INTEGRATING DIFFERENT TYPES OF TARGETING METHODS IN ONLINE ADVERTSING Changyu Wang, School of Business and National Academy of Development and Strategy, Renmin University of China, Beijing, China, [email protected] Bin Zhu, College of Business, Oregon State University, Corvallis, United States, [email protected] Meiyun Zuo, School of Information and National Academy of Development and Strategy, Renmin University of China, Beijing, China, [email protected]

Abstract Rule based targeting, behavioural targeting, and contextual targeting are three different types of targeting methods, delivering customized ads based on different categories of variables. A rule based targeting method uses viewers’ own characteristics, while behavioural targeting delivers ads based on previous content users have browsed. At the same time, contextual targeting uses the characteristics of the webpage on which an ad will be displayed to decide which ads to display. On the other hand, all the three categories of factors and the interaction among them could impact the effectiveness of an online displaying ad. For organizations seeking to optimize their online marketing effort, it therefore becomes crucial to have a thorough understanding on how the variables from individual, behavioural, and contextual aspects moderate the effectiveness of a targeting method based on variables in other categories. But we have not yet found any study provides a theoretic framework to support such understanding. This paper thus proposes to investigate the moderation effect of variables in different categories on the effectiveness of a targeting method based on variables from other categories by using cookie-based data collected by an online marketing firm for its online displaying ads. Keywords: Contextual targeting, Behavioural targeting, Retargeting, Contextual relevance, Purchase funnel, Cookie-based data.

1

INTRODUCTION

Online displaying advertising business is experiencing and will continue to have rapid increase. More than a trillion internet display ads are delivered each quarter in the United States (comScore 2013). And the spending for online displaying advertising is expected to reach $33.4 billion per year in 2017 (eMarketer 2013). At the same time, it has become the trend that people are deliberately avoiding paying attention to banner ads (Drèze & Hussherr 2003), resulting in decreased effectiveness of online advertisements. At the same time, various targeting methods have been proposed to deliver ads that are more relevant to users. All those targeting methods infer users’ preference and deliver ads accordingly. Rule based targeting methods infer users’ preference based on their own demographic (Jansen et al. 2013) or other individual information (Gilbert 2014; Hill 2012), while a behavioral targeting method uses information about the content that users have browsed (Yan et al. 2009) and contextual targeting methods apply the characteristics of the webpage where the ads will be displayed to infer viewers’ preference (Anagnostopoulos et al. 2007; Broder et al. 2007; Joshi et al. 2011). While each type of the targeting methods customizes the displaying ads in different ways, the main goal is to improve the perceived relevance of the ads displayed. Many studies have dedicated themselves to uncover the factors that impact the effectiveness of a targeting method (Al-Natour et al. 2013; Chen 2009; Chen & Stallaert 2014; Doorn & Hoekstra 2013; Magee 2012; Sokolik et al. 2014; Yan et al. 2009) or to identify circumstances under which a targeting method could be effective (Bleier & Eisenbeiss 2015a, 2015b; Chun et al. 2014; Goldfarb & Tucker 2011; Lambrecht & Tucker 2013). At the same time, previous studies have demonstrated that the effectiveness of a targeting method, regardless of the category of variables it is based on, could be impacted by factors from other categories (Aguirre et al. 2015; Bleier & Eisenbeiss 2015b; Hoban & Bucklin 2015). As a result, using one marketing method in conjunction with others does not always lead to higher effectiveness compared with using each targeting method alone (Goldfarb & Tucker 2011). There are both studies reporting higher effectiveness when combining two types of displaying methods (Joshi et al. 2011; Li et al. 2009) and those finding that two types of displaying strategies offset each other’s impact (Goldfarb & Tucker 2011). For organizations seeking to optimize their online marketing effort, it becomes crucial to understand how the variables from different categories, each of which was used by different targeting methods, interact with each other and how such interaction impact the effectiveness of targeting methods. However, we have not yet found a study that investigates the impact of such interaction. Most studies focus either on the optimization of either behavioral targeting (Bleier & Eisenbeiss 2015a, 2015b; Lambrecht & Tucker 2013) or contextual targeting (Chun et al. 2014; Goldfarb & Tucker 2011). This study thus proposes to fulfill the gap by utilizing cookie-based data collected by a marketing firm to investigate the impact of interactions between individual, behavioral, and contextual factors. The data documents not only users’ browsing behavior history but also contextual information under which an ad was displayed, providing a platform to study the impact of these two types of factors. More specifically, we will study the interaction between one type of contextual factor, content consistency (Chun et al. 2014) and one type of behavior factor, dynamic retargeting (Lambrecht & Tucker 2013). At the same time, the data set also records a user’s status in purchase funnel (i.e., create an account, put items in shopping cart, or paid), which could be applied to investigate how a user’s purchase funnel status could moderate the impact of different marketing strategies. We expect the contribution of this study to be two folds. On one hand, it extends existing understanding on existing targeting methods by investigating the interactive effect of variables from different categories, providing a more holistic perspective for the optimization of online ads displaying strategies. On the other hand, although there are studies demonstrating that how the combination of two targeting methods could lead to higher effectiveness under a specific circumstance, we have not yet found studies providing

guidelines on how to integrate different types of targeting methods to maximize the impact of a displaying method. The paper contributes to this aspect of online advertising by identifying circumstances under which the combination of two targeting methods could lead to higher effectiveness. The overall implication of this study will be that it provides practitioners a more holistic perspective about the optimization of online ads displaying effort. Different types of targeting methods have long been considered to be distinctively different strategies for online banner ads. And our study will show that practitioners should consider all different types of factors in order to maximize the return of their marketing budget. The rest of this article is organized as follows. Section 2 reviews related literature and describes the development of theoretic framework, followed by section 3 that depicts the data and the method of this study. And finally section 4 reports the status of the project and discusses possible contributions and implications of the study.

2

BACKGROUND

As online displaying ads have become one of the major paths to reach potential customers, internet users are also starting avoiding paying attention to banners displayed (Cho & Cheon 2004; Sun et al. 2013). Many methods have been proposed to draw people’s attention to the ads, among which selecting appropriate ads that could be interesting to a user to display appears to be an effective strategy. There are three categories of ads selection strategies, each of which uses different types of variables to infer users’ preference. We will review each type in detail below. 2.1

Rule based targeting

A rule based targeting strategy chooses ads for a specific user based on the user’s own attributes, including demographics (Jansen et al. 2013), psychographics (Gilbert 2014), firmographics (Hill 2012), product purchase history (Bleier & Eisenbeiss 2015b), and status in purchase funnel (Hoban & Bucklin 2015). Given that the data used in this study was collected from marketing campaign where all users belong to the same pre-selected segment, this paper thus focuses users’ status in purchase funnel. Purchase funnel is a theoretic model that describes a customer’s journey from his first contact of a brand to the final purchase (Hoban & Bucklin 2015). Hoban & Bucklin (2015) extends this model into the context of internet, dividing people into categories of visitors, authenticate users, and converted customers. Visitors are those who have visited the website of a company but do not provide any identifiable personal information. Compared with those in other categories, users in this category usually have least amount of awareness about the company and its product/service (Hoban & Bucklin 2015). And they usually also have least trust about the company and its website. Authenticated users are those who have created an account at the website. People in this category have more awareness and more trust about the company (Hoban & Bucklin 2015). Authenticated users reveal their personal information by creating an account with the website but usually have not yet indicated their product preference. And converted users, on the other hand, reveal their product preference with the website by completing a transaction there (Hoban & Bucklin 2015). This is the group that has the most amounts of awareness and trust about the website and the company. And the website also has the most information about them. It is apparent people in different purchase funnel stages may react differently to a displaying ad about the website. Previous studies have shown that the purchase funnel stage does impact the effectiveness of a behaviour targeting strategy (Hoban & Bucklin 2015). But we have not yet found studies investigating how it moderates the effectiveness of contextual strategies. At the same time, it is also equally interesting to understand how the impact of purchase funnel stages varies with other behavioural or contextual variables.

2.2

Behavioural targeting

Behavioural targeting denotes to delivering ads based on users’ previous surfing behaviours (Yan et al. 2009). Behavioural targeting usually involves anonymously tracking content read and sites visited by an IP address. The system thus infers a user’s preference and interest based on his surfing history and then selects ads to display accordingly (Yan et al. 2009). It is beyond the scope of this paper to discuss all the behavioural targeting methods here. This paper focuses on one of the widely adopted behavioural targeting methods, dynamic retargeting (Lambrecht & Tucker 2013). This is a method of personalizing the ads delivered to a user by dynamically selecting the products that the user has checked at other websites before (Bleier & Eisenbeiss 2015b; Lambrecht & Tucker 2013). Such high degree of personalization is believed to enhance the relevance of an ad and make it more appealing to its viewer (Ansari & Mela 2003; Malheiros et al. 2012). At the same time, there are also studies suggesting high degree of personalization could cause the concerns for privacy (Bleier & Eisenbeiss 2015a). As a result, dynamic retargeting could enhance users’ perceived vulnerability (Aguirre et al. 2015) thus lead to high degree of resistance (Bleier & Eisenbeiss 2015a) and lower effectiveness of the ads. Consequently, many studies have dedicated themselves to uncover circumstances under which the dynamic retargeting could be effective. For example, it has been found that high-fit with the need of a user could mitigate the concern for the privacy of the user (Doorn & Hoekstra 2013). And the effectiveness of a dynamic retargeting method varies with the stage in purchase funnel (Bleier & Eisenbeiss 2015b; Hoban & Bucklin 2015; Lambrecht & Tucker 2013), characteristics of the ad itself (Aguirre et al. 2015; Doorn & Hoekstra 2013), users’ perception of the retailor (Aguirre et al. 2015; Bleier and Eisenbeiss 2015a), and users’ perception of the website in which the ad was displayed (Aguirre et al. 2015; Bleier & Eisenbeiss 2015b; Tucker 2014). In summary, a dynamic retargeting method could be very effective when it displays on a website that users trust with an ad from a reputable retailer and with a product/service that precisely fulfils users’ need at the moment. At the same time, it has been shown that the contextual factors of the webpage on which an ad is displayed could also have significant impact on the effectiveness of the ad (Chun et al. 2014; Shen & Chen, 2007). But few studies have investigated the impact of the content of the webpage where the ad is displayed on the effectiveness of a dynamic retargeting method. 2.3

Contextual targeting

Contextual targeting denotes to delivering ads to users based on the content they are reading. The main rationale is that matching ads to the content a user is consuming could increase the perceived relevance of the ads (Anagnostopoulos et al. 2007; Broder et al. 2007; Joshi et al. 2011), resulting more favourable attitude toward the ad and advertised brand (Chun et al. 2014; Shen & Chen, 2007), higher click through rate (Chatterjee, Hoffman, & Novak, 2003) and higher degree of purchase intention (Goldfarb & Tucker 2011). And the effectiveness of a contextual targeting largely depends on the algorithm applied to match ads and web page content. Many matching methods have been proposed to enhance the perceived relevancy of ads displayed (Broder et al. 2007; Lacerda et al. 2006; Ribeiro-Neto et al. 2005; Yih et al. 2006). Most of the proposed methods were developed from the perspective of computer science, seeking to identify computation algorithms to enhance the precision of the ads delivered to users. However, if improving perceived relevance is the ultimate goal of contextual targeting, we need to realize that perceived relevance could be very subjective, varying with characteristics of users, webpages, and the products/services advertised. It therefore is almost impossible to find a universal optimal algorithm that can improve the perceived relevancy for any ad to any user of a webpage. In another word, the perceived relevance of the same ad-page pair may vary with different users. It therefore is crucial to understand how other factors impact the effectiveness of a contextual targeting method. And we have yet found very limited studies that investigate this aspect of contextual targeting.

2.4

Integration of different targeting methods

In summary, rule based targeting, behavioural targeting, and contextual targeting deliver customized displaying ads to users. The difference between these three types of targeting methods comes from the variables they use to infer users’ preference. A rule based targeting method selects ads using users’ own characteristics, while behavioural targeting delivers ads based on previous content users have browsed. And contextual targeting considers the characteristics of the webpage on which an ad will be displayed to decide which ads could be more relevant to a user. The outcome of any type of targeting strategies could be moderated by factors from the other two categories or the interaction between them. And consequently combining two different types of marketing methods could result in either higher (Bleier & Eisenbeiss, 2015a; Joshi et al. 2011; Li et al. 2009) or lower effectiveness (Goldfarb & Tucker 2011) than using each method alone. And all these studies demonstrated the change in performance when two types of targeting methods were combined under a specific circumstance. We still lack a theoretic framework that provides a thorough understanding on how the impact of one type of targeting method varies with the variables in the other categories. Such understanding is important because it provides a guideline to marketers seeking to optimize their marketing strategies. This paper thus proposes to investigate the moderation effect of variables in one category on a targeting method from other categories. We selecte content matching and dynamic retargeting, two most commonly used methods in contextual and behavioural targeting methods, respectively to study how variables used by one type method moderate the impact of the other type of targeting methods.

3

PROPOSED RESEARCH AND STATUS

We will use cookie-based data collected by a marketing firm during one of their marketing campaigns in September 2014. The campaign lasted for 19 days. The data documented online activities of users who agreed to participate. The campaign displayed banner ads that promote an online store at various websites. For those who have visited the site of the online store, the system recorded the items a user had selected to check in detail. Therefore, the items displayed on the banner ads could either be a set of randomly selected items or those have been checked by the user before. We will use the binary variable of click through as dependent variable, the content consistency between the content of a webpage and that of a banner ad as the measure of contextual relevancy, and number of items displayed on the banner that have been checked by a user before as the measure of degree of personalization for dynamic retargeting. The data set also contains information about a user’s activities at the online store website, including visiting the site, creating an account with the site, or completing a transaction. We will also use this information as the measure for user’s purchase funnel stage. The research is currently at the stage of developing theoretic model for the project. We are also conducting literature review to identify appropriate computational algorithm to capture the content of a webpage and to calculate the content consistency between a page and a displaying ad. We are also in the process of preparing the data collected for data analysis.

4

SUMMARY

In summary, we propose to analyze a cookie-based data collected by a marketing firm to understand the interactions between variables that measure different aspects of an online displaying ad. More specifically, we will study how the variables used by one type of targeting method moderate the effectiveness of other types of targeting methods. Findings from this study expect to contribute to existing understanding of online displaying ads from several perspectives. It extends existing research in behavioral and contextual targeting by uncovering the context in which each type of targeting methods could result in optimal

outcome. In addition, our study is also expected to shed a light on how different targeting methods should be combined to maximize the return of spending on online displaying ads. For practitioners, the findings from this study could also provide them a more detailed guideline on how to optimize their online marketing strategies.

ACKNOWLEDGEMENT This work was supported in part by the Fundamental Research Funds for the Central universities and the Research Funds of Renmin University of China under Grant 10XNJ065, National Natural Science Foundation of China under Grant 71273265, National Social Science Foundation of China Major Program under Grant 13&ZD184, and part by the Fund of China Scholarship Council. The third author is the corresponding author, and his e-mail is [email protected].

Reference Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., and Wetzels, M. (2015). Unraveling the Personalization Paradox: The Effect of Information Collection and Trust-Building Strategies on Online Advertisement Effectiveness. Journal of Retailing, 91 (1), 34–49. Al-Natour, S., Gemino, A., and Krider, R. (2013). Effective Online Ads: The Role of Placement and Animation, In Proceedings of ICIS 2013, p.65, Milan (available at http://aisel.aisnet.org/icis2013/proceedings/ResearchInProgress/65). Anagnostopoulos, A., Broder, A. Z., Gabrilovich, E., Josifovski, V., and Riedel, L. (2007). Just-in-time Contextual Advertising, In Proceedings of the Sixteenth ACM Conference on Conference on Information and Knowledge Management (CIKM 07), p. 331–340, ACM, New York, NY, USA. Ansari, A., and Mela, C. F. (2003). E-Customization, Journal of Marketing Research, 40 (2), 131–145. Bleier, A., and Eisenbeiss, M. (2015a). The Importance of Trust for Personalized Online Advertising, Journal of Retailing, 91 (3), 390–409. Bleier, A., and Eisenbeiss, M. (2015b). Personalized Online Advertising Effectiveness: The Interplay of What, When, and Where, Marketing Science, 34 (5), 669–688. Broder, A., Fontoura, M., Josifovski, V., and Riedel, L. (2007). A Semantic Approach to Contextual Advertising, In Proceedings of the 30th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’07), p. 559–566, ACM, New York, NY, USA. Chatterjee, P., Hoffman, D. L., & Novak, T. P. (2003). Modeling the Clickstream: Implications for WebBased Advertising Efforts. Marketing Science, 22 (4), 520–541. Chen, J. V. William H.Yen, David C.Akhapon, Lerdsuwankij. (2009). The Effect of Types of Banner Ad, Web Localization, and Customer Involvement on Internet Users’ Attitudes, Cyber Psychology & Behavior, 12 (1), 71–73. Chen, Y., Pavlov, D., and Canny, J. F. (2009). Large-scale Behavioral Targeting, in Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’09), p. 209–218, New York, NY, USA. Cho, C.-H., and Cheon, H. J. (2004). Why Do People Avoid Advertising on the Internet? , Journal of Advertising, 33 (4), 89–97. Chun, K. Y., Song, J. H., Hollenbeck, C. R., and Lee, J.-H. (2014). Are contextual advertisements effective? The moderating role of complexity in banner advertising, International Journal of Advertising, 33 (2), 351–371. Doorn, J. van, and Hoekstra, J. C. (2013). Customization of online advertising: The role of intrusiveness, Marketing Letters, 24 (4), 339–351.

Drèze, X., and Hussherr, F.-X. (2003). Internet advertising: Is anybody watching?, Journal of Interactive Marketing, 17 (4), 8–23. eMarketer. (2013). Advertisers Continue Rapid Adoption of Programmatic Buying, (available at http://www.emarketer.com/Article/Advertisers-Continue-Rapid-Adoption-of-ProgrammaticBuying/1010414; retrieved March 5, 2016). Gilbert, S. 2014. How to Use Psychographic Data in Your Online Marketing, (available at http://susangilbert.com/use-psychographic-data-online-marketing/; retrieved March 5, 2016). Goldfarb, A., and Tucker, C. (2011). Online Display Advertising: Targeting and Obtrusiveness, Marketing Science, 30 (3), 389–404. Hill, B. (2012). “firmographics" that help pinpoint prime selling opportunities, (available at http://www.businessbrief.com/6-firmographics-that-help-pinpoint-prime-selling-opportunities/; retrieved March 5, 2016). Hoban, P. R., and 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. Jansen, B. J., Moore, K., and Carman, S. (2013). Evaluating the performance of demographic targeting using gender in sponsored search, Information Processing & Management, 49 (1), 286–302. Chen, J. and Stallaert, J. (2014). An Economic Analysis of Online Advertising Using Behavioral Targeting, MIS Quarterly, 38 (2), 429–A7. Joshi, Amruta, Bagherjeiran, Abraham, and Ratnaparkhi, Adwait. (2011). User Demographic and Behavioral Targeting for Content Match Advertising, In Proceedings of the Fifth International Workshop on Data Mining and Audience Intelligence for Adervertising (ADKDD 2011), p. 53–60, Association for Computing Machinery, New York, NY, USA. Lacerda, A., Cristo, M., Gonçalves, M. A., Fan, W., Ziviani, N., and Ribeiro-Neto, B. (2006). Learning to Advertise, In Proceedings of the 29th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’06), p. 549–556, ACM, New York, NY, USA. Lambrecht, A., and Tucker, C. (2013). When Does Retargeting Work? Information Specificity in Online Advertising,” Journal of Marketing Research, 50 (5), 561–576. Li, T., Liu, N., Yan, J., Wang, G., Bai, F., and Chen, Z. (2009). A Markov Chain Model for Integrating Behavioral Targeting into Contextual Advertising, in Proceedings of the Third International Workshop on Data Mining and Audience Intelligence for Advertising (ADKDD ’09), p. 1–9, ACM, New York, NY, USA. Magee, R. G. (2012). Impression formation online: How web page colors can interact with physical temperature, Journal of Media Psychology: Theories, Methods, and Applications, 24 (3), 124–133. Malheiros, M., Jennett, C., Patel, S., Brostoff, S., and Sasse, M. A. (2012). Too Close for Comfort: A Study of the Effectiveness and Acceptability of Rich-media Personalized Advertising, in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’12), p. 579–588, ACM, New York, NY, USA. Ribeiro-Neto, B., Cristo, M., Golgher, P. B., and Silva de Moura, E. (2005). Impedance Coupling in Content-targeted Advertising, in Proceedings of the 28th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’05), p. 496–503, ACM, New York, NY, USA. Sokolik, K., Magee, R. G., and Ivory, J. D. (2014). Red-Hot and Ice-Cold Web Ads: The Influence of Web Ads’ Warm and Cool Colors on Click-Through Rates, Journal of Interactive Advertising, 14 (1), 31–37. Tucker, C. E. (2014). Social Networks, Personalized Advertising, and Privacy Controls, Journal of Marketing Research, 51 (5), 546–562.

Yan, J., Liu, N., Wang, G., Zhang, W., Jiang, Y., and Chen, Z. (2009). How Much Can Behavioral Targeting Help Online Advertising?, in Proceedings of the 18th International Conference on World Wide Web (WWW ’09), p. 261–270, ACM, New York, NY, USA. Yih, W., Goodman, J., and Carvalho, V. R. (2006). Finding Advertising Keywords on Web Pages, in Proceedings of the 15th International Conference on World Wide Web (WWW ’06), p. 213–222, ACM, New York, NY, USA. Sun, Y., Lim, K. H., and Peng, J.Z. (2013). Solving the Distinctiveness–Blindness Debate: A Unified Model for Understanding Banner Processing, Journal of the Association for Information Systems, 14 (2), 49–71.