities for budget allocations or identify wasted marketing spend (Lovett, 2013). However, most marketers are still unable to quantify the impact of digital advertising and face the challenge of understanding the effect of different touchpoints across different devices on different digital channels (Hall, 2019, Lovett, 2013).
To understand the effect of each touchpoint across different devices in the online path-to-purchase in the travel industry, the study addresses the following research question: To what extent do branded customer-initiated touchpoints, generic customer-initiated touchpoints and firm-initiated touchpoints influence an online purchase decision and do these effects differ between using mobile or fixed devices?
To address the research question, a binary logistic regression is used. A binary logistic
regression predicts the relationship between the touchpoints people are exposed to, the type of device people has used and whether a purchase decision has been made.
The study makes a few contributions to the academic literature. First of all, the study contains a richer data set to evaluate the effect of touchpoints on a purchase decision, resulting in more insights. Secondly, the study tests whether the effect of touchpoints on a purchase decision strengthen or weaken across different devices, resulting in useful information since only a few studies have done this before. Finally, as a managerial contribution, the study clarifies
opportunities for budget allocation for online travel agencies.
2. Theoretical background
The use of a digital medium can provide a well-arranged view of the customers’ online path-to-purchase in which the specific touchpoints people are exposed to and the specific devices people have used are shown (Abhishek et al., 2017, de Haan et al., 2018). As a result,
companies can reach valuable customers at lower costs and be more efficient (Hallikainen et al., 2019). However, the increase in the number of touchpoints and devices during the path-to-purchase in today’s digitally-enriched markets makes evaluating the online path-to-path-to-purchase increasingly complex and companies face the challenge of deciding which digital marketing channels resources should be allocated to (Kuehnl et al., 2019, de Haan et al., 2018,
Hallikainen et al., 2019). Therefore, understanding the impact of different touchpoints across different devices during the path-to-purchase in the travel industry is becoming very
important (Kannan et al., 2016).
The online path-to-purchase
Anderl et al. (2016a) define the online path-to-purchase or online customer journey as “the inclusion of all contacts of any individual customer with a retailer over all online marketing channels, prior to a potential purchase decision”. Li and Kannan (2014) assume that the inclusion of all contacts in the online path-to-purchase consists of a purchase decision hierarchy and that people move through the purchase decision hierarchy when making an online purchase decision. The purchase decision hierarchy can be divided into three stages: (1) consideration stage, (2) visit stage and (3) purchase stage. In the consideration stage, consumers recognize their needs and consider different channels to search for information. In the visit stage, consumers visit the website through a specific channel in order to search for information and to evaluate the alternatives, such as visiting the website directly, using search channels to compare better prices and options or using both. Finally, in the purchase stage, consumers make a purchase (Li and Kannan 2014). Lemon and Verhoef (2016) combine the consideration stage and the visit stage into one overall stage, the pre-purchase stage.
The different forms of touchpoints
In each of the stages in the online path-to-purchase, consumers encounter different direct or indirect contacts with the firm. By identifying and evaluating these contacts, firms discover specific contact moments that encourage customers to continue or discontinue a purchase decision (Lemon & Verhoef, 2016). These contacts refer to touchpoints and can be defined as “an episode of direct or indirect contact with the firm” (Baxendale et al., 2015). These
touchpoints influence people’s perception about a product, service or brand, but only some of these touchpoints are under the firm’s control (Patterson, 2009). Therefore, touchpoints can be distinguished between customer-initiated touchpoints and firm-initiated touchpoints (Li and Kannan, 2014). In customer-initiated touchpoints consumers search for information on their own initiative, while in firm-initiated touchpoints firms initiate marketing
communications, such as emails and display advertising and determines timing and exposure (Anderl et al., 2016a, Anderl et al., 2016b, Li and Kannan, 2014). De Haan et al. (2016) try to understand how customer-initiated and firm-initiated touchpoints generate traffic, influence purchase decisions and contribute to revenue. The authors find that customer-initiated touchpoints are more effective than firm-initiated touchpoints since customer-initiated touchpoints exhibit higher sales elasticities. This is likely because customer-initiated
touchpoints seem less intrusive and more relevant since they require a level of interest from the consumer. However, the authors claim that firm-initiated touchpoints affect people that have not yet recognized a specific need. The study of Anderl et al. (2016a) examines the effect of customer-initiated and firm-initiated touchpoints on purchase decisions. The authors find an increase in purchase probability if people first react to a firm-initiated touchpoint and then visit the website via a customer-initiated touchpoint. The switch is likely because people limit their choices by actively searching for useful information, resulting in a shorter purchase time.
touchpoint. A decrease is likely since it might be that potential customers are in the process of identifying alternatives (Anderl et al., 2016a).
The different forms of devices
A device can be distinguished into fixed and mobile devices (de Haan et al., 2018). Fixed devices include laptops and desktops and mobile devices include smartphones and tablets. The authors argue that each stage of the online path-to-purchase can be performed on a different device. However, the choice for a device depends on the risk involved in using the device and factors that influence or helps in reducing the risk (de Haan et al., 2018).
After analyzing the online path-to-purchase across different devices, de Haan et al. (2018) find mobile devices as the preferred option in the pre-purchase stage. The benefits of searching for information on a mobile device more easily outweigh the risk of making a wrong purchase. However, in the purchase stage, the authors find the use of a fixed device more suitable. Compared to the pre-purchase stage, the need for information processing is much more fine-grained in the purchase stage. Besides, entering detailed payment information without making mistakes may require a larger and higher-resolution screen of a laptop or desktop. In addition, if the risk of making a transaction on a mobile device is high, customers prefer waiting to access a device that is perceived as more secure. Finally, the authors find that purchase probability is significantly higher when customers switch from a mobile device to a fixed device. Especially when the perceived risk within a product category are higher, the price of a product is higher, and the experience with the product and the online retailer is lower (de Haan et al., 2018).
The interaction between different touchpoints and devices
Understanding the effect of customer-initiated and firm-initiated touchpoints across different devices is becoming increasingly important since the effect of touchpoints on purchase probability differ across different devices (Kaatz et al., 2019). Kaatz et al. (2019) study the effect of different devices on customer-initiated and firm-initiated touchpoints and find that customer-initiated touchpoints such as direct type-in, search engine advertising, search engine optimization and referrer account for the highest purchase probability across all devices. This finding is in line with a finding of de Haan et al. (2016) who find that customer-initiated touchpoints are more effective than firm-initiated touchpoints.
advertising account for the highest purchase probability across mobile devices. A possible explanation for a high purchase probability of direct type-ins across mobile devices may be the emergence of the implementation of an omnichannel strategy in which retailers provide seamless shopping experiences in which people can use the integrated channels constantly, interchangeably, and simultaneously (Quix, 2019). The ubiquitous services provided by an omnichannel strategy are perceived as favorably across mobile users (Hubert et al., 2017). A high purchase probability for search engine advertising across mobile devices is due to the limit presentation of search results since the screen size of mobile devices are smaller, resulting in clicking on sponsored search results more frequently than non-sponsored results when searching for a keyword (Kaatz et al., 2017). Kaatz et al. (2019) find the effect of customer-initiated touchpoints such as search engine optimization and referrer account for the highest purchase probability across fixed devices. Due to larger screen sizes on fixed devices, the presentation of search results is no longer limited, resulting in clicking on organic search results more frequently than sponsored search results (Kaatz et al., 2017). Referrer customer-initiated touchpoints are more favorably across fixed devices since a fixed device is more suitable for inducing casual browsing behavior (Kaatz et al., 2019). Finally, mobile devices are more suitable for purchases that do not require extensive information search, such as well-known products, while fixed devices are more suitable for purchases that do require extensive information search, such as high-involvement products (Kaatz et al., 2019).
Hypotheses
De Haan et al. (2016) examines the impact of customer-initiated and firm-initiated
touchpoints and find that customer-initiated touchpoints are more effective compared to firm-initiated touchpoints since customer-firm-initiated touchpoints exhibit higher sales elasticities. However, they also claim that people who have not yet recognized a specific need can be affected by firm-initiated touchpoints. In another study, de Haan et al. (2018) examines the impact of different devices on the online path-to-purchase. The authors find that mobile devices are the preferred option in the pre-purchase stage since the risk of making a wrong purchase is lower. Fixed devices are the preferred option in the purchase stage since there are high risks involved in making a transaction. Based on these findings, the following
hypotheses will be tested:
H1b: The effect of firm-initiated touchpoints is stronger across fixed devices compared to mobile devices.
H2a: Customer-initiated touchpoints positively affects a customer’s purchase probability. H2b: The effect of customer-initiated touchpoints is stronger across fixed devices compared to mobile devices.
H3: Customer-initiated touchpoints are more effective than firm-initiated touchpoints. A further distinction between customer-initiated touchpoints can be made based on brand usage. The distinction results in branded customer-initiated touchpoints, including direct type-ins and branded search, and generic customer-initiated touchpoints, including comparisons sites and generic search (Kaatz et al., 2019). The authors find that branded customer-initiated touchpoints account for the highest purchase probability across mobile devices and generic customer-initiated touchpoints account for the highest purchase probability across fixed devices. Based on these findings, the following hypotheses will be tested:
H4a: Branded customer-initiated touchpoints positively affect a customer’s purchase probability.
H4b: The effect of branded customer-initiated touchpoints is stronger across mobile devices compared to fixed devices.
H5a: Generic customer-initiated touchpoints positively affect a customer’s purchase probability.
Figure 1. Conceptual model 1
Finally, Anderl et al. (2016a) show that the purchase probability increases if the path-to-purchase started with firm-initiated touchpoints and ended with customer-initiated
touchpoints. The switch between both touchpoints indicate a shorter purchase time because people limit their choice and are actively searching for new information. Furthermore, de Haan et al. (2018) show that purchase probability is significantly higher when customers switch from a mobile device to a fixed device. Especially when product category-related perceived risks are higher, the product price is higher, and the experience with the product category and online retailer is lower. Fixed devices are more suitable for high-involvement products. Since the purchase decision in the travel industry is associated with higher risks, higher prices, complex choices and thus high-involvement (Lin et al., 2009), the study assumes the following hypotheses:
H6: The purchase probability is higher if people first encounter firm-initiated touchpoints and then use customer-initiated touchpoints compared to only firm-initiated touchpoints or
customer-initiated touchpoints.
3. Research design
This chapter starts with a detailed explanation of the data collection process. After that, the data preparation for analysis and methodology will be discussed in detail.
3.1 Data collection
To estimate the effectiveness of different touchpoints across different devices in the online path-to-purchase, an individual-level data set is used consisting of online purchase journey travel data. The data set is provided by the company Growth of Knowledge (GfK) and collected in collaboration with a travel agency in the Netherlands. The data collection started on May 6, 2015 and ended on September 31, 2016. The data were collected by using plug-ins and thus registered on a personal level. Using plug-ins enables measuring Internet behavior passively since plug-ins register what is shown on the consumer’s screen, such as URLs, search queries, search results, search ads, clicks, display ads, content/keywords, videos, and pre-rolls. Subsequently, due to plug-ins, measuring advertising exposure, website visiting behavior and search behavior are possible. Altogether, plug-ins allow for gaining deeper insights about the consumer decision journey and making a link with purchase acts. The data include detailed information about all touchpoints consumers are exposed to, the type of devices on which the touchpoints are reached and whether the touchpoints were followed by a purchase or not. In the dataset, touchpoints differentiate between firm-initiated touchpoints and customer-initiated touchpoints. Firm-initiated touchpoints consist of
affiliates, banners, emails, pre-rolls and retargeting and customer-initiated touchpoints consist of accommodations, information/comparison, tour operator/travel agency and flight tickets. Anderl et al. (2016a) classify customer-initiated touchpoints into branded customer-initiated touchpoints and generic customer-initiated touchpoints. Following the example of Anderl et al. (2016a), if the customer-initiated touchpoint contains the focus brand or a competitor, the customer-initiated touchpoints are branded; otherwise, the customer-initiated touchpoint is generic. A clear overview of the classification of touchpoints is provided in Appendix 1. The data set covers 29011 orientations and 3674 purchases of which 192 purchases with the focus brand. In table 1, a description of the online consumer purchase journey data is provided.
Description Online customer purchase journey
Number of journeys 29011
Number of conversions any brand 3674
Conversion rate any brand 13%
Number of conversions focus brand 192
Conversion rate focus brand 1%
Number of touchpoints: Firm-initiated
Customer-initiated
43069 2348842 Journeys per device:
Fixed Mobile
22054 6957
Table 1. Descriptive statistics
3.2 Data preparation
Before estimating the effectiveness of different touchpoints across different devices in the online path-to-purchase, the dataset needs to be prepared. A first step is to add new columns to the dataset to count how often each touchpoint across each device occurs in the data. Next, unnecessary columns will be removed from the dataset. Then, individual-level data will be aggregated, resulting in a clear overview of all customer journeys and exposure to specific touchpoints, devices and purchases. Finally, the dataset will be checked for outliers and multicollinearity.
3.3 Methodology
𝑙𝑜𝑔𝑖𝑡 𝑝 = log 𝑜 = log 𝑝 − 1 − 𝑝 = 𝛽/+ 𝛽1𝐶𝐼𝐺1+ 𝛽1𝐶𝐼𝐺1∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽:𝐶𝐼𝐺:+ 𝛽:𝐶𝐼𝐺:∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽;𝐶𝐼𝐺;+ 𝛽;𝐶𝐼𝐺; ∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽<𝐶𝐼𝐺<+ 𝛽<𝐶𝐼𝐺<∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽=𝐶𝐼𝐺=+ 𝛽=𝐶𝐼𝐺=∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽>𝐶𝐼𝐺>+ 𝛽>𝐶𝐼𝐺> ∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽?𝐶𝐼𝐵?+ 𝛽?𝐶𝐼𝐵?∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽A𝐶𝐼𝐵A+ 𝛽A𝐶𝐼𝐵A∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽B𝐶𝐼𝐵B+ 𝛽B𝐶𝐼𝐵B ∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1/𝐶𝐼𝐵1/+ 𝛽1/𝐶𝐼𝐵1/∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽11𝐶𝐼𝐵1:+ 𝛽11𝐶𝐼𝐵1:∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1:𝐶𝐼𝐺1; + 𝛽1:𝐶𝐼𝐺1;∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1;𝐶𝐼𝐺1<+ 𝛽1;𝐶𝐼𝐺1<∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1<𝐶𝐼𝐺1=+ 𝛽1<𝐶𝐼𝐺1=∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1=𝐶𝐼𝐺1>+ 𝛽1=𝐶𝐼𝐺1>∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1>𝐹𝐼1A+ 𝛽1>𝐹𝐼1A∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1?𝐹𝐼1B+ 𝛽1?𝐹𝐼1B ∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1A𝐹𝐼:/+ 𝛽1A𝐹𝐼:/∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽1B𝐹𝐼:1+ 𝛽1B𝐹𝐼:1∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽:/𝐹𝐼::+ 𝛽:/𝐹𝐼:: ∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝛽:1𝑗𝑜𝑢𝑟𝑛𝑒𝑦𝑙𝑒𝑛𝑔𝑡ℎ
In model 1, each touchpoint is classified as initiated generic (CIG), customer-initiated branded (CIB) or firm-customer-initiated (FI). The variable device is classified as a factor with two levels. 0 indicates a fixed device and 1 indicates a mobile device. To account for
influential observations in the continuous predictors, the variable journey length is included as a control variable in the model. The variable indicates the length of each customer journey. In this model, the interaction effect between each touchpoint and each device will be tested.
To test hypotheses 6 and 7, model 2 is specified as following:
𝑙𝑜𝑔𝑖𝑡 𝑝 = log 𝑜 = log 𝑝 − 1 − 𝑝 = 𝛽/+ 𝛽1𝐹𝐼 + 𝛽:𝐶𝐼 + 𝛽;𝐹𝐼 ∗ 𝐶𝐼 + 𝛽<𝑚𝑜𝑏𝑖𝑙𝑒 + 𝛽=𝑓𝑖𝑥𝑒𝑑 + 𝛽>𝑚𝑜𝑏𝑖𝑙𝑒 ∗ 𝑓𝑖𝑥𝑒𝑑 + 𝛽?𝑗𝑜𝑢𝑟𝑛𝑒𝑦𝑙𝑒𝑛𝑔𝑡ℎ
In model 2, touchpoints are classified as firm-initiated touchpoints (FI) and customer-initiated touchpoints (CI). In this model, the variable device is no longer classified as a factor, but as a numeric variable. The variable journey length is included as control variable to account for influential observations in the continuous predictors. The model test whether there is an interaction effect between firm-initiated touchpoints and customer-initiated touchpoints, and between mobile devices and fixed devices.
probability of observing the dependent variable. Second, the coefficients can be interpreted based on the odds ratio. The odds ratio indicates the ratio between the probability of
somebody purchasing to the probability of somebody not purchasing.
4. Results
From May 2015 to September 2016, online purchase journey travel data was collected in The Netherlands. This chapter first starts with an exploration of the online purchase journey travel data and then aims to find touchpoints that influence the choice for purchasing on specific devices.
4.1 Data exploration
First, the data were checked for outliers. Within the variables that showed potential outliers, the touchpoint flight tickets website was the only column that shows a real outlier (Appendix 2). The maximum journey length for this touchpoint is 64.501. Looking at the maximum journey length of all other touchpoints, the maximum journey length for this touchpoint differs far from the rest. The journey length for this variable is 10 times the size of the
second-highest journey length which is a journey length of 6.192. Removing this observation from the dataset results in a decrease in average from 7.35 to 5.13, which is more in line with the observations of other touchpoints in the dataset. The observation shows that this consumer has been exposed 64.501 times to the touchpoint flight tickets website. Therefore, it is
assumed that this observation is due to incorrect measurements and can have a huge impact on the results. The observation belongs to purchase ID 11878 and has therefore been removed from the dataset.
Looking at the correlations between variables, a higher value indicates a stronger
positive/negative direct relationship between two variables. For model 1 (Appendix 3a), the highest correlation coefficient in the dataset is .7, which is considered to be a strong
Therefore, when estimating the model, the variables accommodations website, tour
operator/travel agent website competitor and journey length will be used as control variables. Furthermore, there are moderate relationships between the following variables:
information/comparison website and journey length, tour operator/travel agent website focus brand and journey length, generic search and journey length, information/comparison website and tour operator/travel agent website competitor, accommodations search, tour
operator/travel agent search competitor, flight ticket search and generic search, and tour operator/travel agent website focus brand and retargeting.
For model 2 (Appendix 3b), the highest correlation coefficient in the dataset is 1.0, which is considered to be a strong relationship. To reduce high correlation among variables, the variable journey length is transformed into a log variable. This results into the following observations: There is a strong relationship between the variables customer-initiated touchpoints and fixed devices (.9). The correlation coefficient for customer-initiated touchpoints and mobile devices is .5 and the correlation coefficients for firm-initiated touchpoints and fixed devices is .4. Thus, there is a moderate relationship between customer-initiated touchpoints and mobile devices and between firm-initiated touchpoints and fixed devices. The high correlations between variables will be taken into account when interpreting the results.
In total, the dataset consists of 29011 journeys. Of these 29011 journeys, 3674 purchases were made, of which 192 with the focus brand. The conversion rate for purchases with the travel agency or a competitor is 13% and the conversion rate for purchases with the travel agency is 1%. The most common touchpoints in the dataset are accommodations website,
information/comparison website, tour operator/travel agent website competitor, tour
operator/travel agent website focus brand, flight tickets website, and flight tickets app. Of the 29011 journeys, 22054 journeys have been made on a fixed device and 6967 journeys have been made on a mobile device. A clear overview of how many times a touchpoint occurs on a specific device is provided in Appendix 4.
4.2 Model specification 4.2.1 Specification Model 1
quadratic specification (BIC = 20622.38 vs. 20787.85). The variable CIG2 was specified in a quadratic form as the difference in BIC between quadratic and linear resulted in a better fit (BIC = 20853.72 vs. 20859.57). The variable CIG3 was first specified in a linear and then in a quadratic fashion. This showed that the linear specification resulted in a better fit (BIC = 20862.16 vs. 20884.54). The variable CIG4 was found to perform statistically better in a quadratic specification, compared to a linear specification (BIC = 20859.59 vs. 20878.65). The variable CIG5 was specified in a linear form as the difference in BIC between linear and quadratic resulted in a better fit (BIC = 20854.24 vs. 20873.82). The same applies to the variables: CIG6 (BIC = 20891.41 vs. 20892.56), CIB8 (BIC = 20875.31 vs. 20879.23), CIB9 (BIC = 20893.16 vs. 20894.25), CIB12 (BIC = 20888.34 vs. 20892.79), CIG15 (BIC = 20879.96 vs. 20890.24), CIG16 (BIC = 20880.33 vs. 20890.95), FI18 (BIC = 20894.43 vs. 20895.86), FI19 (BIC = 20889.52 vs. 20891.8) and FI20 (BIC = 20885.48 vs. 20889.71). Therefore, these variables are specified in a linear fashion. The variable CIB7 was found to perform statistically better in a quadratic specification, compared to a linear specification (BIC = 20667.2 vs. 20876.22). The same applies to the variables: CIB10 (BIC = 20832.92 vs. 20892.02), CIG13 (BIC = 20862.87 vs. 20875.66), CIG14 (BIC = 20846.55 vs. 20850.7), FI21 (BIC = 20878.02 vs. 20881.63) and FI22 (BIC = 20884.39 vs. 20893.2). Therefore, these variables are specified in a quadratic fashion.
4.2.2 Evolutionary process Model 1
𝐶𝐼𝐺:TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺;+ 𝐶𝐼𝐺;∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺<+ 𝐶𝐼𝐺<TUVWXVYZ[+ 𝐶𝐼𝐺<∗ 𝑚𝑜𝑏𝑖𝑙𝑒 +
𝐶𝐼𝐺<TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺=+ 𝐶𝐼𝐺=∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺>+ 𝐶𝐼𝐺>∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵?+ 𝐶𝐼𝐵?TUVWXVYZ[+ 𝐶𝐼𝐵?∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵?TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵A+ 𝐶𝐼𝐵A∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵1/+ 𝐶𝐼𝐵1/TUVWXVYZ[+ 𝐶𝐼𝐵1/∗
𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵1/TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐵1:+ 𝐶𝐼𝐵1:∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1;+ 𝐶𝐼𝐺1;TUVWXVYZ[+ 𝐶𝐼𝐺1;∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1;TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1<+ 𝐶𝐼𝐺1<TUVWXVYZ[+ 𝐶𝐼𝐺1<∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1<TUVWXVYZ[∗
𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1=+ 𝐶𝐼𝐺1=∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐶𝐼𝐺1>+ 𝐶𝐼𝐺1>∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼1A+ 𝐹𝐼1B∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼:/+ 𝐹𝐼:/∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼:1+ 𝐹𝐼:1TUVWXVYZ[+ 𝐹𝐼:1∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼:1TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼::+ 𝐹𝐼::TUVWXVYZ[+ 𝐹𝐼::∗
𝑑𝑒𝑣𝑖𝑐𝑒 + 𝐹𝐼::TUVWXVYZ[∗ 𝑑𝑒𝑣𝑖𝑐𝑒 + 𝑗𝑜𝑢𝑟𝑛𝑒𝑦𝑙𝑒𝑛𝑔𝑡ℎ".
A null-model results in a BIC of 22055.87. By using this model, the BIC decreases to
20543.24. An ANOVA test indicates that the model significantly performs better compared to the null-model (p = .00).
4.2.3 Specification Model 2
During estimation, variables were specified in linear and quadratic forms to find optimal specifications. The calculation of the quadratic variables in is: (x-µ(x))^2, where x indicates the value of independent variable, and µ indicates the mean of that independent variable. The variable FI was specified in a quadratic form as the difference in BIC between quadratic and linear resulted in a better fit (BIC = 18406.26.72 vs. 18406.51). The variable CI was first specified in a linear and then in a quadratic fashion. This showed that the linear specification resulted in a better fit (BIC = 18331.08 vs. 18357.28). The variable fixed was found to perform statistically better in a quadratic specification, compared to a linear specification (BIC = 18382.60 vs. 18400.73). The variable mobile was specified in a linear form as the difference in BIC between linear and quadratic resulted in a better fit (BIC = 18269.67 vs. 18353.99). For model 2, the best performing model is an estimation with the following variables: "𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒_𝑎𝑛𝑦 ~ 𝐹𝐼 + 𝐹𝐼TUVWXVYZ[+ 𝐶𝐼 + 𝐹𝐼 ∗ 𝐶𝐼 + 𝐹𝐼TUVWXVYZ[∗ 𝐶𝐼 + 𝑓𝑖𝑥𝑒𝑑 +
𝑓𝑖𝑥𝑒𝑑TUVWXVYZ[+ 𝑚𝑜𝑏𝑖𝑙𝑒 + 𝑓𝑖𝑥𝑒𝑑 ∗ 𝑚𝑜𝑏𝑖𝑙𝑒 + 𝑓𝑖𝑥𝑒𝑑TUVWXVYZ[∗ 𝑚𝑜𝑏𝑖𝑙𝑒 + 𝑗𝑜𝑢𝑟𝑛𝑒𝑦𝑙𝑒𝑛𝑔𝑡ℎ".
A null-model results in a BIC of 22055.87. By using this model, the BIC decreases to
4.3 Model estimation 4.3.1 Estimation Model 1 Interpretation of the coefficients
First, the variables CIG14quadratic, journeylength, device:CIG2, deviceCIG2quadratic, device:CIG5, device:CIB8, deviceCIG14 and device:CIG14quadratic have been removed from the model, because the coefficients resulted in NA’s. A possible explanation for the NA’s may be that there are not enough data points for these variables. The results of the logit model are shown in table 2. The variables CIG1, device:CIG1, CIB7, CIB7quadratic,
device:CIB7 and device:CIB7quadratic are used as control variables and therefore not been used for interpretation. Table 2 shows that for firm-initiated touchpoints on fixed devices, e-mail and retargeting have a positive and significant relationship with purchases, while pre-rolls have a negative and significant relationship with purchases. In contrast to firm-initiated touchpoints of fixed devices, firm-initiated touchpoints on mobile devices show that pre-rolls have a positive and significant relationship with purchases. However, the quadratic term of pre-rolls shows a negative relationship with purchases, indicating that pre-rolls have a positive effect on purchases until a turning point is reached (1.534e+00 / (2*6.091e-01) = 1.259235).
For generic customer-initiated touchpoints on fixed devices, accommodations search and flight tickets search have a positive and significant relationship with purchases.
Accommodations app also has a positive and significant relationship with purchases. However, the quadratic term of accommodations app shows a negative relationship with purchases, indicating that accommodations app has a positive effect on purchases until a turning point is reached (5.059e-03 / (2*4.803e-06) = 526.65). For generic customer-initiated touchpoints on mobile devices, information/comparison search show a positive and
significant relationship with purchases, while accommodations search and flight ticket search show a negative and significant relationship with purchases. Information/comparison website has a positive and significant relationship with purchases. However, the quadratic term of information/comparison website shows a negative relationship with purchases, indicating that information/comparison website has a positive effect on purchases until a turning point is reached (5.757e-03 / (2*2.147e-05) = 134.0708).
purchases until a turning point is reached (4.622e-03 / (2*1.396e-06) = 1655.444). For branded customer-initiated touchpoints on mobile devices, tour operator/travel agent search focus brand has a negative and significant relationship with purchases. The quadratic term of tour operator/travel agent website focus brand also has a negative sign, indicating that for low values the relation might be positive, but for high values the relation becomes negative.
Interpretation of odds ratio
Table 2 also shows the odds ratio for the final logit model. For firm-initiated touchpoints on fixed devices, the odds ratio for pre-rolls is 0.882, which means it decreases the probability of purchasing compared to not purchasing with 11.8%. The odds ratio for e-mail is 1.020, which indicates it increases the probability of purchasing compared to not purchasing with 2%. The odds ratio for retargeting is 1.004, which indicates it increases the probability of purchasing compared to not purchasing with 0.4%. For firm-initiated touchpoints on mobile devices, the odds ratio for pre-rolls is 1.635 and the odds ratio for pre-rolls squared is 0.544, which means that for the values of pre-rolls ∈ (−∞,1.259235), the odds are increasing as pre-rolls
increases, but afterwards odds decreases as pre-rolls decreases.
For generic customer-initiated touchpoints on fixed devices, the odds ratio for
accommodations search is 1.039 and the odds ratio for flight tickets search is 1.035, which indicates it increases the probability of purchasing compared to not purchasing with 3.9% and 3.5%. The odds ratio for accommodations app and accommodations app squared is 1.00, which indicates that there is no relation between purchasing compared to not purchasing. For generic customer-initiated touchpoints on mobile devices, the odds ratio for
odds ratio for tour operator/travel agent search focus brand is 0.773, which indicates it decreases the probability of purchasing compared to not purchasing with 27.8%.
Coefficients: Estimate Std. Error Z value Pr(|z|) Odds ratio
(Intercept) -2.257e+00 2.447e-02 -92.252 < 2e-16 *** 0.1046236
CIG1 5.635e-03 2.614e-04 21.561 < 2e-16 *** 1.0056513
Mobile -2.759e-01 5.483e-02 -5.031 4.88e-07 *** 0.7589205 CIG2 5.059e-03 1.52e-03 3.485 0.000492 *** 1.0050723 CIG2quadratic -4.803e-06 1.910e-06 -2.515 0.011891 * 0.9999952 CIG3 3.803e-02 1.632e-02 2.330 0.019790 * 1.0387638
CIG4 -5.692e-04 1.206e-03 -0.472 0.636984 0.9994310
CIG4quadratic 3.990e-07 1.729e-06 0.231 0.817483 1.0000004
CIG5 6.207e-04 4.652e-04 1.334 0.182116 1.0006209
CIG6 -8.414e-03 5.375e-02 -0.157 0.875602 0.9916210
CIB7 4.585e-03 3.303e-04 13.879 < 2e-16 *** 1.0045953
CIB7quadratic -3.229e-06 2.497e-07 -12.929 < 2e-16 *** 0.9999968
CIB8 -1.060e-03 1.620e-03 -0.654 0.512983 0.9989408
CIB9 -1.434e-02 3.202e-02 -0.448 0.654186 0.9857577
CIB10 4.622e-03 6.082e-04 7.600 2.97e-14 *** 1.0046330 CIB10quadratic -1.396e-06 2.155e-07 -6.479 9.22e-11 *** 0.9999986
CIB12 1.570e-01 1.099e-01 1.428 0.153176 1.1700071
CIG13 9.686e-04 6.631e-04 1.461 0.144115 1.0009690
CIG13quadratic -1.736e-07 1.575e-07 -1.103 0.270195 0.9999998
CIG14 -2.744e-03 1.676e-03 -1.637 0.101587 0.9972593
CIG15 3.448e-02 1.454e-02 2.371 0.017760 * 1.0350787
CIG16 -2.081e-03 3.613e-03 -0.576 0.564643 0.9979210
FI18 1.394e-02 1.406e-02 0.991 0.321586 1.0140349
FI19 -4.498e-02 3.106e-02 -1.448 0.147568 0.9560197
FI20 1.958e-02 9.344e-03 2.096 0.036116 * 1.0197742 FI21 -1.254e-01 4.474e-02 -2.803 0.005064 ** 0.8821473
FI21quadratic 8.836e-04 6.861e-04 1.288 0.197840 1.0008840
FI22 3.620e-03 1.817e-03 1.992 0.046354 * 1.0036267
FI22quadratic -2.953e-06 2.128e-06 -1.387 0.165309 0.9999970
device:CIG3 -4.364e-02 2.222e-02 -1.964 0.049539 * 0.9573005 device:CIG4 5.757e-03 2.677e-03 2.151 0.031482 * 1.0057737 device:CIG4quad -2.147e-05 7.825e-06 -2.743 0.006083 ** 0.9999785 device:CIG6 1.096e-01 6.262e-02 1.750 0.080117 . 1.1158235
device:CIB7 2.322e-03 8.454e-04 2.747 0.006022 ** 1.0023246
device:CIB7quad -3.091e-06 9.519e-07 -3.247 0.001166 ** 0.9999969
device:CIB9 5.082e-02 4.175e-02 1.217 0.223432 1.0521368
device:CIB10 1.755e-03 1.694e-03 1.036 0.300218 1.0017566
device:CIB10quad -5.467e-06 1.813e-06 -3.015 0.002569 ** 0.9999945 device:CIB12 -3.254e-01 1.503e-01 -2.166 0.030349 * 0.7222526
device:CIG13 3.569e-03 2.263e-03 1.577 0.114774 1.0035757
device:CIG13quad -6.575e-06 4.265e-06 -1.542 0.123149 0.9999934
device:CIG15 -3.668e-02 1.767e-02 -2.076 0.037905 * 0.9639874
device:CIG16 -2.889e-03 4.661e-03 -0.620 0.535399 0.9971153
device:FI19 -1.575e-02 7.326e-02 -0.215 0.829775 0.9843741
device:FI20 -8.778e-03 1.759e-02 -0.499 0.617803 0.9912607
device:FI21 1.534e+00 4.722e-01 3.248 0.001163 ** 1.6354491 device:FI21quad -6.091e-01 1.995e-01 -3.052 0.002271 ** 0.5438461
device:FI22 4.968e-03 9.017e-03 0.551 0.581684 1.0049800
device:FI22quad 1.889e-05 4.936e-05 0.383 0.701989 1.0000189
Table 2. Results logit model 1
4.3.2 Estimation Model 2 Interpretation of the coefficients
Table 3 shows that mobile devices have a negative and significant relationship with purchases. The interaction between using mobile and fixed devices has a positive and significant relationship with purchases. However, the quadratic term of the interaction between using mobile and fixed devices shows a negative relationship with purchases,
indicating that the interaction between using mobile and fixed devices has a positive effect on purchases until a turning point is reached (6.078e-06 / (2*2.863e-09) = 1061.474).
The odds ratio for mobile devices, the interaction between using mobile and fixed devices, and the interaction between using mobile and fixed devices squared is 1.00, which indicates that there is no relation between purchasing compared to not purchasing.
Coefficients: Estimate Std. Error Z value Pr(|z|) Odds ratio
(Intercept) -4.318e+00 6.551e-02 65.907 < 2e-16 *** 0.01333182
FI -3.772e-04 1.637e-03 -0.230 0.817803 0.99962292
CI -2.644e-04 1.648e-04 -1.604 0.108769 0.99973567
FIquadratic -1.284e-06 2.378e-06 -0.540 0.589131 0.99999872
Mobile -2.336e-03 2.485e-04 -9.404 < 2e-16 *** 0.99766624
Fixed NA NA NA NA NA
Fixedquadratic -8.278e-08 5.710e-08 -1.450 0.147189 0.99999992
journeylength 7.062e-01 1.886e-02 37.449 < 2e-16 *** 2.02628554
FI:CI 9.737e-07 1.242e-06 0.784 0.433008 1.00000097
FIquadratic:CI 1.311e-10 1.341e-09 0.098 0.922074 1.00000000
Mobile:Fixed 6.078e-06 1.188e-06 5.115 3.13e-07 *** 1.00000608 Mobile:Fixedquad -2.863e-09 7.956e-10 -3.598 0.000321 *** 1.00000000
Table 3. Results logit model 2
4.4 Model validation 4.4.1 Validation Model 1
The estimation of the logit model indicates that the logit model outperforms a null-model (𝑋: = 12204.9, p = .00). Concerning face-validity, the interaction between the variable
accommodations app and fixed devices is expected to be insignificant. The results show that this variable is significant. This seems illogical because the variable accommodations app cannot be reached via fixed devices. Therefore, the estimations of the coefficients might not be valid. The Pseudo R-square statistics confirm this notion of having a better fit compared to a null-model. The McFadden R-squared has a value of .091, the Cox & Snell a value of .067 and the Nagelkerke R-squared a value of .126. The BIC decreases from 22055.87 for the null-model to 20543.24 for the final null-model. The hit rate is 87.15%, the Gini-coefficient is .58 and the TDL is 3.3. Table 4 gives an overview of the correctly predicted outcomes. Appendix 5a shows the lift curve for the correctly predicted outcomes.
0 25036 301
1 3342 332
Table 3. Predicted outcomes Model 1
4.4.2 Validation Model 2
The estimation of the logit model indicates that the logit model outperforms a null-model (𝑋: = 3883.8, p = .00). The Pseudo R-square statistics confirm this notion of having a better fit compared to a null-model. The McFadden R-squared has a value of .176, the Cox & Snell a value of .125 and the Nagelkerke R-squared a value of .235. The BIC decreases from
22055.87 for the null-model to 18274.87for the final model. The hit rate is 87.84%, the Gini-coefficient is .64 and the TDL is 3.5. Table 5 gives an overview of the correctly predicted outcomes. Appendix 5b shows the lift curve for the correctly predicted outcomes.
0 1
0 20544 293
1 3368 306
Table 4. Predicted outcomes Model 2
4.5. Logistic regression assumptions
To improve the accuracy of both models, the following assumptions should hold true for the data: the outcome variable is binary or dichotomous, there is a linear relationship between the logit of the outcome and the predictor variables, there are no influential values in the
continuous predictors and there are no high inter-correlations among the predictors
(Kassambara, 2018). For both models, the outcome variable is binary, purchasing versus not purchasing. Thus, the first assumption holds for the data.
As can be seen in Appendix 6a, the scatter plots show that touchpoints accommodations app and pre-rolls are quite linearly associated with the purchase outcome in the logit scale. For the other continuous variables, the linearity assumption does not hold. Therefore, continuous variables have been tested for linear, quadratic, logarithm and exponential specifications. Transforming into logarithm and exponential specifications led to an error, because there are 0’s in the predictor variables. Thus, continuous variables have been specified into linear and quadratic specifications, resulting in a better prediction model compared to only linear
customer-initiated and fixed are quite linearly associated with the purchase outcome in the logit scale. Also in this model, the continuous variables have been tested for linear, quadratic, logarithm and exponential specifications. Transforming into logarithm and exponential specifications led to an error, because there are 0’s in the predictor variables. Thus, continuous variables have been specified into linear and quadratic specifications, resulting in a better prediction model compared to only linear continuous variables. Therefore, it is assumed that the linearity assumption partially holds for each continuous predictor variable
As can be seen in appendix 7a and 7b, there are influential observations in the continuous predictors for model 1 and model 2. To account for influential observations in the continuous predictors, a control variable is included in the model. The variable represents the length of each customer journey. Therefore, the assumption for no influential values in the continuous predictions does hold.
Finally, as can be seen in appendix 3a and 3b, there are high inter-correlations among the predictors. Transforming these variables into percentages does not change the results. However, as the collinear variables are only used as control variables, there is no problem (Allison, 2020). Therefore, the assumption for no high inter-correlations does hold for model 1. The assumption does not hold for model 2, because these variables cannot be used as control variables except for the variable journey length. However, the high correlations between variables will be taken into account when interpreting the results.
5. Conclusion
This study aimed to identify the effectiveness of firm-initiated touchpoints and customer-initiated touchpoints that are subdivided into branded customer-customer-initiated and generic customer-initiated across mobile and fixed devices on a purchase decision. Based on
quantitative analysis of online purchase journey travel data, it can be concluded that generic customer-initiated touchpoints accommodations search, flight tickets search, and
firm-initiated touchpoints e-mail and retargeting result in a higher purchase probability across fixed devices. The same effect holds for generic customer-initiated touchpoint accommodations app and branded customer-initiated touchpoint tour operator/travel agent website focus brand. However, the effect decreases for accommodations app after a turning point of 527 and a turning point of 1655 for tour operator/travel agent website focus brand. The variable generic customer-initiated touchpoint information/comparison results in a higher purchase probability across mobile devices. The same effect holds for variables generic customer-initiated
touchpoint information/comparison website and firm-initiated touchpoint pre-rolls, but the effect decreases after a turning point of 134 for information/comparison website and a turning point of 1 for pre-rolls.
Based on the literature review, I expected that firm-initiated touchpoints and customer-initiated touchpoints positively impacts purchase probability. This assumption holds for customer-initiated touchpoints that are subdivided into branded customer-initiated touchpoints and generic customer-initiated touchpoints. Thus, hypotheses 2a, 4a and 5a are confirmed. For firm-initiated touchpoints, e-mail and retargeting positively impact the purchase probability. However, pre-rolls negatively impact the purchase probability. Therefore, hypothesis 1a is partially confirmed.
According to the analysis, customer-initiated touchpoints are more effective than firm-initiated touchpoints. Therefore, hypothesis 3 is confirmed. This finding is in line with the findings of De Haan et al. (2016). These authors argue that customer-initiated touchpoints seem less intrusive and more relevant to customers. Compared to firm-initiated touchpoints, customer-initiated touchpoints require a level of interest from the customer.
accommodations app, accommodations search, tour operator/travel agent website focus brand and flight tickets search are most effective on fixed devices, while information/comparison website and information/comparison search are most effective on mobile devices. Therefore, hypothesis 2b is partially confirmed.
Since there are more customer-initiated touchpoints than firm-initiated touchpoints in the dataset, a further distinction is made between customer-initiated touchpoints. Generic customer-initiated touchpoints accommodations app, accommodations search and flight tickets search are most effective on fixed devices, while information/comparison website and information/comparison search are most effective on mobile devices. Therefore, hypothesis 5b is partially confirmed. Branded customer-initiated touchpoint tour operator/travel agent website focus brand is most effective on fixed devices. There is no significant evidence that other branded customer-initiated touchpoints are effective on mobile devices. Therefore, hypothesis 4b is not confirmed. According to the literature, fixed devices are more suitable for purchases that do require extensive information search, such as high-involvement products, and mobile devices are more suitable for well-known products (Kaatz et al., 2019). Because online travel bookings are high-involvement services that require extensive information search, it seems logical that fixed devices are preferred.
It might be that people switch between firm-initiated touchpoints and customer-initiated touchpoints during their online customer journey. The switch between firm-initiated touchpoints to customer-initiated touchpoints indicates a shorter purchase time because people limit their choice (Anderl et al., 2016). However, I do not find evidence that the purchase probability is higher if people first encounter firm-initiated touchpoints and then use customer-initiated touchpoints. Therefore, hypothesis 6 is not confirmed.
People can also easily switch between fixed and mobile devices during their online customer journey. The switch from a mobile device to a fixed device indicates a higher purchase probability if the perceived risks are higher, the product price is higher, and the experience with the product and retailer is lower (de Haan et al., 2018). I do find evidence that the interaction between mobile and fixed devices indicate a higher purchase probability. However, the effect decreases after a turning point of 1061. Therefore, hypothesis 7 is confirmed.
require a level of interest of the consumers. These consumers search for information on their own initiative. Therefore, customer-initiated touchpoints seem less intrusive and more relevant to consumers. For firm-initiated touchpoints, firms initiate marketing
communications and determine timing and exposure. Therefore, firm-initiated touchpoints seem more intrusive. However, firm-initiated touchpoints might affect people that have not yet recognized a specific need (Anderl et al., 2016a, Anderl et al., 2016b, Li and Kannan, 2014). A further distinction between customer-initiated touchpoints shows that branded customer-initiated touchpoints are more effective on a final purchase decision compared to generic customer-initiated touchpoints. This is likely because branded customer-initiated touchpoints use the retailer’s name to initiate the contact. Generic customer-initiated touchpoints use unbranded keywords and therefore consumers can identify alternatives (Anderl et al., 2016a). For the interaction between touchpoints and devices, I find that fixed devices are the preferred option for a purchase decision among all touchpoints. This is likely because entering detailed payment information without making mistakes requires a larger and higher-resolution screen. Moreover, when the perceived risk within a product category is higher, the price of a product is higher and the experience with the product and the online retailer is lower, a fixed device is preferred (de Haan et al., 2018). For the interaction between mobile and fixed devices, I find that first encountering mobile devices and then using fixed devices result in a higher purchase probability. According to the literature, this is the case when product category-related perceived risks are higher, the product price is higher, and the experience with the product category and online retailer is lower (Lin et al., 2009). Since the purchase decision in the travel industry is associated with higher risks, higher prices, complex choices and thus high-involvement, this finding is in line with the theory.
Recommendations
For managers, the findings suggest a few recommendations. The study has shown that firm-initiated touchpoints are less effective compared to customer-firm-initiated touchpoints. To increase the effect of firm-initiated touchpoints on a purchase decision, it is therefore
recommended that firms invest in e-mail and retargeting. The effect is highest among people reached via fixed devices. If firms invest in pre-rolls, the effect is highest among people reached via mobile devices. However, it is most effective to use one pre-roll at a time.
highly recommended to invest in an omnichannel strategy as people use mobile and fixed devices interchangeably.
Contributions
The study makes a few contributions to the academic literature. First of all, the study contains a richer data set to evaluate the effect of all forms of touchpoints on a purchase decision, resulting in more insights. Secondly, the study tests whether the effect of touchpoints on a purchase decision strengthen or weaken across different devices, resulting in useful information since only a few studies have done this before. Thirdly, as a managerial contribution, the study clarifies opportunities for budget allocation for travel agencies.
6. Limitations and further research
References
Abhishek, V., Fader, P., & Hosanagar, K. (2012). Media exposure through the funnel: A model of multi-stage attribution. Available at SSRN 2158421.
Abhishek, V., Despotakis, S., & Ravi, R. (2017). Multi-channel attribution: The blind spot of online advertising. Available at SSRN 2959778.
Allison, P. (2012, September 10). When can you safely ignore multicollinearity? Retrieved from https://statisticalhorizons.com/multicollinearity
Anderl, E., Schumann, J. H., & Kunz, W. (2016a). Helping firms reduce complexity in multichannel online data: a new taxonomy-based approach for customer
journeys. Journal of Retailing, 92(2), 185-203.
Anderl, E., Becker, I., Von Wangenheim, F., & Schumann, J. H. (2016b). Mapping the customer journey: Lessons learned from graph-based online attribution
modeling. International Journal of Research in Marketing, 33(3), 457-474. Baxendale, S., Macdonald, E. K., & Wilson, H. N. (2015). The impact of different touch
points on brand consideration. Journal of Retailing, 91(2), 235-253.
Boomtown. (2019, September 15). From touchpoints to journeys: transforming your customer experience with automation. Retrieved from
https://www.goboomtown.com/blog/customer-experience-automation
Dalessandro, B., Perlich, C., Stitelman, O., & Provost, F. (2012, August). Causally motivated attribution for online advertising. In Proceedings of the Sixth International Workshop on Data Mining for Online Advertising and Internet Economy (p. 7). ACM.
De Haan, E., Wiesel, T., & Pauwels, K. (2016). The effectiveness of different forms of online advertising for purchase conversion in a multiple-channel attribution
De Haan, E., Kannan, P. K., Verhoef, P. C., & Wiesel, T. (2018). Device switching in online purchasing: examining the strategic contingencies. Journal of Marketing, 82(5), 1-19. Dehmamy, K. (2019, April 25). Binary choices [PowerPoints]. Retrieved from
https://nestor.rug.nl/bbcswebdav/pid-9374228-dt-content-rid-17666448_2/courses/EBM076A05.2018-2019.2B/CM_Lecture_2_v2.pdf
eMarketer Editors. (2019, February 19). US digital ad spending will surpass traditional in 2019. Retrieved from https://www.emarketer.com/content/us-digital-ad-spending-will-surpass-traditional-in-2019
from https://nestor.rug.nl/bbcswebdav/pid-9009371-dt-content-rid
12797663_2/courses/EBM080A05.2018-2019.1A/MRM1819a%2007%20ModerationMediation.pdf
Hall, G. (2019, May 9). 7 takeaways for measuring the complex customer journey. Retrieved from https://www.visualiq.com/about/blog/7-takeaways-measuring-customer-journey
Hallikainen, H., Alamäki, A., & Laukkanen, T. (2019). Individual preferences of digital touchpoints: A latent class analysis. Journal of Retailing and Consumer Services, 50, 386-393.
Hubert, M., Blut, M., Brock, C., Backhaus, C., & Eberhardt, T. (2017). Acceptance of smartphone-based mobile shopping: Mobile benefits, customer characteristics, perceived risks, and the impact of application context. Psychology &
Marketing, 34(2), 175-194.
Kaatz, C. P., Brock, C., & Blut, M. (2017). Developing a conceptualization of mobile service quality.
Kaatz, C., Brock, C., & Figura, L. (2019). Are you still online or are you already mobile?- Predicting the path to successful conversions across different devices. Journal of Retailing and Consumer Services, 50, 10-21.
Kannan, P. K., Reinartz, W., & Verhoef, P. C. (2016). The path to purchase and attribution modeling: Introduction to special section.
Kassambara, A. (2018). Logistic regression assumptions and diagnostics in R. Retrieved from
http://www.sthda.com/english/articles/36-classification-methods-essentials/148-logistic-regression-assumptions-and-diagnostics-in-r/
Kuehnl, C., Jozic, D., & Homburg, C. (2019). Effective customer journey design: consumers’ conception, measurement, and consequences. Journal of the Academy of Marketing Science, 47(3), 551-568.
Lath, A. (2019, May 11). The importance of online advertising. Retrieved from
https://www.bbntimes.com/en/companies/the-importance-of-online-advertising
Law, R., Leung, K., & Wong, R. (2004). The impact of the Internet on travel
agencies. International journal of contemporary hospitality management, 16(2), 100-107.
Lemon, K. N., & Verhoef, P. C. (2016). Understanding customer experience throughout the customer journey. Journal of marketing, 80(6), 69-96.
environment: An empirical model and a field experiment. Journal of Marketing Research, 51(1), 40-56.
Lin, P. J., Jones, E., & Westwood, S. (2009). Perceived risk and risk-relievers in online travel purchase intentions. Journal of Hospitality Marketing & Management, 18(8), 782-810. Lovett, J. (2013). Attribution methods and models: a marketer’s framework. Retrieved from
http://media.dmnews.com/documents/52/attribution_methods_and_models_12971.pdf
McComas, A. (2017, August 8). The imperfections of last click attribution. Retrieved from
https://blog.thecenterforsalesstrategy.com/the-imperfections-of-last-click-attribution
Mialki, S. (2018, July 13). What is a pre-roll ad and why should marketers use them in campaigns? Retrieved from https://instapage.com/blog/pre-roll-ads
Patterson, L., & Marketing, V. (2009). Managing Touch Point Value: 10 Steps to Improve Customer Engagement. Customer THINK.
Quix, F. W., & Sloot, L. (2019). Retailmarketing. Noordhoff Uitgevers.
Shao, X., & Li, L. (2011). Data-driven multi-touch attribution models. In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 258-264). ACM.
Appendix 1 – Description of touchpoints
Firm FI18 Affiliates Placing a link from a firm’s
business on a partner’s website.
Firm FI19 Banner Digital graphics embedded
in Web content pages.
Firm FI20 Email Ads within an email or
entirely promotional emails.
Firm FI21 Pre-rolls An automatically played
video advertisement before a featured video (Mialki, 2018).
Firm FI22 Retargeting Delivering personalized
banners based on consumer’s browsing history.
Branded CIB10 CIB12
Tour operator/travel agent Website Tour operation/travel agent Search focus brand
Accessing the focus
brand’s website directly by entering its URL or
indirectly by searching for a focus brand’s keyword. Branded CIB7
CIB8 CIB9
Tour operator/travel agent Website Tour operator/travel agent App Tour operator/travel agent Search competitor
Platform to compare prices and product information from competitors. Generic CIG4 CIG5 CIG6 Information/comparison Website Information/comparison App Information/comparison Search
Platform to compare prices and product information from competitors. Generic CIG1 CIG2 CIG3 Accommodations Website Accommodations App Accommodations Search
Platform to compare prices and product information from competitors. Generic CIG13
CIG14 CIG15
Flight tickets Website Flight tickets App Flight tickets Search
Platform to compare prices and product information from competitors.
Generic CIG16 Generic search Searching for a keyword in
a general search engine.
4
Appendix 3a – Correlations Model 1
Appendix 3b – Correlations Model 2
Before transformation:Appendix 4 – Number of occurrences
Touchpoint No occurrence Occurrence Occurrence on Fixed Occurrence on Mobile Purchase on Fixed Purchase on Mobile Accommodations Website 1552826 903588 775561 128027 319944 28654 Accommodations App 2407269 49145 0 49145 0 11448 Accommodations Search 2448971 7443 4194 3249 1579 605 Information / Comparison Website 2262071 194343 158687 35656 51810 5684 Information / Comparison App 2410569 45845 0 45845 0 10089 Information / Comparison Search 2454656 1758 1085 673 450 168 Tour operator / Travel Agent Website Competitor 1736990 719424 607288 112136 222080 26900 Tour operator / Travel Agent App Competitor 2447923 8491 0 8491 0 1852 Tour operator / Travel Agent Search Competitor 2453010 3404 2005 1399 626 384 Tour operator / Travel Agent Website Focus brand 2276598 179816 161505 18311 81232 5219 Tour operator / Travel Agent Search Focus Brand2455880 534 316 218 154 37
Flight tickets Website
2243083 148830 128729 20102 35515 5408
Flight tickets App 2435485 20929 0 20929 0 2924
Flight tickets Search
2448896 7518 4691 2827 1453 619
Affiliates 2454834 1580 1511 69 451 9
Banner 2454605 1809 1778 31 230 4
Email 2453579 2835 2713 122 1334 97
Pre-rolls 2454485 1929 1867 62 226 1
Appendix 8 – R script
# MODEL 1:#clear environment rm(list=ls())
#set working directory
setwd("/Users/ilsemein/Documents/University of Groningen/MSc Marketing/2019-2020/Master thesis/Data Purchase Journey Travel")
#reading the data
df <- read.csv("TravelData.csv", header=TRUE, na.strings = c(""))
#libraries install.packages("ggplot2") library(ggplot2) install.packages("tidyverse") library(tidyverse) install.packages("data.table") library(data.table) install.packages("GGally") library(GGally) install.packages("cowplot") library(cowplot) install.packages("stargazer") library(stargazer) install.packages("DescTools") library(DescTools) install.packages("mfx") library(mfx) install.packages("lmtest") library(lmtest) install.packages("dplyr") library(dplyr)
#counting number of times a touchpoint appear in a journey df$CIG1 <- df$type_touch == 1
df$CIG1 [df$touch1 == "TRUE"] <- 1
df$CIG2 <- df$type_touch == 2 df$CIG2 [df$touch1 == "TRUE"] <- 2
df$CIG4 <- df$type_touch == 4 df$CIG4 [df$touch1 == "TRUE"] <- 4
df$CIG5 <- df$type_touch == 5 df$CIG5 [df$touch1 == "TRUE"] <- 5
df$CIG6 <- df$type_touch == 6 df$CIG6 [df$touch1 == "TRUE"] <- 6
df$CIB7 <- df$type_touch == 7 df$CIB7[df$touch1 == "TRUE"] <- 7
df$CIB8 <- df$type_touch == 8 df$CIB8 [df$touch1 == "TRUE"] <- 8
df$CIB9 <- df$type_touch == 9 df$CIB9 [df$touch1 == "TRUE"] <- 9
df$CIB10 <- df$type_touch == 10 df$CIB10 [df$touch1 == "TRUE"] <- 10
df$CIB12 <- df$type_touch == 12 df$CIB12[df$touch1 == "TRUE"] <- 12
df$CIG13<- df$type_touch == 13
df$CIG13 [df$touch1 == "TRUE"] <- 13
df$CIG14 <- df$type_touch == 14 df$CIG14 [df$touch1 == "TRUE"] <- 14
df$CIG15 <- df$type_touch == 15 df$CIG15 [df$touch1 == "TRUE"] <- 15
df$CIG16 <- df$type_touch == 16 df$CIG16 [df$touch1 == "TRUE"] <- 16
df$FI18 <- df$type_touch == 18 df$FI18 [df$touch1 == "TRUE"] <- 18
df$FI19 <- df$type_touch == 19 df$FI19 [df$touch1 == "TRUE"] <- 19
