LAUSR

Abstract Consumers often consider multiple alternatives from the same product category prior to making a purchase. Uncovering the predominant patterns of such co-considerations can help businesses learn more about the competitive structure of the market in the mind of the consumer. Extant research has shown that various types of online and offline consumer activity data (e.g., shopping baskets, search and browsing histories, social media mentions) can be used to infer product co-considerations. In this paper, we study a case of uncovering co-consideration patterns using a massive dataset of online price quote requests from U.S. auto shoppers. The main challenge we face is that, for privacy protection, no unique individual identifier (anonymous or otherwise) is contained in the data. Such a data deficiency prevents us from using existing methods such as affinity analysis for inferring co-considerations. However, by leveraging spatiotemporal patterns in the data, we manage to probabilistically uncover the predominant patterns of co-considerations in the U.S. auto market. As a validation and illustration of its usefulness, we embed the inferred market structure in a sales response model and show a substantial improvement in predictive performance.