The Effect of Online Marketing Channels on The Firm Performance
of Business-to-Business Firms & Business-to-Customer Firms
- Measured with Different Attribution Models
Author: Supervisor:
B. Hartsink Dr.H.Gungor
Executive Programme in Management Studies – Digital Business
Student number: 10261745
Amsterdam Business School
2
Table of contents
Abtract ...4 Introduction ... 5-6 Literature Review ... 6-14
Online Marketing Channels ... 6-7 Classification of OMC ... 7-10 Attribution ... 10-12 Firm focus ... 12-13 Hypothesis ... 13-14 Methodology ... 14-18 Data ... 14-15 Attribution models ... 15-18 Model fit ... 18 Results ... 18-25 Paths to conversion ... 19 Results of OMC on firm performance (revenue) ... 19-22 Results of OMC on firm performance (conversion) ... 22-25 Conclusion ... 25-26 Limitations and Future research ... 26-27 References ... 28-30 Appendix ... 31-34
3
Statement of Originality
This document is written by Bob Hartsink who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
4
Abstract
Customers are browsing the Internet with different goals in mind. Within their „customer journey‟ they get in touch with multiple channels. Advertisers face the challenge to determine the contribution of each channel to the performance of the firm. The firm focus (B2B or B2C) is taken into account while analyzing the contribution of the various channels. Heuristic attribution models and data-driven attribution models have been introduced within the Business-to-Customer domain in academia and used in the practice. In this research the models are applied to the datasets of the different types of firms, focusing on the „Markov model‟. Nine different channels are analysed and classified into two channel categories. The results show differences in the effectiveness of the online marking channels within the two different firms. The channels in which the contact is initiated by the firm (FICs) have a higher contribution to the conversions and revenue of the B2B firm than the customer-initiated-channels (CICs). For the B2C firm the CICs have a higher contribution to the performance than the FICs. This research indicates the difference in effectiveness of the channels within the different types of firms. Marketers can use these insights to develop their online marketing strategy.
5
Introduction
Advertisers‟ spend on online advertising increased by 13%, from €843 million in 2013 to €954 million in the first half year in 2017 within The Netherlands. The largest growth rate (18%) was in the Search category which is the largest category of online advertising with a 45% market share and a total of €438 million spend revenue (IAB, 2017). The number of contact points between customer and firms have increased through the emergence of new communication channels during the last two decades (Lemon & Verhoef, 2016). Many consumers visit company websites multiple times before they convert, using various channels before the conversion. These previous visits may influence how the visitor returns to the website, such that the customer revisits the website through the same channel (carryover effects) or through different channels (spill-over effects). As the number of online channels and the complexity of the customer journey increase, measuring the contribution to a conversion (e.g. purchase, subscription) of each channel becomes more interesting (Anderl, Becker, von Wangenheim & Schumann, 2016). Attribution models can be used by marketers to have better insights in how online marketing channels contribute to conversions (Zhao, Mahboobi & Bagheri, 2017). Attribution is commonly considered as a measurement issue, but this method will have a heavy impact on the ability to enhance the effectiveness of campaigns. Attribution models use bias measurements that do not take the strategic behaviour of publishers into consideration (Berman, 2015). Consumers
increasingly use online media to find information. Therefore firms are motivated to spend more of their online marketing budget online. Justifying the online marketing spend is simplified, especially in comparison to offline media spend, by the introduction of online metrics such as click-through ratio (CTR) and cost per acquisition (CPA) (Kireyev, Pauwels & Gupta, 2016). Customer-firm interactions are the starting point of the relationship and present a source of competitive advantage from a
marketing perspective. Researchers and managers need to have an understanding of the effect of factors such as the channel and frequency on interactions and relationships as a whole
(Pérez-Aróstegui, Bustinza-Sánchez & Barrales-Molina, 2015). Allocating the budgets and resources within a firm across (multi)channels and communication activities becomes an important question (Neslin &
6 Shankar, 2009). The net profit can be increased by up to fourteen times by shifting marketing budgets among the channels (Wiesel, Pauwels & Arts, 2011).
Literature review
Marketing contacts are often initiated by customers in the online world, but traditionally marketing messages were pushed to the consumers by the firm. For a long time the firms pushed ads, promotions, and offers to customers. Nowadays firms still push marketing messages, but managing the non-push initiatives are becoming more important (Shankar & Malthouse, 2007). The contact with customers that is initiated by the firm are increasingly unwanted (Blattberg, Kim & Neslin, 2008). Firms must now work with search engines (e.g. Google) to attract the browsing customer who have indicated interest in the firm with the search words they use to their website. The internet empowers the consumers to interact with firms and other customers. The opinion of customers, reviews and discussions about products and services are now present online (Shankar & Malthouse, 2007). Previous research and recent analysis has shown that advertising in the Search category has a lot of potential and has become an important marketing channel for firms (Ghose & Yang, 2006; IAB, 2017).
Online Marketing Channels (OMC) Overview
European marketers report that they use, on average, seven online marketing channels in parallel (Anderl, Schumann & Kunz, 2016). In order to clarify the online marketing channels that are available for firms who are operating in the online environment, each channel will be described briefly. Direct is the channel in which the customer visits a website by directly typing-in the URL of the firm. This channel is related to Search, in which customers use a general search engine with search keywords in order to let the search engine come up with results. These results can be presented in two different types, organic results are ranked by the algorithm of the search engine while sponsored results are paid for by the firms. The process of improving the visibility of a website or a webpage in the organic results of search engines is „Search Engine Optimization‟ (SEO) (Gupta, Miglani & Sundriyal, 2013). Firms can bid on specific keywords within search engines for a sponsored search result, which is
7 shown on top of the organic search results on a search result page. The advertiser pays the assigned price for the visitor who clicks on the link, the paid search result, to their website (Ghose & Yang, 2009). The Direct type-in of an URL and Search are related, as customers use the URL as a search string in a search engine. According to Lee and Sanderson (2010) there are differences in URL queries and non-URL queries. Customers click more frequently on higher ranked results of URL queries in comparison to non-URL queries. In terms of user intent, URL queries were more navigational (nearly 86%) than the non-URL search (58% were informational). Using URL queries are common in Search, mostly for navigational reasons, but a notable number of users search this way with a different intent in mind (Lee & Sanderson, 2010). Display are banner advertisements which are embedded in the content of websites. Customers can click on such ads to be redirected to the website of the advertiser. A special form of display advertisement is „retargeting‟. Personalized banners will be delivered to consumers based on their browsing history. These ads can be a generic message of the firm or
dynamic, a customized based on the pages and products a consumers visited. Using other platforms to promote links to their websites is known as Affiliate (Referral, Lead Generation) marketing. These platforms earns commissions when people on their platform click on these links and convert (registration, purchase). Firms can also reach their customers with promotions or information by sending Emails. The consent of the customer is needed to send these emails. The Social Network platforms of a firm can be used as marketing channels. These social platforms are managed by the companies themselves. These services are based on the active participation of users. Social Network platforms include various types of very influential and visited services such as social networks (Facebook), systems of content aggregation (news.google.com), multimedia social services
(YouTube), social bookmarking systems (delicious.com) and many others (Petrov, Zubas & Milojevic, 2015).
Classification of OMC
The customer experience is the journey of the customer across multiple channels over time with a firm during the purchase cycle. Customers might interact with different channels at each stage of the journey. Depending on the nature of the product/service or the customer‟s own journey, the strength or
8 importance of each channel may differ at each stage (Lemon & Verhoef, 2016). The online marketing channels that organizations are using can be divided into two forms of channels. Customer-initiated-channels (CICs) are Customer-initiated-channels in which consumers seek out information through their own initiative, for example when a customer is using a search engine and clicks on an organic search result (Organic Search). Firm-initiated-channels (FICs) are initiated by the firm to have contact with (potential) customers. A form of a FIC is when the firm is sending out an Email to their newsletter subscribers (Anderl et al. 2016).
Figure 1 Classification of OMC
A central idea in marketing is that customers follow a few stages before completing a purchase, including need recognition, information search, evaluation of alternatives and choice. FICs can reach potential consumers, while CICs assist customers who have a need for the product category. This is evidenced by the search behaviour (search engines), evaluation of alternatives (price comparisons) or by specific behaviour on the website (retargeting). CICs are more effective in generating revenue, as a result of assisting the customer in the online purchase funnel (De Haan, Wiesel & Pauwels, 2016). When a potential customer visits the website for the first time through a FIC and returns via a CIC, his or her choice may be narrowed down by actively searching for new information. This switch indicates a progress in purchase decision and thus an increase in purchase propensity (Anderl et al. 2016). In the literature alternative approaches are developed. According to Klapdor (2013) channels should be
Online Marketing Channels Direct CIC Social Network CIC Organic Search CIC Paid Search FIC Affiliate FIC Display FIC Email FIC
9 classified on the browsing goal of a user to predict the purchase propensity. The browsing goal can either be informational or navigational. When the user wants to learn something by visiting or reading web pages, the search goal is informative. The search goal is navigational when the user wants to visit a specific website. This classification is also the case for other online marketing channels. A direct type in of the website and the emails from a firm are classified as navigational, whereas the display ads, price comparison websites, affiliates and retargeting channels are likely to be used as acquisition for information (Klapdor, 2013).
Kannan and Li (2014) introduced a methodology, the Bayesian model, which attributes the
incremental value of each online marking channel in an online environment. Within this framework the authors take into account the carryover and spill-over effects across online marketing channels that are used by the customer to visit the website of a firm. The attribution and estimation of the effects of the customer – and firm-initiated-channels is done by using data on an individual level, such as the visits, touches and purchases, through these channels over time. It helps to allocate the credit for conversion among the used channels. This is important for firms in order to accurately determine the effect of each channel to the conversions on the website. It will help the managers of a firm to efficiently allocate marketing budgets among the online marketing channels. Next to that it helps to create targeting strategies. Kannan and Li (2014) find significant spill-over effect of FICs to CICs at the stage of visit and purchase. Interventions from the firm affect visits through other channels in the short run, which implies that managers must take a more inclusive and macro view of the return on investments in the interactions initiated by the firm. The attribution of the channels with a last-click metric significantly underestimates the contribution of the FIC‟s to conversions. The last channel a visitor comes from is often a CIC, but this initiative can be the result of the use of a FIC by the firm. The real contribution of organic search on conversions is much lower than the last-click attribution presents. This channel should be used by the customer, who initially visited the website through a different channel, as a navigational tool to visit the website for completing the purchase. The same applies for the paid search channel and direct type-in. The model of Kannan and Li (2014) can also form the basis for the determination of the acquisition costs through each channel in each period. The results of their research show that Email and Display ads are effective in the short run, but when using
10 these channels as retargeting strategy it is not always the best strategy. Using Email and Display ads based on the behaviour of a customer, visits on the website and earlier clicks on an Email or Display ad, can decrease the purchase probability of customers in some cases. Their model can be used to determine in what way these channels can contribute to more conversions (Kannan & Li, 2014). De Haan et al. (2016) state that with FICs 53,3% of the elasticities are significant, while CICs have 70% significant elasticities. This is in line with Kannan and Li (2014) who show that CICs are more effective than FICs. In their dataset they found that CICs are 26.7 times more effective than FICs. Prospective customers are likely to be more attentive to information that is directly relevant to what they seek. Firm-initiated advertisements reach the customers often at the wrong time and with a sub-optimal message (De Haan et al. 2016).
Attribution
While companies use multiple online marketing channels in parallel, advertisers face the fact that the advertisements interact in a non-trivial manner to influence consumers. It becomes a team effort of the different publishers in getting consumers to respond to ads. In that case publishers can make use of the efforts of the other publishers, creating moral hazards (Holmstrom, 1982). Interpreting the influence of multiple advertisements of these publishers to the user‟s decision process is called the „attribution problem‟ of advertisers (Shao & Li, 2011, Anderson & Cheng, 2017). Advertisers often have
performance based contracts for the online marketing campaigns. In 2013 performance based pricing took 65% of the online advertisement industry revenue compared to 41% in 2005. Publishers promise a share of the observed output of the campaign, „Cost-per-Action‟ (CPA) e.g. clicks, visits or
conversions. Attribution is needed to analyse the results of the channels, in order to allocate compensation (Berman, 2015). Several rule-based attribution models have been proposed for the multi-touch attribution, in order to distribute the credit to all related advertisements based on their corresponding contributions (Zhang, Wei & Ren, 2014). The last-click attribution model is a popular rule-based method to analyse the credit to a conversion. This model allocates the credit of a conversion to the last shown/clicked ad, the last touch with the consumer, prior to the conversion. The last-click method can be compared to the traditional sales compensation schemes, in which the salesperson who
11 closes the deal receives the commission (Berman, 2015). The adverse of rule-based methods is they may not fit the reality well, the rules of such methods are derived from some simple intuition. As a result of enhanced capability of tracking the advertisement placement and the interaction of the customer with the ads, data driven multi-channel attribution models were proposed recently. These models attempt to infer the contribution from real user interaction data (Zhang et al. 2014). In order to study the conversion attribution problem, Shao & Li (2011) used two data driven approaches. The first one, a bagged logistic regression model, demonstrated a more stable estimate of individual advertising channel contributions with a similar classification accuracy as a usual logistic regression. They also proposed a probabilistic model based on a combination of first and second-order conditional probabilities to directly quantify the attribution of different advertising channels. These models resulted in several important insights for the advertising team. The probabilistic model was
implemented in the production advertising platform of the company after their research (Shao & Li, 2011). In the context of search engine marketing, an attribution framework based on Markovian graph-based data mining techniques is applied in the research of Anderl et al. (2016). The authors modelled multichannel customer journeys on the individual level as first- and higher-order Markov graphs, using the property removal effect to determine the contribution of online channels. By using such a model, carryover and spill-over effects both within and across channel categories are identified. These effects can help online advertisers to develop integrated online marketing strategies (Anderl et al. 2016). In the research of Kireyev et al. (2016), a multivariate times-series model is used, in order to estimate the effectiveness of Display ads on Search ads. This model is designed for aggregate-level time-series data (daily advertising expenditures and revenues). The result of their research showed that the „Return-on-Investment‟ (ROI) estimates are 10% (Display) and 38% (Search) higher than those obtained by standard metrics. The company may have under-invested in these channels, by acting on the results of the standard metrics. After accounting for attribution and dynamics, their results show how companies can shift the online marketing budget optimally (Kireyev et al. 2016). The research of De Haan et al. (2016) differs from the others, because they analysed the data using a vector
autoregressive (VAR) model. The sales impact of allocation shifts among online and offline
12 attribution, the budget allocation on online advertising proposed by the VAR model yields a 21% revenue increase. In table 1 an overview of used data-driven attribution models within the current literature is presented.
Data-Driven Models Authors
Logistic regression model Shao & Li, 2011
Game Theory-based models Dalessandro, Perlich, Stitelman & Provost, 2012
Bayesian models Kannan & Li, 2014
Mutually exciting point process model Xu, Duan & Whinston, 2014
Survival Models Zhang et al. 2014
Markov models Anderl. et al. 2016
Vector Autoregression (VAR) models De Haan et al. 2016 Multivariate time-series models Kireyev et al. 2016
Table 1 Overview data driven attribution models
Firm focus
Business-to-business (B2B) firms are different from business-to-customer (B2C) firms in many ways. Four unique characteristics for B2B firms are defined, in comparison to B2C firms. B2B firms have a limited number of business customers with large sales orders, higher risks in business transactions, higher switching costs for business customers and greater involvement and significant knowledge on the product and industry of business customers. The B2C consumers have the tendency to shop around (Wang & Xu, 2017). The customers of B2C firms, consumers, use the web as a critical channel for information and engage more in web browsing and search. The browse and search frequency is lower for customers of B2B firms than for the B2C customers. B2B customers prefer information over accessibility when they interact with a website. Wang & Xu (2017) show that website visits are more valuable for B2C firms than B2B firms. Cortez & Johnson (2017) prospected the future challenges for B2B marketers and argue that research should be done in six major topics: Innovation, Customer Journey and Relationship Value, Data Analytics, Harnessing Technology, Marketing/Finance Interface and Revenue Growth, and Industry Context or Ecosystem. This research focuses on those topics. In
13 previous research, reviewed in the Literature Review, the main focus is on the online marketing behaviour of the B2C firms. The effect of the online marketing channels for B2B firms has not been researched within the literature yet. This research will address the gap in the current literature.
Hypothesis
Research is already done on the effect of different online marketing channels, using different
attribution models. The current literature lacks research on the use of these channels for B2B firms. In this research the focus is on the effect (contribution to conversions and sales revenue) of the OMC within a B2B company, compared to the effects for a B2C company (figure 2).
What is the effect of the different online marketing channels on the performance (sales revenue/conversions) of the different firm types (B2B vs B2C)?
As previously stated, European marketers use, on average, seven online marketing channels. While looking at the same channels for both types of firms, do their customers behave differently among these channels? Wang & Xu (2017) state that the browser and search frequency of B2B-customers is lower than the customers of B2B firms:
H1. B2B-customers use less online marketing channel touch points within the path to a conversion than the customers of B2C firms.
The browsing behaviour of the B2C-customers differs from the B2B-customers. B2B firms have a limited number of customers who drive the sales revenue. A customer of B2B firms is more important for the firm performance than a customer for a B2C firm. Therefore the B2B firm needs to be in contact with their customers in the online environment (Wang & Xu, 2017). Expected is that this contact is initiated by the firm. For B2B firms the FICs will have a higher impact on the performance of the firm than the CICs.
H2a. The contribution of FICs to the revenue of the firm are higher than the contribution of CICs for B2B firms.
14 H2b. The contribution of FICs to the conversions of the firm are higher than the contribution of CICs for B2B firms.
Customers of B2C firms tend to shop around and browse the web for information. The CICs are used by the customers to get in contact with a firm. These channels will have a higher impact on the performance of the firm than the FICs, even when 45% of the online marketing budget is spent on the Search channel (IAB, 2017):
H3a. The contribution of CICs to the revenue of the firm are higher than the contribution of FICs for B2C firms
H3b. The contribution of CICs to the conversions of the firm are higher than the contribution of FICs for B2C firms
Figure 2 Conceptual model
Methodology
Data
The datasets for the case study are provided by a business-to-business online shop for tools and
technical parts and a business-to-customer online fashion retailer for the date range from the 3rd of July
2017 (Q3) to the 31st of December 2017 (Q4). Within this timeframe a total of 40726 conversions are
15 effectiveness of the online marking channels in a multi-channel environment. The datasets are
analyzed within Rstudio, an integrated development environment (IDE). Before importing the datasets to this tool, some adjustments were made. Within the dataset of the B2C company, two outliers (one positive and one negative) were present. The positive outlier, an incorrectly calculated conversion with a significant sales revenue, was rectified the day after with the negative outlier. Both conversions were deleted from the dataset. The dataset of the B2B company contained extensive use of online marketing channels. The channels „Shopping‟, „SEA Non Branded‟ and „SEA Branded‟ were marked as Paid Search, in order to have consistency in the use of channels for both companies. These channels are all forms of paid results within a search engine. The B2B company also reported the „Metasearch‟ as a separate channel, which is marked as Organic Search. These adjustments were discussed with an expert in the field of data analysis and online marketing. After the data cleansing and adjustments, the following default channels are defined and used for this research: {Direct, Organic search, Paid search, Email, Affiliate, Social Network, Display}. Customer journeys contains one or more „touches‟ across a variety of channels (Anderl et al. 2016). The journey of a customer can lead to a conversion. The (top 10) paths among these channels that result in a conversion for both firms are added in the appendix. The first channel mentioned is the starting channel of the customer journey. These conversion paths contain the data about the activities and journeys of the customer on the different online marketing channels (Schultz & Dellnitz, 2017). Different attribution models can be applied to determine the impact of each channel on the amount of conversions and the revenue.
Description Dataset 1 Dataset 2
Firm focus B2C B2B
Industry Fashion Retail Tools and technical parts
Number of channels 9 8
Number of conversion paths 3385 2687
Maximum number of touch points 161 133
Number of conversions 20752 19974
Number of conversions 1 path journey 6942 7645 Number of conversions 2 paths journey 4252 4216 Number of conversions 5+ paths journey 5241 4051
Table 2 Descriptions
16 Two different multi-channel attribution models will be applied to achieve the research objective, determining the effect of the firm focus on the effectiveness of online marketing channels. Last-click, First-click and Linear attribution will be applied as rule-based models. These models determine the individual impact of the different online marketing channels by using pre-defined rule sets. With these rules the models imply that some channels/customer contacts are more relevant than others to
conclude for a conversion (Schultz & Dellnitz, 2017). The Last-click and First-click models presume that the last or first channel is the only relevant touch of the customer for the decision to convert. In other words, the full reward for the conversion is given to that last or first channel, without rewarding the other channels. In the Linear model, all channels touched by the customer are rewarded with an equal credit for the conversion (Schultz & Dellnitz, 2017). These heuristic models, Last-click in particular, are often used by marketers to report the results of the channels, while prior research indicates that these approaches of attributing conversion can calculate incorrect conclusions (Kannan & Li, 2014).
Figure 3 Use of attribution models in 2015 (Schultz & Dellnitz, 2017).
The Markov-chain attribution model will be used as a data driven attribution model. Markov chain, named after the Russian mathematician Andrey Markov, is “a stochastic model describing a sequence of possible events in which the probability of each event depends only on the state attained in the previous event” (Qi & Luo, 2017). This probabilistic model can represent dependencies between sequences of observation of a random variable (Anderl et al. 2016). Data driven attribution models are
17 based on more sophisticated calculations of the observed customers‟ movement patterns across
different touch points, than the heuristic models (Schultz & Dellnitz, 2017).
Within this research, the customer journeys are represented as chains, visualized in Markov graph (figure 4). A Markov graph M = <S,W> is defined by a set of states S = {S1,....,Sn}. As stated in table 2,
the customer journeys within the firms exists of eight or nine different channels. These channels are included as the state in the journey of the customer. The „start‟ state, the starting point of the journey, and a „conversion‟ state as end state of a successful conversion are included in this journey as well. The full set of states S in the B2B dataset is as follows: S = {Start, Conversion, Direct, Organic search, Paid search, Email, Affiliate, Social Network, Display, Other}. Within the B2C dataset one additional channel is present {Unavailable}. The source of this touch point could not be set as one of the others. The actual results of the Markov Chain analysis of this research will be represented in a graph, in which the channels are classified in the two forms.
18 An attribution model can be developed by using the representation as Markov graphs, which allows identifying structural correlations in the customer journey data (Anderl et al. 2016). The transition probabilities are defined by Anderl et al. (2015), used in the analysis within Rstudio, as follows:
P(Xt = St|Xt-1 + St-2,....,X1 + S1 ) = P(Xt = St|Xt-1 + St-1 ,Xt-2 = St-2 ,....,Xt-k = St-k )
The longest path to a conversion by the customer of the B2C firm was 161 touches among the different channels, while the customer of the B2B firm touched the channels 133 times in the path to the{Conversion} state. The top 10 conversion paths of both firms are mentioned in Appendix A.
Model fit
As figure 3 indicates, Last-click model is most often used by marketers to report the results of the channels. Previous research implies that these approaches of attributing conversion, heuristic models, can calculate incorrect conclusions (Kannan & Li, 2014). In order to evaluate the model fit of different models, the predictive accuracy and robustness can be measured (Anderl et al. 2016). The receiver operating characteristics (ROC) is used as measure to evaluate predictive accuracy. Secondly the robustness, the ability of a model to deliver stable and reproducible results if the model is run multiple times and is therefore indispensable for sustainable attribution results, is applied to two measures. First the robustness of the predictive accuracy across all cross-validation repetitions is measured. Next to that, the robustness of the „Removel Effect‟ is measured, in order to evaluate the stability of the attribution results of each channel (Anderl et al. 2016). The results of the model fit of the Markov model meets the following criteria: objectivity, predictive accuracy, robustness, interpretability, versatility and algorithmic efficiency (Kakalejčík et al. 2018).
Results
Based on the report of the model fit by Anderl et al. (2016), the results of the Markov model will be used for the final reporting of the results. The results of the analysis on Last-click model will be reported and compared to the Markov model results as well, as this model is most often used by marketers in the field.
19 Paths to conversion
In table 2 the customer journeys of both the B2B and B2C customers are summarized. The data in this table gives insights into the behaviour of the customer among the different channels used by the companies. For almost a similar amount of conversions, the customers of the B2B company used less different paths (2687) to convert, compared to the customers of the B2C company (3385). Also the path with the most touches among the channels by a customer before converting was longer for the B2C firm. This indicates that the customers of B2C companies visited the website more often before placing the online order. By focusing on the amount of conversions after touching different amounts of channels, the journeys of the customers are analysed. 38% of the conversions are placed by the B2B customer after touching one channel, compared to 33% of conversions for the B2C customers with that same „single journey‟. The B2B customers with longer journeys, touching five channels or more, contribute for 20% of the total conversions in that time period, while the customers of the B2C firms with a similar journey contribute for 25% of the conversions. The hypothesis that B2B-customers use less online marketing channel touch points within the path to a conversion than the customers of B2C firms (H1), based on the statement of Wang & Xu (2017) that the B2C consumers have the tendency to shop around while the browse and search frequency of customers in B2B field are lower, can be accepted. This hypothesis is supported by the datasets of this research.
Results of OMC on Business-to-Business and Business-to-Consumer performance (revenue)
The performance of the firm is divided into the conversions and the revenue. In order to analyse the effect of the firm focus on the firm performance, the results of the online marketing channels (CICs and FICs) are compared between both firms. The results of the Markov model analyses, with the aggregated datasets of the firms with the data range of Q3 and Q4 of 2017, on the revenue are
combined in graph 1. Both hypothesis 2a and hypothesis 3a focus on the effect of the online marketing channels on the revenue of the firm. As illustrated (graph 1), a difference in effect of the channels on the revenue of the firms is present. The FICs account for 60% of the revenue generated by the B2B firm, while those channels account for 43% of the revenue generated by the B2C firm. In addition, the CICs contribute to 57% of the revenue for the B2C firm. The channels in which the contact is coming
20 from the B2B firm contribute to 40% of the revenue of the firm. When focusing solely on the B2B firm, the contribution of the FICs on the revenue are higher than the contribution of the CICs on the revenue. The B2B customer who touched the FICs generated more revenue for the firm than the customers who touched the CICs in their customer journey. The hypothesis that the FICs has a higher contribution on the revenue than the CICs for the B2B firm (H2a) can be accepted. The exact opposite effect is hypothesised for the B2C firm. The customers who touched the CICs within their customer journey generated more revenue than the customer who touched the FICs. The hypothesis that the CICs have a higher contribution on the revenue than the FICs for the B2C firm (H3a) can be accepted as well. This hypothesis is supported by the datasets of this research.
Graph 1 Results of OMC on revenue with Markov
The results of the channels on the revenue of the firm is also analysed on the weekly datasets for both firms. The Markov model is applied to the weekly conversion paths from week 40 to 50. The results of the analysis on the effect of the firm focus on the revenue generated by the different online marketing channels, are reflected in both graph 2a and graph 2b. The contribution of the FICs on the revenue of the B2B firm within these weeks is 59,8%, with a maximum deviation of 3,85% (week 42) and a standard deviation of 2,10. The contribution of the CICs on the B2B firms revenue within this period is 40,2% (graph 2a). The contribution on the revenue of the B2C firm by the CICs within these weeks is 54,6%, with a maximum deviation of 4,24% (week 45) and a standard deviation of 2,44. The
21 contribution of the FICs on the B2C firms revenue within this period is 42,6% (graph 2b). The overall effect of online marketing channels on the firms revenue is also present within the weekly datasets.
Graph 2a Weekly results of OMC on B2B revenue with Markov
Graph 2b Weekly results of OMC on B2C revenue with Markov
The results of the Markov model are also illustrated in table 3, which is an overview of all models applied to the datasets (Q3 and Q4 2017). The analysis of the heuristic models show different results of the contribution, by the different channels, on the revenue for both B2B and B2C firms. Within this overview the contribution of the channels which were not defined as a FIC or CIC („Other‟ and „Unavailable), are summarized as well.
22
Table 3 Overview models: Results of OMC on revenue
Results of OMC on Business-to-Business and Business-to-Consumer performance (conversions)
A different indicator for the firm performance used in this research is the amount of conversions. The same approach is used to analyse the contribution of the different online marketing channels on the conversions of the B2B and B2C firm. The difference in effect of the channels on the conversions between both firms is analysed and reflected in graph 3. Within the B2B firm the contribution of both types of channels on the revenue is even more out of ratio. The contribution of the FICs on the conversions generated by the B2B firm is higher (64,5%) than the contribution of the CICs (35,5%). The hypothesis that the FICs have a higher contribution on the conversions than the CICs for the B2B firm (H2b) can be accepted. Graph 3 also illustrates a difference in contribution of the channels for the B2C. Similar to the contribution on the revenue of the B2C firm, the CIC contributes for more
conversions (54%) than the FIC (43%). The remaining 3% of the conversions is contributed to the „Other‟ channels. The hypothesis that the CICs have a higher contribution on the conversion than the FICs for the B2C firm (H3b) can be accepted and is supported by the datasets of this research.
The results of the channels on the conversion of the firm is also analysed within the weekly (week 40 to 50) datasets. The contribution of the type of channel for both firms are illustrated in graph 4a and 4b. The contribution of the FICs on the conversion of the B2B firm within these weeks is 63,4%, with a maximum deviation of 5% (week 40) and a standard deviation of 2,38. The contribution of the CICs on the conversion of the B2B firms within this period is 38,5% (graph 4a). Within the B2C firm the
23 effect of the channels on the conversions are more balanced, yet more conversions are contributed to the CICs on a weekly basis. The contribution of these channels is 53,7%, with a maximum deviation of 4,48% (week 45) and a standard deviation of 2,48, within these eleven weeks. The FICs contribute for 43,2% on the conversions of the B2C firm (graph 4b). The remaining 3,1% is the contribution of the „Other‟ channels. The results of the analysis on the Q3 + Q4 2017 datasets are also supported by the weekly datasets.
Graph 3 Results of OMC on conversion with Markov
24
Graph 4b Weekly results of OMC on B2C conversion with Markov
The results of all used attribution models are illustrated in table 4, applied to the datasets (Q3 and Q4 2017). The contribution of the heuristic models to the conversions for both firms are different from the Markov model analysis. Within this overview the contribution of the channels which were not defined as a FIC or CIC („Other‟ and „Unavailable), are shown in this table. The 3,1% of contribution of these channels for the B2C firm, mentioned in the previous part, is reflected in this overview (the first bar in table 4).
Table 4 Overview models: Results of OMC on conversion
Conclusions
Multichannel attribution helps to gain more insights into the effectiveness of the online marketing channels to the performance of the firm. As the attribution measures this in a specific setting, the
25 results are conditional upon a number of management decisions. These include the choice in the channels used or budgets per channels. Having a correct attribution is a pre-requisite for the optimization of the budget allocation among the channels (Anderl et al. 2016). This research is the starting point of an analysis on the effect of the online marketing channels on the different firm types, business-to-business and business-to-customer. The customers of a B2C firm have different
characteristics, they tend to shop around more often, than the customers of a B2B firm (Wang & Xu, 2017). These differences are supported by this research, the effect of online marketing channels on both the conversions and the revenue of the firms are different. The contribution of the CICs on the conversions and revenue for B2C firms are higher than the FICs, based on the Markov model. Previous research states that the CICs, „Direct‟, results in more conversion, which is in line with the findings of this research (Kannan & Li, 2014; De Haan et al. 2016). However the FICs, „Paid Search‟, contribute to more conversions and revenue than the CICs for B2B firms. The FICs contribute to 64,5% of the conversions while these channels account for 60% of the revenue. The average order value of a conversion coming from a customer who touched a FIC in the journey is lower than a conversion from a customer who touched a CIC within a B2B firm. The CICs are more effective on the performance (conversions and revenue) of B2C firms, while the FIC are more effective for B2B firms.
Managerial implications
The marketers of B2B firms need to take into account that contact with the customer which is initiated from the firm result in better performance (both conversion and revenue). The channels in which the contact is initiated by the customer itself results in better performance for the B2C firms. With this given, those responsible for the performance of the different online marketing channels can optimize the budgets among the channels. These findings are based on the Markov model analysis. As figure 3 illustrates, the most frequently used model by the marketers is the Last-click model. The analysis of this Last-click model on the datasets show almost similar results on the contribution of the different types of channels on the different firms. However, for the B2C firm, the contribution of the CICs on the firm performance (both conversion and revenue) is higher than the analysis of the Markov model.
26 Therefore the results of the FICs on the firm performance would be underestimated. As a result, less budget will be allocated by marketing managers to these channels. By using the Markov model as attribution model instead, the potential of the FICs will be taken into account by allocating more budget to these channels.
Limitations and Future research
Although this research focuses on a gap in the literature, the effect of the firm focus on the firm performance by the different online marketing channels, the findings are based on the datasets of two firms. Extensive further research on this topic, on the datasets of more firms, need to be done to generalize the findings. The firms within this research operate in two different fields, online retail (B2C) and tools and maintenance (B2B). The datasets for the analysis can also be extended with the customer journeys that do not lead to a conversion. By taking these paths into account, the fit of the model can be analysed („receiver operating characteristics‟ and the „removal effect‟), which improves the internal validity of the research. In this research the findings of the model fit by Anderl et al. (2016) are used. Within this research seven different main channels are used for the analysis. A deeper analysis on some main online marketing channels can be done in future research. The „Paid Search‟ channel, advertising on the search engine, can be split into separate channels based on the keyword approach. Paid search using „branded‟ keywords, such as the name of the company, and paid search by using „non-branded‟, more generic, keywords are examples of keyword approaches. Earlier research has shown that „branded‟ search has a positive effect on the click-through rate, returning customers and conversion rate (Rutz, Trusov & Bucklin, 2011). „Social Network‟ was used as a channel initiated by the customer. However, a firm can also use this channel to promote posts or advertise with
sponsored posts. A distinction between „Organic‟ and „Paid‟ social network channels can be made in future research. The channels themselves can be analysed individually in further research, to report on the effectiveness of the individual channel level, instead of dividing the channels into two types.
Finally, the interpretation of the customer journey is difficult, because alternative explanations for correlations between conversions and advertising exposure may exist (Anderl et al. 2016). The correlations observed between the dependent variable, in this research the firm performance, and
27 advertising might be due to selection effects: the difference in outcomes of the exposed and unexposed populations may be for reasons having nothing to do with the advertising (Lewis, Rao & Reiley, 2011). Although research on a large-scale is needed to establish a causal relationship, this research is a stepping stone for the difference in the B2B and B2C online marketing environment. The approach of this research, together with the indicated focus points, would be valuable to explore in future research.
28
References
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.
Anderl, E., Schumann, J. H., & Kunz, W. (2016). Helping firms reduce complexity in multichannel online data: A new taxonomy-based approach for customer journeys. Journal of retailing, 92(2), 185-203.
Anderson, C., & Cheng, M. (2017). Multi-Click Attribution in Sponsored Search Advertising: An Empirical Study in Hospitality Industry. Cornell Hospitality Quarterly, 58(3), 253-262.
Berman, R. (2015). Beyond the last touch: Attribution in online advertising.
Blattberg, R. C., Kim, B. D., & Neslin, S. A. (2008). Database marketing: analyzing and managing customers. Heidelberg, Germany: Springer.
Bryl, S. (2016). Marketing Multi-Channel Attribution Model With R. AnalyzeCore.
Dalessandro, B., Perlich, C., Stitelman, O., & Provost, F. (2012). Causally motivated attribution for online advertising. Proceedings of the Sixth International Workshop on Data Mining for Online Advertising and Internet Economy, 1-9.
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.
Gupta, S., Miglani, S., & Sundriyal, V. (2013). Search Engine Optimization Techniques. International Journal of Engineering, 2(9).
Holmstrom, B. (1982). Moral hazard in teams. The Bell Journal of Economics, 324-340.
IAB (2017). Online ad spend study – H1 2017. https://iab.nl/kennisbank/online-ad-spend-study-h1-2017/
29 Kakalejčík, L., Bucko, J., Resende, P. A., & Ferencova, M. (2018). Multichannel Marketing
Attribution Using Markov Chains.
Kannan, P. K., & Li, H. A. (2017). Digital marketing: A framework, review and research agenda. International Journal of Research in Marketing, 34(1), 22-45.
Kireyev, P., Pauwels, K., & Gupta, S. (2016). Do display ads influence search? Attribution and dynamics in online advertising. International Journal of Research in Marketing, 33(3), 475-490.
Klapdor, S. (2013). Effectiveness of online marketing campaigns. An investigation into online multichannel and search engine advertising, Munich: Springer Gabler.
Lee, W., & Sanderson, M. (2010). Analyzing URL queries. Journal of the American Society for Information Science and Technology, 61(11), 2300-2310.
Lemon, K. N., & Verhoef, P. C. (2016). Understanding customer experience throughout the customer journey. Journal of Marketing, 80(6), 69-96.
Lewis, R., Rao, J., & Reiley, D. (2011). Here, there, and everywhere: Correlated online behaviors can lead to overestimates of the effects of advertising. Proceedings of the 20th International Conference on World Wide Web, 157-166.
Mora Cortez, R., & Johnston, W. J. (2017). The future of B2B marketing theory: A historical and prospective analysis. Industrial Marketing Management, 66, 90-102.
Neslin, S. A., & Shankar, V. (2009). Key Issues in Multichannel Customer Management: Current Knowledge and Future Directions. Journal of Interactive Marketing, 23(1), 70-81.
Pérez-Aróstegui, M. N., Bustinza-Sánchez, F., & Barrales-Molina, V. (2015). Exploring the relationship between information technology competence and quality management. BRQ Business Research Quarterly, 18(1), 4-17.
Petrov, S., Zubac, V., & Milojevic, M. (2015). Social networks as marketing channels. Ekonomika, 61(2), 153.
30 Rutz, O. J., Trusov, M., & Bucklin, R. E. (2011). Modeling indirect effects of paid search advertising: Which keywords lead to more future visits?. Marketing Science, 30(4), 646-665.
Shankar, V., & Malthouse, E. (2007). The growth of interactions and dialogs in interactive marketing. Journal of Interactive Marketing, 21(2), 2-4.
Shao, X., & Li, L. (2011). Data-driven multi-touch attribution models. Proceedings of the 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 258-264.
Schultz, C. D., & Dellnitz, A. (2017). Attribution Modeling in Online Advertising. Multi-Platform Advertising Strategies in the Global Marketplace, 226-249.
Wang, F., & Xu, B. (2017). Who needs to be more visible online? The value implications of web visibility and firm heterogeneity. Information & Management, 54(4), 506-515.
Wiesel, T., Pauwels, K. P., & Arts, J. (2011). Marketing‟s profit impact: quantifying online and offline funnel progression. Marketing Science, 30(4), 604–611.
Yang, S., & Ghose, A. (2010). Analyzing the relationship between organic and sponsored search advertising positive, negative, or zero interdependence? Marketing Science : The Marketing Journal of the Institute for Operations Research and the Management Sciences, 29(4), 602-623.
Zhang, Y., Wei, Y., & Ren, J. (2014). Multi-touch Attribution in Online Advertising with Survival Theory. Data Mining (ICDM), 2014 IEEE International Conference on, 687-696.
Zhao, K., Mahboobi, S., & Bagheri, S. (2017). Revenue-based Attribution Modeling for Online Advertising.
31
Appendix A
Top 10 conversion paths
The paths of the customers that most often result in a conversion for both firms are listed in the tables below. The {Direct} channel is present in seven paths that result in a conversion. This channel is a CIC, which has the most impact on both conversions and revenue for the B2C firm.
MKT Path – B2C Conversions Conversion Value
Direct 2898 13,28%
Paid Search 2536 11,14%
Direct > Direct 1035 5,18%
Organic Search 857 4,12%
Paid Search > Direct 759 3,50%
Direct > Direct > Direct 571 2,94% Paid Search > Paid Search 442 1,82%
Direct > Referral 432 2,55%
Paid Search > Direct > Direct 362 1,72% Direct > Direct > Direct > Direct 347 1,69% Organic Search > Direct 294 1,49%
Table 5 Top 10 conversions path B2C firm
In addition, the {Paid Search} channel is most often touched by the customers of the B2B firm within the conversion paths. This channel is part of the FIC, which contributes for almost two-thirds of the conversions (and 60% on the revenue).
MKT Path - B2B Conversions Conversion Value
Paid Search 5934 22,11%
Paid Search > Paid Search 2270 8,30%
Paid Search > Paid Search > Paid Search 1026 3,97%
Direct 1013 5,65%
32
Organic Search 494 3,06%
Paid Search > Paid Search > Paid Search > Paid Search 480 1,96%
Direct > Direct 276 2,24%
Paid Search > Paid Search > Direct 325 1,87% Paid Search > Direct > Direct 259 1,59%
Table 6 Top 10 conversions path B2B firm
Appendix B
Rstudio Syntax
The conversion paths for both firms are analyzed within Rstudio with both the heuristic models (First click, Last click, Linear) and Markov model. The formula of the Markov model used in Rstudio is similar to the formula of Anderl et al. (2015),
P(Xt = St|Xt-1 + St-2,....,X1 + S1 ) = P(Xt = St|Xt-1 + St-1 ,Xt-2 = St-2 ,....,Xt-k = St-k )
The syntax and output for both analyses on both firms are presented below (output on the conversion value is censored). The weekly datasets of both firms, week 40 to 50, are analysed in a similar way. The syntax and outputs (22 in total) of these weekly datasets are not added in the Appendix.
33
Illustration 2 Syntax and output of B2C firm within Rstudio
Appendix C
Revenue Model Direct Paid
search
Organic Search
Affiliate Email Social Network
Other Display UNA%
B2B FIRST 18,12% 71,05% 6,96% 1,33% 1,78% 0,09% 0,01% 0,66% 0,00% B2B LAST 39,08% 49,03% 6,09% 3,37% 1,32% 0,01% 0,00% 1,11% 0,00% B2B LINEAR 29,29% 59,15% 6,37% 2,17% 1,63% 0,04% 0,00% 1,36% 0,00% B2B MARKOV 30,18% 48,48% 9,25% 4,68% 2,90% 0,08% 0,01% 4,41% 0,00% B2C FIRST 44,11% 33,67% 11,35% 0,73% 5,91% 2,74% 0,94% 0,41% 0,14% B2C LAST 59,49% 18,89% 6,68% 7,87% 3,56% 1,80% 1,38% 0,29% 0,05% B2C LINEAR 54,41% 24,14% 8,22% 4,96% 4,48% 2,17% 1,15% 0,40% 0,07% B2C MARKOV 41,24% 25,22% 10,60% 8,44% 8,25% 2,34% 2,67% 1,10% 0,15%
34
Conversion Model Direct Paid search
Organic Search
Affiliate Email Social Network
Other Display UNA%
B2B FIRST 12,33% 79,13% 5,29% 1,39% 1,28% 0,10% 0,02% 0,48% 0,00% B2B LAST 28,81% 60,27% 5,26% 3,00% 1,05% 0,03% 0,01% 1,57% 0,00% B2B LINEAR 21,40% 68,76% 5,15% 2,03% 1,16% 0,05% 0,01% 1,44% 0,00% B2B MARKOV 26,54% 52,88% 8,82% 4,43% 2,66% 0,10% 0,02% 4,56% 0,00% B2C FIRST 42,30% 35,90% 10,85% 0,69% 5,85% 2,51% 1,29% 0,48% 0,15% B2C LAST 58,06% 20,81% 6,85% 6,80% 3,65% 1,73% 1,76% 0,30% 0,04% B2C LINEAR 52,55% 26,23% 8,19% 4,46% 4,55% 2,04% 1,48% 0,42% 0,06% B2C MARKOV 40,72% 26,07% 10,68% 7,90% 8,21% 2,30% 2,87% 1,10% 0,15%
Table 8 Channel contribution on firm conversion Q3 + Q4
Based on the results of the analysis within Rstudio, the contribution of each channel is calculated in table 7 and 8. The contribution of the „Direct‟, „Organic Search‟ and „Social Network‟ channels are combined for the total customer-initiated-channel contribution on the firm performance. The „Paid Search‟, „Affiliate‟, „Email‟ and „Display‟ channels are combined for the total firm-initiated-channel contribution on the firm performance.
