Modeling Cross-Category Purchases in Sponsored Search Advertising

Abstract:     
A unique aspect of sponsored search advertising is that it allows firms to track what products consumers initially search for and what products they eventually buy after visiting their website - something that is typically hard to acquire in physical world settings. Based on a unique and unprecedented data set that contains information on what keyword advertisements induced consumers to arrive on a firm's website and what products they eventually bought in that session, we build a model to map consumers' search-to-purchase behavior in the online world. We analyze the relationship between consumers' search for products in a specific category and their propensity to buy products within that category. In addition, we examine spillovers from search resulting in purchases across other product categories in the same session. The model is estimated on purchases from four categories (bath, bedding, kitchen, and home decor) using a unique 6 month panel data set of several hundred keywords collected from a large nationwide retailer that advertises on Google. Our model allows us to decompose the latent utility of a product category into intrinsic utility from that product category and extrinsic utility from joint purchases of that category with other ones. With regard to the intrinsic utilities, we find that there is a considerable amount of spillovers between the initial search and the final purchase behavior such that consumers who start a search for a product in one category purchase products from a different category, in addition to purchasing from the original category they searched for. However, such search-purchase spillover effects accruing from a given advertisement are not necessarily symmetric between any two given product categories. With regard to the extrinsic utilities, we see evidence of positive cross-category interdependence for retailer-specific keywords. In addition, we find that brand-specific and generic keywords are less likely to induce cross-category purchases. Based on our empirical estimates, we conduct counter factual experiments to derive insights into changes in overall profitability when firms were to engage in personalized price discounts and targeting via sponsored search advertisements. Our analyzes suggest that retailers can benefit from providing personalized targeting and recommendations based on consumer search and cross-category purchase patterns observed during sponsored search advertising.

Keywords: Online advertising, Search engine marketing, Sponsored search, Conversion rates, Electronic commerce, Spiilovers, Hierarchical Bayesian estimation

Working Paper Series