Get inspired! Inspirational content in the customer journey
Investigating the effect of inspiration pages on the dropout probability of customers in an e-commerce setting
by
Senne Aarssen 15-06-2020
University of Groningen
Faculty of Economics and Business (FEB) Master Thesis
MSc Marketing Intelligence
First supervisor: Dr. F.T. Beke Second supervisor: Prof. Dr. T.H.A. Bijmolt
2
ABSTRACT
In a world where online shopping becomes more and more the standard, companies are on a mission to develop smooth customer journeys, guiding customers from the very first touchpoint to a purchase. In reality however, the majority of journeys will end in a dropout. These customer dropouts frustrate not only the companies, but also the customers. One of many ways to prevent customers from dropping out, is by using inspiration pages to inspire customers. These pages show products “in use”, for example through recipes, interiors or outfits. Inspiration pages have
a unique place in the beginning of the customer journey. Here, companies can offer consumers new ideas while increasing their consumption intentions. This thesis studies the effect of these inspiration pages on the dropout probability of customers. We dive deeper into this phenomenon by looking at the number of times customers visited inspiration pages and what type of inspiration pages they visited during their journey. We also investigate whether or not small screened devices have a positive moderating effect on the dropout probability.
We found that inspiration pages do indeed decrease the dropout probability. Although the effect is fairly small, we would still advise online retailers to invest in these pages. However, more research should be done on the generalizability of these results to see if they also hold for other industries.
Keywords: inspiration pages, customer journey, dropout probability, touchpoints, e-commerce,
3
PREFACE
In 2014 I started my bachelor in Business Administration at the University of Groningen. Here, I started to grow an interest in marketing, which resulted in active membership of the marketing study association MARUG. Towards the end of my bachelor, I started to learn more about big data and the MSc Marketing Intelligence proved to be the perfect fit for my interests. The combination of marketing and big data was something that I really enjoyed, and the courses were challenging and fun. After doing an internship at a large online retailer, I learned about the customer journey in practice and saw how data could help us understand the way customers move through their purchase funnel.
This together made the fit with this thesis subject complete. I want to thank dr. Frank Beke, my first supervisor for the constructive and useful feedback I received during this project. Even though the corona pandemic happened, and we all had to be a bit more flexible in the way we could do our research, I look back at a very constructive cooperation. Also, I would like to thank my thesis group for the help and feedback they provided during and after the online group meetings we had and my parents for proofreading my thesis. Finally, I would like to thank Prof. Dr. Tammo Bijmolt for his time and effort in being the second supervisor of this project.
I hope you enjoy reading this thesis,
Senne Aarssen
4 TABLE OF CONTENTS 1. INTRODUCTION 5 2. LITERATURE REVIEW 7 2.1 Customer Journey 7 2.2 Inspiring Content 9 2.3 Touchpoints 11 2.4 Customer Dropouts 11 2.5 Device Type 11 2.6 Conceptual Model 11 3. RESEARCH DESIGN 13 3.1 Data Collection 13 3.2 Plan of Analysis 14 3.3 Logistic Regression 15 3.4 Validation 15 4. RESULTS 19 4.1 Data Preparation 19 4.2 Descriptive Statistics 21 4.3 Logistic Regression 23 4.3.1 Multicollinearity 24 4.3.2 Coefficients 25 4.3.3 Marginal Effects 26 4.3.4 Interaction Effect 29 4.4 Model Performance 30 5. DISCUSSION 31 5.1 Inspiration Pages 31
5.2 Number of Inspiration Pages 32
5.3 Type of Inspiration Page 32
5.4 Interaction Effect of Device Category and Inspiration Pages 33
6. CONCLUSIONS & MANAGERIAL IMPLICATIONS 34
6.1 Limitations & Suggestions for Future Research 35
7. REFERENCES 37
5
1. Introduction
The customer journey has been a buzzword for the last couple of years. Not strange, considering that everyone who wants to buy a product or service goes through a customer journey. Instead of just looking at the final step of the funnel, the actual purchase, it is important for companies to look at the complete journey leading up to a conversion (Wooff & Anderson, 2015). The purchase someone makes today could very well be the consequence of multiple touchpoints over a longer span of time.
The explosion of potential customer interaction points makes it almost impossible to guarantee consistency in service and experience across all channels when managing individual touchpoints. Companies should, instead, focus on managing the customer journey as a whole (Maechler et al., 2016). Connecting the actions of users across the various channels and touchpoints a company is active in, and to realise how to leverage the strengths and weaknesses of each of these touchpoints, can give a firm a significant advantage (Bucklin & Sismeiro, 2008).
The global e-commerce conversion rate averages around 2.8% (Monetate, 2018). That means that very few people who visit a web shop eventually make a purchase. A critical challenge for companies with regard to the customer journey, is therefore to make sure the customers stay in their journey (Chau et al., 2006). If a company knows what drives customers to drop out of the customer journey, they could interfere and try to keep these customers aboard. Eventually, this could lead to higher conversion rates, which makes this research very relevant for managerial purposes.
6
These customer dropouts are not only frustrating to online retailers, but also to the customers. After all, they are the ones that have invested time in information search and product comparison (Chau et al., 2006). Understanding which customers have a greater risk of dropping out of the journey is therefore very interesting for online retailers, and in the end, customers alike.
One possible way companies can try to prevent customers from dropping out, is by using inspiration pages. The idea of inspiring content is to show products “in use”. For example
through recipes, interiors or outfits. By visiting these pages, customers will be shown new ideas about consumption possibilities (Bottger et al., 2017). These inspiration pages have the ability to combine offering customers a new idea with increased consumption intentions, which in turn can decrease the dropout probability (Bottger et al., 2017).
Where the paper by Bottger et al. (2017) focused on developing a scale to measure customer inspiration, the goal of this thesis is to go beyond this, and dive deeper into this kind of customer behaviour. Our aim is to see how this might affect customers to either stay or drop out of their customer journey. This leads to the following research question:
RQ: What effect do inspiration pages have on the probability of a customer to drop out of the customer journey?
7
The scientific relevance of this paper will be to contribute to the current literature on attribution within the customer journey (Anderl et al., 2016; de Haan et al., 2016; Kaatz et al., 2019; Kakalejč et al., 2018; Li & Kannan, 2014). Although the base of these studies is the same,
looking at the customer journey as a combination of touchpoints and using this to give credit to the individual touchpoints that have been interacted with, our research will instead focus on what contributes to customers dropping out. Specifically, we will look more into one of these possible touchpoints, the concept of inspiration pages, which has not been studied in this manner before.
After this chapter we will build a theoretical framework by reviewing the current literature on all the relevant subjects. Here we will also introduce and explain our hypotheses and draw our conceptual model. After that we will explain the choices we made regarding the research design. We will then move to our results part, where we describe our results which we will then analyse. We will explain our main findings in the discussion chapter. Lastly, we will move to our conclusions and recommendations based on the earlier chapters.
2. Literature Review
In this chapter, the theoretical background of this paper will be covered. We will elaborate on the constructs of the customer journey, inspirational content, touchpoints, customer dropouts and device types. In this chapter we will also describe and explain our hypotheses. Finally, we will draw our conceptual model.
2.1 Customer Journey
8
2014). They all describe the journey a customer makes with a company from the very first contact moment up to and including a purchase or even until after the purchase.
Anderl et al. (2015) define a customer journey as to include all contact of any individual customer with a retailer through all online marketing channels, prior to a potential purchase decision.
Furthermore, Lemon & Verhoef (2016) conceptualize the customer journey as the experiences of a customer with a firm over time, during the purchase cycle, across multiple touchpoints. They define three stages in this journey, the pre-purchase, purchase and post-purchase phase.
Maechler et al. (2016) state that a customer journey can include anything that happens before, during and after the experience of buying a product or using a service. They state that journeys could be stretching across multiple channels and touchpoints, often lasting days or weeks.
Kaushik (2013) proposed a four-step model of the customer journey, the so called “See, Think, Do, Care” model. This model looks at the group of potential customers, and classifies them into
one of these four stages. A customer in the “see” stage of the customer journey does not have any commercial intent, but is somehow sparked with a company or product. The second stage is the “think” stage, in which someone actually has some commercial intent. The third stage is the “do” stage, in which someone makes a purchase. The final stage, “care”, is everything that
comes after the purchase, in which a company will try to retain a customer. Because of its clear distinctions of the different phases, and the fact that this model is widely used in practice, we will use this conceptualization of the customer journey in our study.
9
previously possible (Anderl et al., 2015). We will use these methods to help understand customer behaviour during the customer journey.
2.2 Inspiring Content
Bottger et al. (2017) found that inspiration can lead to a higher purchase intention. They define customer inspiration as “a customer’s temporary motivational state that facilitates the transition
from the reception of a marketing-induced idea to the intrinsic pursuit of a consumption-related goal”.
Our dataset is limited to the furniture market, and with furniture being an occasionally purchased product, characterised by high levels of consumer involvement (Andreu et al., 2010), we expect inspiring content to have an even bigger effect on dropout behaviour compared to lower involvement product categories. This is in line with the findings of Herhausen et al. (2019), who studied the effect of customer inspiration on customer loyalty. They found that customer inspiration mainly works as a predictor for customer loyalty in high involvement markets.
10
Because of the unique position inspiration pages have in the customer journey, by being able to link the ability to offer customers a new idea with consumption intention (Bottger et al., 2017), we hypothesise that customers that have looked at inspiration pages have a lower chance of dropping out of their customer journey.
H1: Customers that looked at inspiration pages have a smaller chance of dropping out of their customer journey.
Further, we will dive deeper into the behaviour of looking at inspiration pages. First, we will investigate whether the amount of times a customer visits an inspiration page has any impact on the dropout probability. We hypothesise that when visiting multiple inspiration pages, there is a larger chance for a customer to get inspired, which will contribute positively to the purchase intentions and therefore it will lower the dropout probability.
H2: Customers that visit more inspiration pages, both the same page or different pages, will have a smaller chance of dropping out of their customer journey
Next to that, we will investigate if the effect on dropout probability differs for certain types of inspiration pages. We will make a split between inspiration pages that show products in use, for example blogs or Pinterest pages. And inspiration pages that do not, for example the news page or the brochure. As stated by Bottger et al. (2017), a good way to get customers inspired is to show products in use. We therefore hypothesise that customers that look at this type of inspiration pages have a smaller chance of dropping out compared to customers that look at inspiration pages without products in use.
11
2.3 Touchpoints
Each customer journey consists of one or multiple touchpoints. A touchpoint could be defined as direct or indirect contact with a brand (Baxendale et al., 2015). Lemon & Verhoef (2016) identify four categories of touchpoints: brand-owned, partner-owned, customer-owned and social/external/independent. They state that it is important for a firm to identify the touchpoints that occur throughout a customer journey, and pinpoint which touchpoints might trigger customers to drop out of their purchase journey. Inspiration pages are usually brand-owned pages, but could theoretically be any of the other categories. For this study we will focus on brand-owned inspiration pages.
2.4 Customer Dropouts
To overcome the problem that it is hard to determine if someone really dropped out of the customer journey, or is merely idling in between interactions, we chose to classify customers that have not interacted with any touchpoints for a long time as dropouts. Anderl et al. (2016) used a timeframe of thirty days from the last interaction as the threshold to classify journeys as unsuccessful. We will use this same timeframe of thirty days in determining if customers dropped out.
2.5 Device Type
12
positive moderating effect on the relation between inspiration pages and the dropout probability. We hypothesise this because mobile devices have constraints in both user interface and system resources (Wu et al., 2007) and inspiring content often consists of rich media, best suited on a large screen.
We even expect inspiration pages to have a negative effect on the dropout probability, when seen on a mobile device. The fact that the rich media does not fit very well on a small screen might frustrate customers, resulting in a dropout. This leads to the following hypotheses:
H4: Using a small screen device while looking at inspiration pages has a positive moderating effect on the probability of a customer to drop out of their customer journey
We will divide all possible devices into two categories: small screen devices, which include mobile, and large screen devices, which include tablet, laptop and desktop.
2.6 Conceptual Model
13
Figure 1: Conceptual Model 3. Research Design
In this chapter we will elaborate on our data collection methods, variables and type of analysis we will use to answer our research question. We will also discuss how we will validate our results.
3.1 Data Collection
To answer our research question, we will use a large clickstream dataset from an omni-channel furniture retailer. Bucklin & Sismeiro (2008) define clickstream data as “the electronic record of a user’s activity on the internet, which is capable of tracing the exact navigational path a customer takes while navigating the web”.
14
Next to the online clickstream data, we will also have offline sales records available which we can link to the customer ID’s in our clickstream data. We will use this, in combination with
online sales records, to determine if a customer journey ends in a dropout or not.
3.2 Plan of Analysis
In this study we will calculate the probability of dropping out of the customer journey considering if inspiration pages were visited, the number of times inspiration pages were visited, what type of inspiration pages were visited and the type of device a customer used. The dependent variable of this study is a binary variable, with either a customer dropping out of the customer journey or not. As stated earlier, we will use the thirty day timeframe that Anderl et al. (2016) used to determine if a customer really dropped out of their customer journey.
For the number of times an inspiration page is visited we will look at the Client ID. First we check if a customer visited any kind of inspiration page during their journey by looking at their PagePath. If that is the case, we will check how many times they did and what kind of inspiration page(s) they visited. We will make a split between inspiration pages that show products in use and inspiration pages that do not. In our dataset we have 6 different inspiration pages, in Table 1 we show which pages belong to what group.
Page Type of inspiration page
Blog PIU
Folder (brochure) Other Binnenkijken bij (showing
existing interiors)
PIU
Magazine Other
Trends PIU
Woonstijlen (style pages) PIU
Table 1. Types of inspiration pages *PIU = products in use
15
devices with a large screen and devices with a small screen. The classification is shown below in Table 2.
Device type Category
Mobile Small screen size
Desktop, laptop, tablet Large screen size
Table 2: classification of devices
3.3 Logistic Regression
To estimate the effects of our independent variables on the dropout probability, we will apply logistic regression. This type of model fits very well to our study because of the dichotomous nature of our dependent variable. Customer either drop out or they do not. Traditionally these type of research questions were analysed by either OLS regression or linear discriminant function analysis, but these techniques were found to be less than ideal for handling binary dependent variables due to their strict statistical assumptions (Peng et al., 2002). Another option that would fit our data would be the probit model. Research by Gibbons & Hedeker (1997) found that probit and logistic regression models lead to virtually identical conclusions. Nonetheless, the interpretation of parameters is found to be more straightforward for logistic regression. Also, logit models are more often used in marketing research where probit is used more often in economic research. For this study we therefore chose to apply logistic regression. However, in order to test the robustness of our results, we will also estimate a probit model and compare the results of both models.
3.4 Validation
16
(Papies et al., 2017). One way to partly deal with these issues, is by adding control variables. It would be naive to think that dropout behaviour is only caused by the variables that we want to study. In this chapter we will describe the variables we have included into our model in order to keep the omitted variable bias to a minimum and to contribute to the causality of our results and why we chose these specific variables.
Website speed
One driver of drop out behaviour that has already been studied is website speed (Ladhari, 2010; Selvidge, 1999). Selvidge (1999) found, in an experimental setting, that if a customer has to wait longer for a webpage to load, the probability of someone closing the page is higher. This variable is out of scope for this study but will be included into our model as a control variable.
Number of sessions
We will also add some variables based on high involved customers. People who are more involved already have a smaller chance of dropping out of their customer journey (Downling, 2002). Because they are highly involved, they browse more pages and have a higher probability of looking at inspiration pages. The problem here would be that when customers stay in their journey, this might not be due to the fact that this customer looked at inspiration pages, but is actually due to the fact that these customers are more involved. To account for this effect, we start by adding the total number of sessions a customer had during our observation period. As we know, a customer journey typically takes place over multiple sessions and touchpoints (Lemon & Verhoef, 2016). A high number of sessions could be an indicator of high involvement. Therefore, in order to control for this effect, we will add this variable into our model as a control variable.
17
Next, we will add the average session duration of a customer over their complete journey. A low average session duration could be an indicator of higher involved customers, as high involved customers could have more but shorter sessions as they know how to navigate the website. As stated above, this could in turn affect their drop out probability. In order to control for this, we add this control variable into our model.
Channel type
Finally, our last variable for high involved customers is the channel type used by customers to start their session. These marketing channels are the instruments retailers use to reach potential customers on the internet and get them to visit their website (Anderl et al., 2015). If a customer used mainly organic channels, we label them as higher involved. If a customer used mainly paid channels, we label them as regular customers. Table 3 shows the categorization of channels that was used.
Organic channels Paid channels
Email Affiliates
Offline Branded Paid Search
Organic Search Criteo
Referral Display
Social Generic Paid Search
Direct type-in Paid Social
Table 3: Categorization of channels
Customers in the buying phase
18
for this effect, we looked at customers that had products in their basket while looking at inspiration pages. If this was the case, we labelled them as customers in the buying phase and added this as a control variable into our model in order to control for this.
Unfortunately, it is almost impossible to enumerate all possible drivers of a dependent variable, measure them, and include them in the model. So we cannot expect control variables to deal with endogeneity alone (Papies et al., 2017). But as Papies et al. (2017) also state, even when accounting for endogeneity in multiple ways, none of these are perfect. Bound et al. (1993) even state that the cure can be worse than the disease, meaning that in some cases correcting for endogeneity only lessens the causality of the results. Because of this, we will follow the conclusions made by Papies et al. (2017), even though endogeneity can be a serious problem, the best approach is to establish a best practice around it. We hope that by thinking carefully about our possible omitted variables, and adding the right control variables into our model, we keep endogeneity to a minimum. Further, we do recognize that because our data is observational, not being able to fully account for endogeneity will be one of the limitations of this study.
19
discuss if this might bias our results. Furthermore, we will calculate the pseudo R2 and the hit rate of our model. These tests and their results will be further discussed in the results chapter.
4. Results
This chapter will start off with describing how the data was prepared for analysis, followed by providing some descriptive statistics of our dataset. Furthermore, the results of the logistic regression and the corresponding tests will be given.
4.1 Data Preparation
Before the dataset could be properly used for analysis, a lot of data cleaning and preparation needed to take place. First of all, multiple datasets with information on different levels of aggregation needed to be merged. After this was done, a lot of columns that were not needed for this study were deleted so that we ended up with a more compact dataset. Because we want to study behaviour over the customer journey, it is important that we can link the journeys to a client. To be able to do so, we deleted all observations that did not have a client ID recorded so that we ended up with only full journeys.
Furthermore, we looked for other irregularities or deviations in the data. We found that some observations did not have a date recorded. As we need to know the date in order to check if the journey has ended according to the 30-day time frame, we chose to delete these observations. After deleting those observations, we still had a very large dataset to work with.
20
Furthermore, we checked our continuous variables for outliers. The boxplots of the distribution of these variables can be found in Appendix A. There were some high values, but none of these outliers seemed to be caused by errors in the data so we chose to keep those values in because they contain relevant information. For example, the variable of page load time had the most outliers. This was caused because most customers did not experience any long page load times. Only 4067 customers experienced load times above 1 second (1000 milliseconds). The maximum load time was just above 9 minutes. Although this seems very long, we do not have any reason to believe this is an error in the data. We left these outliers in the dataset as this is exactly the effect that we want to control for: customers that have a long page load time who because of this drop out of their customer journey.
Next, we had to change some variables in order to make them ready for analysis. The variable of device type had three levels: mobile, tablet and desktop. We changed this to two levels: small screen size and large screen size. We are interested in the mobile device someone used on the moment of looking at inspiration pages, so we derived the device used at the moment a client had a page path with an inspiration page in it. It is possible that a customer used multiple devices when looking at inspiration pages, but as there are only a handful of customers for which this occurred, we chose to handle this the following way: we looked at the device type that was used the most when looking at inspiration pages and labelled that customers device type accordingly.
21
Variable Consists of
Dummy_dropout Did this clients journey with more than 30 days since the last interaction end
in either no purchase at Goossens (1) or a purchase, either online or offline (0)?
Dummy_sawAnyInspiration Did the customer visit any type of inspiration page during their journey yes (1)
or no (0)?
Dummy_SawPIU Did the customer visit PIU type inspiration page(s) during their journey yes
(1) or no (0)?
Num_Inspirationpage_Visited The total amount of times any type of inspiration page was visited by a
customer during their journey
AvgSessionDuration The average duration of all sessions in minutes of a customer over their
complete journey
NumberSession The total sum of sessions a customer had over their complete journey
Dummy_Newdevice A dummy variable for which device category (small screen or large screen)
was used when visiting inspiration pages
Dummy_basket Did the customer have a product in their basket when looking at inspiration
pages yes (1) or no (0)
PageLoadTimePerClient The total page load time in milliseconds a customer experienced during their
journey
Dummy_Channeltype Did the customer use mainly paid channels (1) or organic channels (0) during
their customer journey?
Table 4: List of variables
4.2 Descriptive Statistics
After the necessary adjustments had been made to the dataset, we calculated some general descriptive statistics of the dataset. By deleting the observations without a client ID, we still had over 7 million observations to work with. These 7 million interactions were made by 463.238 different client ID’s. Out of this group, 428.406 customers did not visit any inspiration pages during their journey and 34.832 customers did.
22
purchase at the company we have data from (either online of offline). When looking at the customer journey at a broader sense, it could be that these customers still had a successful journey but purchased at a different company. This is out of scope for this study. From the total group, 461.143 customer journeys ended in a drop out and only 2096 journeys were successful. In Table 5, the number of visits and number of customers that saw certain inspiration pages are shown.
Page # of visits # of customers
Blog 529.663 18.262 Binnenkijken_bij 207.776 4.926 Folder 421.410 12.001 Magazine 71.009 1.641 Trends 166.234 3.629 Woonstijlen 259.200 6.655 Total 1.655.292 47.114
Table 5: Number of visits and customers per inspiration page
Before we made our model, we also calculated the “no dropout rate” (customers that did not
23 Number of customers Number of customers
that dropped out
No dropout rate
Customers who did not look at inspiration pages
428.406 426.521 0.44%
Customers who looked
at inspiration pages 34.832 34.622 0.60%
Table 6: Model free evidence
4.3 Logistic Regression
To study our hypotheses, we will estimate the following logistic regression equation:
𝑃(𝑑𝑟𝑜𝑝𝑝𝑖𝑛𝑔 𝑜𝑢𝑡)𝑖 =
1 𝜀−𝑧𝑖
Where P(dropping out)i is the probability that customer i drops out of the customer journey,
and Zi is a combination of the three independent variables, the interaction effect and the control
variables:
𝑍𝑖 = 𝛽𝑜+ 𝛽1𝐼𝑛𝑠𝑝𝑖 + 𝛽2𝐼𝑛𝑠𝑝𝑇𝑦𝑝𝑒𝑖 + 𝛽3𝑁𝑢𝑚𝐼𝑛𝑠𝑝𝑖+ 𝛽4𝐷𝑒𝑣𝑖𝑐𝑒𝑖+ 𝛽5(𝐼𝑛𝑠𝑝𝑖𝐷𝑒𝑣𝑖𝑐𝑒𝑖) +
𝛽 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 + 𝜀
Inspi = Did customer i look at inspiration pages during their customer journey InspTypei = Did customer i look at inspiration pages with products in use
NumInspi = The number of inspiration pages that customer i visited during their customer
journey
Devicei = The device type that was most used by customer i when looking at
inspiration pages
control = The control variables page load time, average session duration, number of sessions, channel type and customers in the buying phase.
24
After estimating our model, we will interpret the outcomes. Because we estimate a non-linear model, we have two options to interpret our results: compute the marginal effects or exponentiate the coefficients to get the odds-ratios (Buis, 2010). We will use the marginal effects to interpret our results, because odds-ratios tend to not work very well for models that include interaction effects (Norton et al., 2004).
4.3.1 Multicollinearity
Next, we will check our model for multicollinearity by calculating the VIF scores and looking at a correlation matrix. The VIF scores are shown in Table 7.
Variable VIF sawAnyInspiration 3.123 SawPIU 3.121 Num_Inspirationpage_Visited 1.884 Device 1.478 PageLoadTime 1.001 Channel_type 1.012 AvgSessionDuration 1.114 Number_sessions 1.224 Buying_phase 1.024 sawAnyInspiration:Device 1.488
Table 7: VIF scores
25
4.3.2 Coefficients
First, we will have a look at the coefficients. These are shown in Table 8 below. Because our logistic model follows a CDF distribution, these coefficients only tell us if there is a positive or negative relation. This means that for the significant negative variables an increase leads to a decrease in the probability of a customer dropping out of the customer journey. For the positive variables this is the other way around. The variables ‘saw any inspiration’, ‘number of inspiration pages visited’, ‘device category’ and the interaction variable of ‘saw any inspiration’ and ‘device category’ are all significant. As interaction effects in non-linear models are hard to interpret correctly (Ai & Norton, 2003), we will interpret this effect separately in chapter 4.3.3. Further, the control variables ‘average session duration’, ‘number of sessions’ and ‘customers in the buying phase’ also show statistically significant effects. To interpret the results further,
we will calculate the marginal effects.
Coefficient St. error P-value
Intercept -8.342 0.103 0.000***
Saw any inspiration -0.342 0.137 0.012*
Saw PIU 0.166 0.164 0.310
Number of inspiration pages visited -0.088 0.024 0.000*** Device category 2.933 0.502 0.000*** Sawany inspiration:device category 2.728 0.886 0.003**
Control: Page load time 0.000 0.000 0.318
Control: Avg session duration
0.001 0.000 0.000***
Control: Number of sessions 0.001 0.000 0.002**
Control: Channel type -0.002 0.048 0.964
Control: Customers in the buying phase
6.066 0.102 0.000***
26
4.3.3 Marginal Effects
To answer our hypotheses, we calculated the marginal effects of our explanatory variables, which is the derivative of the probability with respect to an explanatory variable (Liao, 1994). For the interpretation of the marginal effects, it is important to note that these are always calculated for a specific xi. We use the statistical program R-studio, which takes the average
27 Marginal effects
Logit model
Std. Error P-value Marginal effects Probit model (+ significance level)
Saw any inspiration -0.000064 0.00002 0.006** -0.000086**
Saw PIU 0.000038 0.00004 0.349 0.000048 Number of inspiration pages visited -0.000019 0.00001 0.0001*** -0.000025*** Device category 0.000316 0.00003 0.000*** 0.000312*** Saw any inspiration:device category 0.003014 0.0046 0.008** 0.005291*
Control: Page load time
0.000000 0.00000 0.320 0.000000
Control: Avg session duration 0.000001 0.00000 0.000*** 0.0000001*** Control: Number of sessions 0.000002 0.00000 0.003** 0.0000001*** Control: Channel type -0.000001 0.00001 0.963 -0.000002 Control: Customers in the buying phase
0.069408 0.00495 0.000*** 0.073535***
Table 9: Marginal effects signif. codes: 0.001 ‘***’, 0.01 ‘**’, 0.05 ‘*’
Inspiration pages
28
we can say that our H1 holds: Customers that looked at inspiration pages have a smaller chance of dropping out of their customer journey.
Number of inspiration pages
Our explanatory variable of the total number of inspiration pages viewed by a customer, is significant (p < 0.001). The marginal effect of this variable is -0.000019. This effect, however, is harder to interpret because of its continuous nature. The marginal effects for a continuous variable measure the instantaneous rate of change. This may or may not correspond very well to the effect of a one unit change (Williams, 2019). We will still interpret the marginal effects of this variable in this way, but we acknowledge that strictly speaking these effects might differ a bit in numbers. As the marginal effect is -0.000019, this means that the probability of dropping out of the customer journey is ~0,0019% smaller for every extra inspiration page viewed by a customer, given that all other variables are held at a constant. We can therefore say that our H2 holds: customers that visit more inspiration pages, both the same page or different pages, will have a smaller chance of dropping out of their customer journey.
Type of inspiration page
Our variable ‘type of inspiration page’ is not statistically significant (p > 0.1). We can therefore
not say that there is a significant effect of the type of inspiration page on the probability of customers to drop out of their customer journey. This means we have to reject our H3: customers that look at inspiration pages that show products “in use” have a smaller chance of dropping out of their customer journey.
Device category (direct effect)
29
marginal effect of this variable is 0.000316. This means that the probability of dropping out of the customer journey is 0,0316% larger for customers that used small screen devices.
4.3.4 Interaction Effect
In the academic world, there has been a lot of discussion on how to properly interpret interaction effects in terms of marginal effects (Ai & Norton, 2003; Cornelissen & Sonderhof, 2009; Norton et al., 2004). The big problem is that interaction in nonlinear models cannot be interpreted in the same way as in linear models. The marginal effects of a change in both interacted variables is not equal to the marginal effect of changing just the interaction term (Norton et al., 2004). To correctly interpret the interaction effect of device type and saw any inspiration page, we will have a look at the odds-ratios. These are the exponentiated values of the coefficients. The odds-ratio for saw any inspiration is exp(-0.342) = 0.71. For the interaction effect, this value is exp(2.728) = 15.30. This last value can be interpreted as the ratio by which the odds ratio changes. This means that if a customer is using a small screen device while looking at inspiration pages, and, keeping all other variables at a constant, is not dropping out of the customer journey, the odds-ratio of saw any inspiration (0.71) increases by a factor of 15.30. This in turn, means that the negative effect of looking at inspiration pages becomes a positive effect when the customer is using a device with a small screen. This effect is significant (p<0.05). We can therefore accept our H4: Using a small screen device while looking at inspiration pages has a positive moderating effect on the probability of a customer to drop out of their customer journey.
30
Hypothesis Findings
H1: Customers that looked at inspiration pages have a smaller chance of dropping out
of their customer journey.
Accepted
H2: Customers that visit more inspiration pages, both the same page or different pages,
have a smaller chance of dropping out of their customer journey
Accepted
H3: Customers that look at inspiration pages that show products “in use” have a smaller
chance of dropping out of their customer journey
Rejected
H4: Using a small screen device while looking at inspiration pages has a positive
moderating effect on the probability of a customer to drop out of their customer journey
Accepted
Table 10: Overview of our findings
4.4 Model Performance
As a final step, we will evaluate our model performance. To do so, we will conduct a likelihood-ratio test and calculate the pseudo R2, AIC and BIC of our model. The results are shown in Table 11 below.
McFadden R2 AIC BIC
Null model 26803 26814
Logit model 0.490 13702 13823
Table 11: performance measures of our model
First, we performed a LR test to see if our model significantly performs better than a null-model. The results of this test are highly significant (p < 0.001), which means that our model does significantly perform better than a null-model. Next, we calculated the McFadden pseudo-R2. In a logistic regression context, McFadden’s R2 is preferred above other pseudo R2’s due to
31
of McFadden’s R2 tends to be considerably lower than those of the regular R2 index. A pseudo R2 with a value from 0.2-0.4 would represent excellent model fit (McFadden et al., 1979). A pseudo R2 of 0.490 can therefore be seen as a very good score representing good model fit. Further, the AIC and BIC of our model are a lot lower than those of the null model, which confirms that our logistic model is performing better than the null model.
5. Discussion
The objective of this thesis was to study how inspiration pages affect the probability to drop out of the customer journey. In this chapter we will elaborate on our results and, if our hypotheses are not supported, we will provide possible explanations for these results. The output of Tables 9 & 10 is used to discuss the results.
5.1 Inspiration Pages
32
For our dataset, we used data from a large furniture retailer. Even though the effect is small, it could still be worth investing in inspiration pages in this case as the average basket size in euros for furniture is fairly high compared to other industries. It would be interesting to see how inspiration pages affect the dropout probability for other product categories. We will discuss this further in chapter 6.1.
5.2 Number of Inspiration Pages
Another objective of this thesis was to study if the number of inspiration pages customers visited during their journey had an effect on the dropout probability. As can be seen in table 9, we found that there is a significant negative effect. Every extra inspiration page that a customer visited, decreased the probability of dropping out of the customer journey with ~0.0019%.
Customers who visit more inspiration pages have a larger chance to actually get inspired. This can in turn lead to a decrease in the dropout probability. This is in line with the findings of Bottger et al. (2017), who found that inspiration can have a positive effect on consumption intentions. It could be argued that customers who visit more pages in general, would already have a smaller change of dropping out of the journey. By controlling for high involved customers and number of sessions, we hope to have dealt with this problem.
To get customers to visit more inspiration pages, it is important that a firm invests in multiple high-quality inspiration pages. The inspiration pages should also be suitable for all screen sizes, we will come back to this in section 5.4. Some ways of increasing the overall visit frequency of a website might include actively engaging with your customers on social media (Rishika et al., 2013) and establishing brand loyalty (Thorbjornsen & Supphellen, 2004).
5.3 Type of Inspiration Page
33
et al., 2017). This could in turn decrease the dropout probability. We hypothesised that inspiration pages that show products in use will have a larger negative effect on the dropout probability compared to the pages that do not. However, we did not find a significant effect.
It could be that the split we made between pages that show products in use and those that do not were not distinctive enough. The brochure and magazine, the pages we labelled as not showing products in use, still contain certain elements of products in use. This might be the reason we did not find a difference between the two categories.
As we do not have any information about the underlying effect of customer inspiration, we cannot conclude anything about this as of yet. Future research could focus on defining a more extensive conceptualization of inspiration pages. This way, perhaps a better split between page types could be made. We will discuss this further in chapter 6.1.
5.4 Interaction Effect of Device Category & Inspiration Pages
The final objective of this thesis, was to study the possible positive moderating effect of using a small screened device (versus a large screen) while visiting inspiration pages on the probability of a customer to drop out of their customer journey. We found a significant moderating effect between these variables. We found that the negative effect of looking at inspiration pages on the dropout probability, becomes a positive effect when the customer looked at these pages while using a small screened device. These findings contradict the findings of De Haan et al. (2015), who found that mobile devices are more qualified for search purposes (the think phase) and less for purchasing purposes (the do phase).
34
companies might have made their pages better compatible with small screened devices in comparison to others.
6. Conclusions & Managerial Implications
Coming back to our research question: “what effect do inspiration pages have on the probability of a customer to drop out of the customer journey?”, we can now say that there is a significant
negative effect of looking at inspiration pages on dropout probability. Even though the effect we found is very small, inspiration pages could still be of good use to companies. Customer dropouts do not only frustrate the online retailers, but also the customers who invested time in information search and product comparison (Chau et al., 2006). Inspiration pages could therefore help online retailers in their pursuit of a smooth customer journey. In the end, to have less frustrated customers is a win-win situation for both customer and company.
Guiding customers towards more of these inspiration pages could help in keeping them in their journey, as each extra inspiration page a customer visits, decreases their dropout probability. Companies should therefore make these pages easy to navigate to and invest in high quality content, in order to increase traffic and maximize the use of inspiration pages.
35
6.1 Limitations and Suggestions for Future Research
This section will provide some limitations of this research, and give a couple of recommendations for future research on this subject.
First, this study is based on data from one company in one specific industry: furniture. As this is a market that is characterised by high levels of customer involvement (Andreu et al., 2010), the results could differ a lot with industries that have lower levels of customer involvement. This could mean that the results are somewhat less generalizable with other industries. Next to that, the furniture industry is lagging behind in the overall explosive growth of e-commerce. Online sales account for only about 4% of the total sales (Oh et al., 2008). It could be the case that for some other industries, the effect of inspiration pages, therefore, differs from our findings. We would recommend future research to investigate the effect of inspiration pages over a broader range of industries, with different levels of involvement and online sales volumes.
Building on this, because we have data from only one company, we also had to deal with only a limited amount of inspiration pages. Our dataset consists of 6 different types of inspiration pages, but there are way more types of inspiration pages a company could use. An example of another type of inspiration page is one that uses virtual reality. For example, VR-implemented websites offer consumers the possibility to select furniture to set up a virtual living room (Oh et al., 2008). Augmented Reality could even extend this by projecting the furniture into a customer’s own living room. Future research could include companies that use other types of
inspiration pages to see if the effect on the dropout probability differs.
36
anywhere else. We focus on customer journeys at one company, but from a customer perspective, a journey could stretch over multiple companies.
Also, we have to conclude that because our dataset consists of observational data, we are not able to fully account for endogeneity (Papies et al., 2017). We do not know the underlying rationale of certain actions. To really understand the behaviour of customers, more qualitative research is needed. We would recommend future research to follow a number of customers on their complete customer journey. This way they can see if and why they visit inspiration pages, and what effect this has on their probability to drop out of their journey. This could also help with the conceptualization of customer inspiration. As of now, there is little to no literature on what can be seen as inspiration pages. Future research could help by defining what pages cause customers to get inspired. This could in turn help with making a better split in inspiration page types, in order to study this further.
Another shortcoming of our dataset was the lack of demographics. If we had more information about the customers, we might have been able to segment our dataset to see if certain customers react differently to inspiration pages in comparison with other groups. This could be interesting for future research to explore.
37
7. References
Ai, C., & Norton, E. C. (2003). Interaction terms in logit and probit models. Economics
Letters, 80(1), 123–129. https://doi.org/10.1016/S0165-1765(03)00032-6
Anderl, E., Becker, I., von Wangenheim, F., & Schumann, J. H. (2016). Mapping the customer journey: Lessons learned from graph-based online attribution modeling.
International Journal of Research in Marketing, 33(3), 457–474.
https://doi.org/10.1016/j.ijresmar.2016.03.001
Anderl, E., Schumann, J. H., & Kunz, W. (2015). Helping Firms Reduce Complexity in Multichannel Online Data: A New Taxonomy-Based Approach for Customer Journeys.
Journal of Retailing, 92(2), 185–203. https://doi.org/10.1016/j.jretai.2015.10.001
Andreu, L., Sánchez, I., & Mele, C. (2010). Value co-creation among retailers and consumers: New insights into the furniture market. Journal of Retailing and Consumer Services,
17(4), 241–250. https://doi.org/10.1016/j.jretconser.2010.02.001
Baxendale, S., Macdonald, E. K., & Wilson, H. N. (2015). The Impact of Different Touchpoints on Brand Consideration. Journal of Retailing, 91(2), 235–253. https://doi.org/10.1016/j.jretai.2014.12.008
Bottger, T., Rudolph, T., Evanschitzky, H., & Pfrang, T. (2017). Customer inspiration: Conceptualization, scale development, and validation. Journal of Marketing, 81(6), 116– 131. https://doi.org/10.1509/jm.15.0007
Bound, J., Jaeger, D., & Baker, R. (1993). The Cure Can Be Worse than the Disease: A Cautionary Tale Regarding Instrumental Variables. National Bureau of Economic
Research. https://doi.org/10.3386/t0137
Bucklin, R. E., & Sismeiro, C. (2008). Click Here for Internet Insight: Advances in
Clickstream Data Analysis in Marketing. Journal of Interactive Marketing, 23(1), 35–48. https://doi.org/doi:10.1016/j.intmar.2008.10.004
Buis, M. (2010). Interpretation of interactions in nonlinear models. Stata Journal, 10(2), 305– 308. http://ideas.repec.org/a/tsj/stataj/v10y2010i2p305-308.html
Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics Using Stata. Stata Press, 23(4), 2–6. www.stata.com
Chau, P. Y. K., Hu, P. J. H., Lee, B. L. P., & Au, A. K. K. (2006). Examining customers’ trust in online vendors and their dropout decisions: An empirical study. Electronic Commerce
Research and Applications, 6(2), 171–182. https://doi.org/10.1016/j.elerap.2006.11.008
Cornelissen, T., & Sonderhof, K. (2009). Partial effects in probit and logit models with a triple dummy-variable interaction term. Stata Journal, 9(4), 571–583.
https://doi.org/10.1177/1536867X0900900404
Criteo. (2018). Global Commerce Review EMEA, Q2 2018. https://www.criteo.com/wp-content/uploads/2018/09/18_GCR_Q2_Report_EMEA.pdf
Davis, S., & Toney, L. (2020). How Coronavirus Is Impacting Ecommerce. ROI Revolution. https://www.roirevolution.com/blog/2020/05/coronavirus-and-ecommerce/
38
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.
https://doi.org/10.1016/j.ijresmar.2015.12.001
Downling, G. (2002). Customer Relationship Management: In B2C Markets, Often Less Is More. California Management Review, 44(3).
Gibbons, R. D., & Hedeker, D. (1997). Random Effects Probit and Logistic Regression Models for Three-Level Data. Biometrics, 53(4), 1527. https://doi.org/10.2307/2533520 Herhausen, D., Kleinlercher, K., Verhoef, P. C., Emrich, O., & Rudolph, T. (2019). Loyalty
Formation for Different Customer Journey Segments. Journal of Retailing, 95(3), 9–29. https://doi.org/10.1016/j.jretai.2019.05.001
Hoban, P. R., & Bucklin, R. E. (2015). Effects of internet display advertising in the purchase funnel: Model-based insights from a randomized field experiment. Journal of Marketing
Research, 52(3), 375–393. https://doi.org/10.1509/jmr.13.0277
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(April), 10–21.
https://doi.org/10.1016/j.jretconser.2019.04.005
Kakalejč, L., Bucko, J., Resende, P., & Ferencova, M. (2018). Multichannel Marketing Attribution Using Markov Chains. Journal of Applied Management and Investments,
7(1), 49–60.
Kaushik, A. (2013). See-Think-Do: A Content, Marketing, Measurement Business Framework. Occam’s Razon By Avinash Kaushik Blog, 1–21.
https://www.kaushik.net/avinash/see-think-do-content-marketing-measurement-business-
framework/%0Ahttp://www.kaushik.net/avinash/see-think-do-content-marketing-measurement-business-framework/
Kleinlercher, K., Emrich, O., Herhausen, D., Verhoef, P. C., & Rudolph, T. (2018). Websites as Information Hubs: How Informational Channel Integration and Shopping Benefit Density Interact in Steering Customers to the Physical Store. Journal of the Association
for Consumer Research, 3(3), 330–342. https://doi.org/10.1086/698415
Ladhari, R. (2010). Developing e-service quality scales: A literature review. Journal of
Retailing and Consumer Services, 17(6), 464–477.
https://doi.org/10.1016/j.jretconser.2010.06.003
Leeflang, P. S. H., Wieringa, J. E., Bijmolt, T. H. ., & Pauwels, K. . (2016). Modeling
Markets. Springer.
Lemon, K. N., & Verhoef, P. C. (2016). Understanding customer experience throughout the customer journey. Journal of Marketing, 80(6), 69–96.
https://doi.org/10.1509/jm.15.0420
Li, H., & Kannan, P. K. (2014). Attributing Conversions in a Multichannel. Journal of
Marketing Research, LI(February), 40–56. https://doi.org/10.1509/jmr.13.0050
Liao, T. (1994). Interpreting Probability Models: Logit, Probit, and Other Generalized
39
Maechler, N., Neher, K., & Park, R. (2016). From touchpoints to journeys: Seeing the world as customers do. McKinsey Digital, March, 1–14.
Mason, C. H., & Perreault, W. D. (1991). Collinearity, Power, and Interpretation of Multiple Regression Analysis. Journal of Marketing Research, 28(3), 268.
https://doi.org/10.2307/3172863
McFadden, D., Hensher, D., & Stopher, P. (1979). Behavioural travel modelling (3rd ed.). Menard, S. (2000). Coefficients of determination for multiple logistic regression analysis.
American Statistician, 54(1), 17–24. https://doi.org/10.1080/00031305.2000.10474502
Monetate. (2018, December 6). The data types that drive the most relevance: EQ3 2018 out
now.
https://monetate.com/blog/the-data-types-that-drive-the-most-relevance-eq3-2018-out-now/
Norton, E. C., Wang, H., & Ai, C. (2004). Computing Interaction Effects and Standard Errors in Logit and Probit Models. The Stata Journal: Promoting Communications on Statistics
and Stata, 4(2), 154–167. https://doi.org/10.1177/1536867x0400400206
O’Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors.
Quality and Quantity, 41(5), 673–690. https://doi.org/10.1007/s11135-006-9018-6
Oh, H., Yoon, S. Y., & Shyu, C. R. (2008). How can Virtual Reality reshape furniture retailing? Clothing and Textiles Research Journal, 26(2), 143–163.
https://doi.org/10.1177/0887302X08314789
Pantano, E., & Priporas, C. V. (2016). The effect of mobile retailing on consumers’ purchasing experiences: A dynamic perspective. Computers in Human Behavior, 61, 548–555. https://doi.org/10.1016/j.chb.2016.03.071
Papies, D., van Heerde, H., & Ebbes, P. (2017). Addressing Endogeneity in Marketing Models. In Advanced Methods for Modeling Markets (pp. 581–624).
Peng, C. Y. J., Lee, K. L., & Ingersoll, G. M. (2002). An introduction to logistic regression analysis and reporting. Journal of Educational Research, 96(1), 3–14.
https://doi.org/10.1080/00220670209598786
Raymond, J. (2001). No more Shoppus Interruptus. American Demographics, 23(5), 39–40. Rishika, R., Kumar, A., Janakiraman, R., & Bezawada, R. (2013). The effect of customers’
social media participation on customer visit frequency and profitability: An empirical investigation. Information Systems Research, 24(1), 108–127.
https://doi.org/10.1287/isre.1120.0460
Ruiz-Mafe, C., Sanz-Blas, S., Hernandez-Ortega, B., & Brethouwer, M. (2013). Key drivers of consumer purchase of airline tickets: A cross-cultural analysis. Journal of Air
Transport Management, 27, 11–14. https://doi.org/10.1016/j.jairtraman.2012.10.010
Selvidge, P. (1999). How Long is Too Long to Wait for a Website to Load? Usability News,
1(2), 1999–2001.
Srinivasan, S., Rutz, O. J., & Pauwels, K. (2016). Paths to and off purchase: quantifying the impact of traditional marketing and online consumer activity. Journal of the Academy of
Marketing Science, 44(4), 440–453. https://doi.org/10.1007/s11747-015-0431-z
40
Usage. Brand Management, 11(3), 199–208. https://doi.org/10.1108/eb025586 Williams, R. (2019). Marginal Effects for Continuous Variables. 1–12.
Wooff, D. A., & Anderson, J. M. (2015). Time-weighted multi-touch attribution and channel relevance in the customer journey to online purchase. Journal of Statistical Theory and
Practice, 9(2), 227–249. https://doi.org/10.1080/15598608.2013.862753
Wu, C. I., Teng, C. M. J., Chen, Y. C., Lin, T. Y., Chu, H. H., & Hsu, J. Y. J. (2007). Point-of-capture archiving and editing of personal experiences from a mobile device. Personal
41
8. Appendices
Appendix A: Boxplots of our continuous variables
