Oliver J. Rutz

From Generic to Branded: A Model of Spillover in Paid Search Advertising

In Internet paid search advertising, marketers pay for search engines to serve text advertisements in response to keyword searches that are generic (e.g., “hotels”) or branded (e.g., “Hilton Hotels”). Although standalone metrics usually show that generic keywords have higher apparent costs to the advertiser than branded keywords, generic search may create a spillover effect on subsequent branded search. Building on the Nerlove–Arrow advertising framework, the authors propose a dynamic linear model to capture the potential spillover from generic to branded paid search. In the model, generic search advertisements serve to expose users to information about the brands ability to meet their needs, raising awareness that the brand is relevant to the search. In turn, this can induce additional future search activity for keywords that include the brand name. Using a Bayesian estimation approach, the authors apply the model to data from a paid search campaign for a major lodging chain. The results show that generic search activity positively affects future branded search activity through awareness of relevance. However, branded search does not affect generic search, demonstrating that the spillover is asymmetric. The findings have implications for understanding search behavior on the Internet and the management of paid search advertising.

A Dynamic Model of the Effect of Online Communications on Firm Sales

Interpersonal communications have long been recognized as an influential source of information for consumers. Internet-based media have facilitated information exchange among firms and consumers, as well as observability and measurement of such exchanges. However, much of the research addressing online communication focuses on ratings collected from online forums. In this paper, we look beyond ratings to a more comprehensive view of online communications. We consider the sales effect of the volume of positive, negative, and neutral online communications captured by Web crawler technology and classified by automated sentiment analysis. Our modeling approach captures two key features of our data, dynamics and endogeneity. In terms of dynamics, we model daily measures of online communications about a firm and its products as contributing to a latent demand-generating stock variable. To account for the endogeneity, we extend the latent instrumental variable technique to account for dynamic endogenous regressors. Our results demonstrate a significant effect of positive, negative, and neutral online communications on daily sales performance. Failure to account for endogeneity results in a severe attenuation of the estimated effects. From a managerial perspective, we demonstrate the importance of accounting for communication valence as well as the impact of shocks to positive, negative, and neutral online communications.

A Latent Instrumental Variables Approach to Modeling Keyword Conversion in Paid Search Advertising

The authors present a modeling approach to assess the purchase conversion performance of individual keywords in paid search advertising. The model facilitates estimation of daily keyword conversion and click-through rates in a sparse data environment while accounting for the endogenous position of the text advertisement served in response to a search. Position endogeneity in paid search data can arise from both omitted variables and measurement error. The authors propose a latent instrumental variable approach to address this problem. They estimate their model on keyword-level paid search data containing daily information on impressions, clicks, and reservations for a major lodging chain. They find that higher positions increase both the click-through and conversion rates. When advertisements are served in higher positions, approximately one-third of new conversions is due to increased click-through while approximately two-thirds are due to increased conversion rates. The authors show that the keyword list generated on the basis of their estimated conversion rates outperforms the status quo list as well as lists generated by observed conversion and click-through rates.

Modeling Indirect Effects of Paid Search Advertising: Which Keywords Lead to More Future Visits?

Many online shoppers initially acquired through paid search advertising later return to the same website directly. These so-called “direct type-in” visits can be an important indirect effect of paid search. Because visitors come to sites via different keywords and can vary in their propensity to make return visits, traffic at the keyword level is likely to be heterogeneous with respect to how much direct type-in visitation is generated. Estimating this indirect effect, especially at the keyword level, is difficult. First, standard paid search data are aggregated across consumers. Second, there are typically far more keywords than available observations. Third, data across keywords may be highly correlated. To address these issues, the authors propose a hierarchical Bayesian elastic net model that allows the textual attributes of keywords to be incorporated. The authors apply the model to a keyword-level data set from a major commercial website in the automotive industry. The results show a significant indirect effect of paid search that clearly differs across keywords. The estimated indirect effect is large enough that it could recover a substantial part of the cost of the paid search advertising. Results from textual attribute analysis suggest that branded and broader search terms are associated with higher levels of subsequent direct type-in visitation.

Zooming In on Paid Search Ads---A Consumer-Level Model Calibrated on Aggregated Data

We develop a two-stage consumer-level model of paid search advertising response based on standard aggregated data provided to advertisers by major search engines such as Google or Bing. The proposed model uses behavioral primitives in accord with utility maximization and allows recovering parameters of the heterogeneity distribution in consumer preferences. The model is estimated on a novel paid search data set that includes information on the ad copy. To that end, we develop an original framework to analyze composition and design attributes of paid search ads. Our results allow us to correctly evaluate the effects of specific ad properties on ad performance, taking consumer heterogeneity into account. Another benefit of our approach is allowing recovery of preference correlation across the click-through and conversion stage. Based on the estimated correlation between price-and position-sensitivity, we propose a novel contextual targeting scheme in which a coupon is offered to a consumer depending on the position in which the paid search ad was displayed. Our analysis shows that total revenues from conversion can be increased using this targeting scheme while keeping cost constant.

The Evolution of Internal Market Structure

We present a dynamic factor-analytic choice model to capture evolution of brand positions in latent attribute space. Our dynamic model allows researchers to investigate brand positioning in new categories or mature categories affected by structural change such as entry. We argue that even for mature categories not affected by structural change, the assumption of stable attributes may be untenable. We allow for evolution in attributes by modeling individual-level time-specific attributes as arising from dynamic means. The dynamic attribute means are modeled as a Bayesian dynamic linear model (DLM). The DLM is nested within a factor-analytic choice model. Our approach makes efficient use of the data by leveraging estimates from previous and future periods to estimate current period attributes. We demonstrate the robustness of our model with data that simulate a variety of dynamic scenarios, including stationary behavior. We show that misspecified attribute dynamics induce temporal heteroskedasticty and correlation between the preference weights and the error term. Applying the model to a panel data set on household purchases in the malt beverage category, we find considerable evidence for dynamics in the latent brand attributes. From a managerial perspective, we find advertising expenditures help explain variation in the dynamic attribute means.

Managing Blood Donations with Marketing

Blood banks rely on marketing to encourage donors to give blood. Many, if not most, blood banks in the United States are community-based not-for-profit organizations with limited marketing budgets. As a result, blood banks increasingly use novel and inexpensive online media, i.e., paid, owned, and earned POE media, in their marketing efforts. We propose a dynamic model to help blood bank marketing managers understand how blood donations can be managed via online POE media. We analytically characterize the optimal forward-looking paid media strategies, taking into account the asymmetric costs related to shortage and excess of blood, as well as the possibility of a cost-free target donation zone. We detail new advertising resource allocation rules for blood banks and show when traditional allocation recommendations do not apply. Additionally, we discover that under certain circumstances, owned/earned media activities hurt the blood banks performance, despite being predominantly free. We validate our analytical model by using daily donation data from a community-based blood bank and measure the effects of POE media activities on the level of blood donated.

Metrics for the New Internet Marketing Communications Mix

The Internet provides marketers with an expanded set of communications vehicles for reaching customers. Two of the most important and fast-growing elements of this new communications mix are online advertising and electronic word-of-mouth. While these vehicles provide new opportunities, they are also challenging marketers to understand how consumers respond and to develop new metrics for assessing performance. The purpose of this chapter is to review recent research developments in marketing that are most relevant to assessing the impact of these communications vehicles. The chapter first discusses the two major forms of Internet advertising, display ads (also known as banners) and paid search. (Paid search refers to the text ads served as sponsored links by Internet search engines.) The existing literature on banner ads provides a body of empirical findings as well as a set of methods that marketers can draw upon to assess performance. Research on paid search is still emerging but early work has developed approaches that marketers can use as the basis for suitable performance metrics. As a social medium, the Internet offers users new ways to communicate more easily and more extensively with others. Online communities, social networking sites, online referral programs, product reviews, and blogs all allow word-of-mouth to spread faster and farther than in the past. Research has shown how electronic records of online word-of-mouth (e.g., product reviews) can be connected, via models, to performance outcome variables such as product ratings and sales levels. Emerging work shows promise at identifying the particular members of online communities who are most likely to influence the actions of others, revealing opportunities for firms to manage

A New Method to Aid Copy Testing of Paid Search Text Advertisements

The authors propose a new approach to evaluate the perceptions and performance of a large set of paid search ads. This approach consists of two parts. First, primary data on hundreds of ads are collected through paired comparisons of their relative ability to generate awareness, interest, desire, action, and click performance. The authors use the Elo algorithm, a statistical model calibrated on paired comparisons, to score the full set of ads on relative perceptions and click performance. The estimated scores validate the theoretical link between perceptions and performance. Second, the authors predict the perceptions and performance of new ads relative to the existing set using textual content metrics. The predictive model allows for direct effects and interactions of the text metrics, resulting in a “large p, small n” problem. They address this problem with a novel Bayesian implementation of the VANISH model, a penalized regression approach that allows for differential treatment of main and interaction effects, in a system of equations. The authors demonstrate that this approach ably forecasts relative ad performance by leveraging perceptions inferred from content alone.