Abstract
Advertisers use multiple online channels to reach consumers over the Internet. Yet, to the current authors' knowledge, little is known about how prior user behavior across these advertising channels can be used to predict conversions. To address this issue, the authors of the current paper drew on advertising-response and purchase decision-making theory, as well as findings about user search and information processing on the web. Analyzing a clickstream data set (electronic records of users' online activity), the authors found that consumer reactions to advertising messages through multiple channels were strong predictors of purchase propensity. Advertisers, the authors believe, can use the results to predict purchasing behavior based on prior reactions to multichannel online advertising and develop individualized targeting strategies.
MANAGEMENT SLANT
Advertising messages delivered through multiple channels are strong predictors of purchase propensity.
Both the total number and number of different sources of website visits positively predict conversion probabilities.
Channel order indicates purchase propensity: A transition from information to navigation channels increases conversion probability, whereas a switch in the other direction negatively affects conversion rates.
Advertisers can use the results to develop individualized targeting strategies.
INTRODUCTION
Consumers increasingly use the Internet to purchase goods and services. Between 2007 and 2013, Internet retailing grew by more than 11 percent per annum1 and another 14 percent from 2013 to 2014.2 To benefit from this growth, firms have invested a major proportion of their advertising budget in digital advertising. Between 2003 and 2013, for example, U.S. spending for search, display, and other forms of online marketing climbed by 19.4 percent annually, reaching U.S.$42.8 billion in 2013.3
Prior research shows that clickstream data—that is, the electronic record of a user's activity on the Internet—can be used to predict purchase behavior based on the number of previous visits to a website (Moe and Fader, 2004; van den Poel and Buckinx, 2005) or in-site behavior (Montgomery, Li, Srinivasan, and Liechty, 2004; Sismeiro and Bucklin, 2004). E-commerce managers can use this information to optimize their advertising strategies across multiple online channels and to target potential consumers with individualized messages.
Prior research on online advertising using individual-level data has focused predominantly on single online advertising channels, such as
display advertising (Braun and Moe, 2013; Goldfarb and Tucker, 2011; Manchanda, Dubé, Goh, and Chintagunta, 2006);
retargeting (Lambrecht and Tucker, 2013);
sponsored search advertisements (Ghose and Yang, 2009; Nottorf and Funk, 2013).
Existing studies examining and comparing the effectiveness of different online advertising channels often rely on aggregated data (Breuer and Brettel, 2012; Breuer, Brettel, and Engelen, 2011) and, therefore, do not allow prediction on an individual consumer level.
Despite this growing body of literature on clickstream analysis and Internet advertising, research on predicting individual user behavior—based on users' prior reaction to advertising in different online channels—is, to the authors' knowledge, still scarce. Specifically, little is known about how such data can be applied to predict conversion behavior, when advertisers use multiple online channels to communicate messages to individual users.
Can user clicks on advertising across multiple online channels be used to predict purchase propensities when targeting users on an individual level? If so, is there a difference in predictive power, depending on the sequence of channels used by individual consumers? Does customer experience with the online retailer affect the diagnosticity of multichannel interactions?
The current study attempted to answer these questions by investigating the relationship between the user journey and purchasing behavior in an online context. A user journey consists of all visits to an online retailer's website preceding a potential purchase decision and the originating channel for each visit.
The current authors analyzed how consumers react to advertiser messages in different channels and particularly how conversion rates of individual users are related to
the number of different online channels in their user journey;
the order of these channels in the journey;
their interactions with past purchase behavior.
The authors' proposed model derived from theories in
marketing (advertising response, purchase decision making) and
information retrieval (user search, information processing on the web)
and tested whether taxonomies developed in an information retrieval context can be applied to online advertising and purchase decision making. They estimated this model using a large data set gathered from an online retailer in the fashion-and-apparel industry, which contained approximately 1.6 million user journeys.
The authors of the current paper used single-source data at a highly granular cookie level and, thus, could identify unique users. Furthermore, the data included visits originating in a wide variety of online channels (See “Order of Channels,” p. 435), as well as users' past and present purchasing behavior. The results showed that the proposed variables were highly diagnostic for purchase behavior and thus provided valuable information for online retailers.
LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT
Prediction Model
To investigate the predictive capability of user reactions to online multichannel communication for individual purchasing behavior, in the prediction model, the authors explored three main characteristics (See Figure 1):
number of different channels involved in a user journey
channel order in the journey
their interactions with previous purchase.
Additionally, the model included well-established predictor variables such as the number and timing of visits and historical purchase behavior (van den Poel and Buckinx, 2005). The analysis covered the following six digital channels:
organic search engine links (organic search)
sponsored search engine links (paid search)
newsletters
paid links from affiliated websites such as fashion blogs (affiliate)
links established through sponsorship agreements from other websites such as beauty and fashion magazines (link)
banner advertising on various websites (display).
Conversion Prediction Model
Number of Different Channels
Several studies have demonstrated the synergetic effects of advertising campaigns through multiple channels (Chang and Thorson, 2004; Havlena, Cardarelli, and de Montigny, 2007; Naik and Peters, 2009; Naik and Raman, 2003).
Unlike consumers who repeatedly view advertiser messages on a single channel, people exposed to messages on different channels report (Chang and Thorson, 2004)
higher attention;
higher message credibility ratings;
more positive thoughts about the advertiser brand.
In the context of online multichannel advertising, consumer reactions in different channels can be a sign that the consumer actively is engaging in information search from different sources.
Some have argued that the purchase probability for customers increases with the number of heterogeneous attributes and signals searched (Branco, Sun, and Villas-Boas, 2012). The current authors, therefore, hypothesized that reaction to advertising messages across different channels positively predicts purchase probability—even when controlling for the total number of contacts.
H1: The higher the number of online channels on which consumers react to advertiser messages, the higher their subsequent conversion probability.
Order of Channels
Research on web-search behavior suggests that users have different goals when using the Internet (Broder, 2002; Jansen, Booth, and Spink, 2008). According to the primary intent of the user, classified web queries have been classified as informational, navigational, or transactional in nature (Broder, 2002).
Informational Channels
The authors of the current paper proposed that this taxonomy for web-search behavior, which has been empirically validated in several subsequent studies (Jansen et al., 2008; Rose and Levinson, 2004), also applies to online user journeys and reactions to advertising messages in different channels. They asserted that consumers in different stages of their purchase decision process have different goals and thus react to advertising on different digital channels.
To examine how the order of clicks in different channels influences purchase probabilities, the authors designated online channels according to their primary use by consumers as information or navigation channels (Broder, 2002). In one scenario, a user's goal was informational if he or she wanted “to learn something by reading or viewing web pages” (Rose and Levinson, 2004, p. 15). Users generally learn from blogs or websites with professional editorial content (e.g., magazines); accordingly, the authors of the current paper classified affiliate, display, and link channels as informational and predicted that when consumers use these information channels, they are in information-acquisition mode.
Navigational Channels By contrast, a navigational-user goal implies that the consumer intends to go to a “specific website that the user has in mind” (Rose and Levinson, 2004, p. 14), by clicking on a search-engine results page (organic search, paid search) or by reacting to a newsletter. When consumers use these channels, they are in navigation mode.
The current study classified search engines as purely navigational because it only considered traffic from keywords including the retailer's brand. Users often use search engines for navigation if they do not know the exact URL or because it is simply more convenient than typing in a known URL (Rose and Levinson, 2004).
Furthermore, the current authors assumed that users with transactional goals—the user intends to reach a website where further interaction will occur (Rose and Levinson, 2004)—also start in navigational mode. They have to browse to the corresponding website before performing the transaction—online shopping in the specific case of the current paper.
A switch from one mode to another affects purchase probabilities, in line with the theory of choice or consideration sets (Hauser and Wernerfelt, 1990; Howard and Sheth, 1969; Narayana and Markin, 1975). Sequential multistage choice processes first were introduced for product choice but nowadays also find application in retailer or store choice situations (Fotheringham, 1988; Spiggle and Sewall, 1987).
Of all retailers in the market that carry a specific product or product category, consumers are only aware of a limited number of retailers, the “awareness set.” The so-called “evoked set” is a subset of the awareness set and contains a limited number of retailers that a consumer actively considers for a specific purchase (Spiggle and Sewall, 1987). To build and refine their evoked sets, consumers seek information—in the given case, they do so online.
Consider the following scenarios:
A user starts browsing in information-acquisition mode by searching magazine websites and blogs. While doing so, he or she is exposed to advertiser information in the form of a banner advertisement or a link. The consumer becomes aware of the relevance of this specific brand for reaching his or her goals, clicks on the link, and proceeds to the retailer's website.
If this consumer later switches to navigation mode and navigates directly to the retailer's website—using a search engine or clicking on a link in a received newsletter—it could indicate that he or she plans to include (or already has included) the retailer into his or her evoked set and wants more information about this seller.
By definition, the purchase likelihood for brands in the evoked set is higher than that for brands outside the set, so conversion probability should increase substantially after a user switches from information acquisition to navigation mode.
If a user, instead, starts off in navigation channels and then switches back to information-acquisition mode, such behavior may indicate that his or her evaluation of the advertiser's website provided no clear preferences, so that the user still needs to refine the evoked set or could even exclude the retailer from the evoked set.
In this case, future advertiser messages in information channels likely are less useful than they were in the prior case, because the consumer already is aware of the advertiser but proactively explores other options.
In the latter case, the authors of the current paper, therefore, predicted a decrease in purchase probability:
H2a: Conversion probability is higher if a user starts his or her user journey in information channels and then switches to navigation channels.
H2b: Conversion probability is lower if the user starts his or her user journey in navigation channels and then switches to information channels.
Interaction of Multichannel Effects With Previous Purchase
Multichannel effects should differ between existing and new customers. The current authors based this statement on previous studies, which have shown that an existing customer relationship moderates the advertising effectiveness of single online channels (Breuer and Brettel, 2012) and can be used to predict current online purchase behavior (Baecke and van den Poel, 2010).
Although the Internet has considerably reduced search costs, consumers only engage in limited information search during the purchase process (Johnson et al., 2004). This phenomenon often is explained by “cognitive lock-in” (Johnson, Bellman, and Lohse, 2003; Murray and Häubl, 2007)—in other words, repeated use of a website increases the probability that the consumer will continue to use it.
Practice and learning make using the site more efficient, which, in turn, leads to lower cognitive costs and, thus, to higher procedural-switching costs (Burnham, Frels, and Mahajan, 2003). If a customer recently has purchased at an online store, he or she already is familiar with the site and therefore needs less additional information to reach a final purchase decision. Therefore, the effect of higher attention and higher message credibility through multichannel communication should be stronger for new customers than for existing customers:
H3: Reactions to advertiser messages across multiple channels are stronger predictors of conversion probability for new than for existing customers.
Implicitly, this line of argumentation for channel order assumes no prior experience of the customer with the provider. In case of a previous purchase, however, customers probably already have included the retailer in question into their evoked set.
Therefore, for existing customers, one can assume that the current study's focal store already is part of their awareness set. A click on a message in an informational channel in this case can be seen as a positive update of the customer's evoked set, which is dynamic over time (Shocker, Ben-Akiva, Boccara, and Nedungadi, 1991). When followed by a reaction in a navigational channel, this sequence might indicate a very imminent final choice and thus a high purchase probability. Thus,
H4a: The increase in conversion probability after a switch from information to navigation channels is greater for existing than for new customers.
If an existing user visits a site through a navigational channel and later comes back via an informational channel, this should indicate a decrease in purchase probability, as the user is proactively exploring alternative options and browsing in informational mode (H2b).
The authors of the current article hypothesized that this switch from navigational to informational channels is less diagnostic for existing customers because they may experience procedural switching costs due to cognitive lock-in (Burnham et al., 2003; Johnson et al., 2003).
H4b: The decrease in conversion probability after a switch from navigation to information channels is greater for new than for existing customers.
In summary, the current hypotheses introduced the number of different channels, channel order, and interactions of previous purchase with multichannel variables as predictors of conversion probability.
METHODOLOGY
Research Model
The authors used a nested logit approach to test whether the model predicts the probability that an individual consumer ultimately will engage in a transaction. The dependent variable was conversion, i.e., the completion of a purchase (CONVi; 1 = consumer converts, 0 = not). To test hypotheses H1, H2a, and H2b, the main effects model (Model 1) introduced the number of different channels CHi and channel order as predictors of conversion probability.
The authors operationalized channel order using the binary dummy variables SWITCH_INi and SWITCH_NIi: if a user started his or her journey in an informational channel and then switched to a navigational channel, the respective variable SWITCH_INi was set to 1 (0 otherwise).
b2 measured how an information–navigation switch affects conversion probability, whereas
b3 measured the same effect for a change from navigation to information channels.
Additionally, the model included previous purchase (PPRi), which is a well-established predictor for purchase behavior in clickstream research (Moe and Fader, 2004; van den Poel and Buckinx, 2005), as a binary variable, indicating whether the focal consumer has shopped at the retailer's online store before (1 = yes, 0 = no).
Two additional covariates,
the number of previous visits (INTi) and
the duration (DURi) of the user journey in hours
helped to account for observed heterogeneity. Including the number of previous visits ensured that the multichannel effect measured in CHi is not confounded with the total number of visits, which is again an established predictor of purchase behavior in clickstream research (van den Poel and Buckinx, 2005).
Finally, there might be systematic differences in the conversion behavior of users with specific first and last channels. The current authors therefore added two sets of K–1 dummy variables (where K denotes the total number of channels):
FCH[2..K],i and
LCH[2..K],i.
Each dummy variable represented an individual channel and took a value of 1 if the browsing journey started (FCH variables) or ended (LCH variables) with that particular channel. Otherwise, it was 0. This control was especially important for the SWITCH variables, which would otherwise capture all effects arising from systematic differences between channels.
To test whether multichannel effects interact with previous purchase (H3, H4a/b), Model 2 extended the base model with the interaction terms CHi × PPRi, SWITCH_INi × PPRi, and SWITCH_NIi × PPRi .
The Data
The authors estimated these two nested models using cookie-level clickstream data from a European fashion-and-apparel retailer. This pure e-commerce player, which requested anonymity in the current study, has no brick-and-mortar stores. Consumers can purchase only from the advertiser's online store, so there should not be any direct effects from online–offline research shopping—that is, when consumers use the Internet for search and an offline channel for purchase (Verhoef, Neslin, and Vroomen, 2007). It is important to mention that the online retailer did not use any behavioral targeting or retargeting mechanisms during the data-collection period.
The data covered all the digital-advertising channels described above and showed, for each channel, when a consumer clicked on an advertising message. They also contained conversion information, such that one can determine whether a consumer has purchased something, the exact time, and the date of the purchase. This information was used to construct user journeys that describe the visits to the retailer's website by an individual consumer originating in a variety of online channels and the reaction in terms of purchasing behavior.
Each user journey reflected an individual consumer and ended with either inactivity or a conversion. For consumers with prior purchase, all interactions since that conversion were included in the data. In an initialization procedure, the authors excluded any journeys that exhibited only one click or interactions that lasted less than one second. This may occur when users double-click on an advertising message.
The data set represented all incoming traffic that generated the site's revenues over a period of six months between 2010 and 2011. In total, it consisted of journeys from 1,664,673 users, of which 12,426 converted (0.746 percent). On average, journeys included 3.9 contacts in 1.3 channels and lasted 9.9 days; 1.2 percent of the cases belong to existing customers (See Table 1).
RESULTS
In this section, the maximum likelihood estimation results are explained (See Table 2). Multicollinearity diagnostics showed that variance inflation factors (VIF) of the six main variables were lower than the critical level of 10 (Kutner, Nachtsheim, Neter, and Li, 2004). In addition, none of the main effects were strongly correlated. To rule out a potential bias due to unobserved factors related to individual channels, the authors additionally conducted a robustness test in which they removed individual channels from the sample. Coefficients varied to some extent across individual models, whereas the magnitude and direction of the effects were highly consistent, further confirming the validity of the results.
As hypothesized in H1, the number of different channels involved had a positive effect on conversion probability (b1 = 0.732, p < 0.001). Switches between informational and navigational channels also influenced purchase probabilities, and both coefficients were highly significant.
Moreover, the large and positive coefficient (b2 = 1.941, p < 0.001) of SWITCH_IN suggested a strong increase in conversion probability (by a factor of nearly 7) when a consumer started by browsing on information channels and then later responded to advertiser messages in navigation channels. This finding supported H2a.
The coefficient of the SWITCH_NI variable (from navigation to information channels) also was significant but with a negative sign (b3 = −0.161, p < 0.001). The exp(b3) of 0.852 implied that conversion probability decreased by about 15 percent when consumers responded to advertiser messages in information channels after starting with the objective to navigate straight to the advertiser, which supported H2b.
The interaction of number of channels involved and previous purchase in the model, including multichannel interactions, was negative and significant (b7 = −0.363, p < 0.001). That is, the effect from multichannel online advertising to conversion probabilities differed for new and existing customers. An additional channel in the user journey of a new customer increased the purchase probability by a factor of around 2.4, whereas the increase for existing customers was only around 1.7, supporting H3.
It is important to note that in the model with the interaction terms included, the coefficients of CH (b1 = 0.864, p < 0.001), SWITCH_IN (b2 = 1.457, p < 0.001), SWITCH_NI (b3 = −0.306, p < 0.001), and PPR (b4 = 3.738, p < 0.001) do no longer represent the average difference between the two groups but rather the simple effects at particular levels of the other variables included in the interaction (Irwin and McClelland, 2001). The result for the interaction between the channel-order variable SWITCH_IN and previous purchase (PPR) was positive and significant (b8 = 1.136, p < 0.001).
As hypothesized in H4a, if an existing customer switched from information to navigation channels, the corresponding effect was substantially stronger than for new customers. In agreement with H4b, the effect from SWITCH_NI was positively influenced by previous purchase (b9 = 0.380, p < 0.001). In fact, as an additional analysis using existing customers as the reference category showed, the negative effect was even completely cancelled out and SWITCH_NI was not significant (b3′=0.075, p > 0.1) for existing customers.
Descriptive Statistics
Estimation Results
DISCUSSION AND CONCLUSIONS
Research Implications
The current study presented and tested a theoretical model to explain how reactions to advertiser messages across different online channels can be applied to predict conversion rates in an e-commerce context. The authors estimated the proposed model using a large, consumer-level data set from an online retailer in the fashion-and-apparel industry. The data included activities across six online channels, together with conversion data, such that they represented actual purchase behavior.
The contribution of this study to online advertising research is at least threefold.
The current research established the number of different channels in a user journey as a new predictor of purchase probability. The estimation results showed that not only the total number but also the sources of website visits significantly predict conversion probabilities in online multichannel settings. Thus, the number of different online advertising channels can be regarded as a proxy for the number of heterogeneous attributes a consumer gathers during information search.
As the purchase threshold decreases with the number of attributes and signals considered (Branco, Sun, and Villas-Boas, 2012), purchase probability increases with the number of channels used.
Following an interdisciplinary approach, the current authors showed that taxonomies developed in an information retrieval context can be applied successfully to predict online purchases. In particular, they relied on an earlier taxonomy of search (Broder, 2002) and its operationalization (Jansen et al., 2008).
The new categorization presented in the current study enables advertisers to group online channels by primary user intent, namely, as informational and navigational channels. With the input of consideration set theory, the authors of the current study explained how channel order can be used to predict purchase probabilities.
Their finding: A transition from information to navigation channels leads to a significantly higher conversion probability, but a switch in the other direction seems to affect conversion rates adversely.
The authors found that previous purchase not only does not moderate advertising effectiveness of individual advertising channels (Breuer and Brettel, 2012) but also is an important moderator for the predictive capabilities of multichannel advertising variables.
Because of cognitive lock-in (Johnson et al., 2003), reactions to advertiser messages by existing customers differ in predictive power. Reactions to advertiser messages across multiple channels have a stronger positive effect on conversion probability for customers without previous purchases, whereas a switch from informational to navigational channels is more predictive for existing customers than for new customers.
A switch in the reverse direction indicates a lower purchase probability for new customers but has no measurable effect for existing customers. Hence, multichannel online advertising may not be equally effective for existing and new customers. Research on online-marketing should consider this fact and include corresponding controls in future studies.
MANAGERIAL IMPLICATIONS
For online advertisers and especially for online retailers, the findings of the current study provide valuable suggestions for advertising strategies across multiple online channels:
Advertisers should coordinate simultaneous advertising messages according to the consumer's previous browsing behavior. Consumers whose first contact is through information channels—affiliates, blogs, or fashion websites, for instance—will exhibit significantly greater purchase probabilities if they subsequently click on an advertiser message in navigation channels. Advertisers, therefore, should ensure their visibility among these consumers on navigation channels.
As search engine providers start offering “remarketing” options (Google Inc., 2014), information on prior multichannel usage enables advertisers to target individual customers, for example, by increasing their bids for users who come from information channels. If, however, potential new customers start their user journey in navigational channels, a subsequent reaction to messages in informational channels leads to a decrease in purchase probability, indicating that the user might still be refining his or her evoked set. Advertisers might, therefore, consider targeting these users with special offers or rebates, for example by adapting the landing pages.
Finally, this study showed that, in addition to previous channels viewed, the user's past purchase behavior needs to be considered when scheduling advertisements, because reactions to online advertising messages differ in predictive ability between new and existing customers. Executing such a strategy requires sophisticated targeting mechanisms for online marketing instruments. To operationalize corresponding advertising scheduling strategies, real-time targeting mechanisms need to be enhanced to enable considerations of previous purchase behaviors of customers.
For publishers, search engines, technology providers, and vendors of ad servers or tracking and optimization tools, the implications of this study are closely related to those for online advertisers: Providers need to enhance current tracking mechanisms to identify unique consumers throughout their entire user journey.
Ideally, this identification would include both advertising exposures (i.e., impressions) and users' reactions (i.e., clicks and conversions). Targeting mechanisms also should support real-time decisions about whether to present an advertisement to a certain user. In an ideal world, targeting mechanisms would work in an integrated manner across all the channels included in an online campaign and also include data from other sources, such as CRM databases.
LIMITATIONS AND OUTLOOK
The current study has several limitations that provide avenues for additional research. Using cookie data naturally imposes two restraints:
The assumption that one cookie represents one consumer is not necessarily true, such as when several users share the same computer. This concern, however, is relatively minor, because
cookies appropriately represent individual users on modern multi-user operating systems (Drèze and Zufryden, 1998), and
cookie data have been successfully used in a number of studies, especially in clickstream research (Bucklin and Sismeiro, 2003; Manchanda et al., 2006; Sismeiro and Bucklin, 2004).
Users can delete cookies at any point in time (and many modern web browsers provide options to automatically remove them), so they have a limited lifetime (Flosi, Fulgoni, and Vollman, 2013).
In the current study, the advertiser expected an average cookie lifetime of 30 days, which should not pose any problem, because the user journeys in the data indicated a median duration of 2.8 days and an average duration of 9.9 days (SD = 14.7).
Additionally, the data set used in the current study featured only one online retailer and a single product category with several distinctive aspects. In general, data from multiple firms might help to control for potential biases from individual, firm-specific effects. Replicating the analyses with a multiple-advertiser data set across different product categories could validate the findings of this study and produce more differentiated insights into the interaction effects.
Furthermore, the data set was restricted to branded search terms and did not include generic keywords. Including this information could help to refine the prediction results. Yet, the current study showed that retailers successfully can predict conversions based on limited information.
Finally, the analysis only covered clicks on advertiser messages and did not include advertising exposures. Clicks are not directly within an advertiser's control, as firms only can influence exposures and not behavioral user responses in the form of clicks. Although the current study on reactions to advertiser messages already gave valuable indications for optimizing multichannel online advertising strategies, future research should include exposures as well as clicks to account for intermediate advertising effects.
In summary, this study showed that advertising response and purchase decision-making theory—as well as findings about user search and information processing on the web—provide a valuable basis for interpreting multichannel clickstream data. As consumer reactions to advertising messages through multiple channels are strong predictors of purchase propensity, advertisers can use this information to develop and optimize individualized targeting strategies.
ABOUT THE AUTHORS
Sebastian Klapdor is associate principal at McKinsey & Company in Munich, Germany. He authored this paper when he was a PhD student at TUM School of Management at Technische Universität in Munich. His research focuses on multichannel online advertising, clickstream analysis, and e-commerce. Klapdor's work has been published in the Journal of Interactive Marketing.
Eva Anderl is senior consultant at FELD M, a digital-marketing consulting firm in Munich. She co-authored the current study when she was a PhD student at Universität Passau, Germany. Anderl specializes in using online marketing data to better understand consumer behavior and help marketers maximize their return on investment. Her research can be found in the Journal of Interactive Marketing.
Jan H. Schumann is professor of marketing and innovation at Universität Passau. His primary research interests are online marketing, technology and innovation, and pricing of services, as well as relationship marketing. Schumann's work has been published in the Journal of Marketing, Journal of Retailing, Journal of Service Research, Journal of Business Venturing, and other publications. Schumann also is a member of the editorial board of the Journal of Service Research and the Thunderbird International Business Review.
Florian von Wangenheim is professor of technology marketing at ETH Zurich, Switzerland. His research specialties are technology-intensive service management and value-based customer management. Von Wangenheim's work has appeared in the Journal of Marketing, MIS Quarterly, Journal of the Academy of Marketing Science, Journal of Retailing, Journal of International Business Studies, Journal of Service Research, Journal of Interactive Marketing, among other journals. He serves on the editorial boards of Journal of Marketing and Journal of Service Research.
Footnotes
↵1 “E-Stats 2013: Measuring the Electronic Economy.” U.S. Census Bureau. Retrieved July 22, 2015, from http://www.census.gov/econ/estats/e13-estats.pdf].
↵3 “IAB Internet Advertising Revenue Report 2013.” Price-WaterhouseCoopers, 2014.
- © Copyright 2015 The ARF. All rights reserved.
REFERENCES
Vol 55 Issue 4
