The Great Equalizer? An Empirical Study of Consumer Choice at a Shopbot

Shopbots are computer agents that aid consumers by comparing prices across online stores. A consumer visits a shopbot web site, inputs a product to search for, such as a book, and then the shopbot automatically queries available online stores, and tabulates and presents the results to the consumer. Consumers consider a subset of these stores by clicking upon a hyperlink displayed with the offer. This evoked consideration set is important to shopbot owners because shopbot owners are generally compensated directly for consideration as opposed to purchase, in the industry this is referred to as payfor-click. In our research we estimate a multivariate probit model to predict a consumer’s evoked consideration set. Our utility model includes the product attributes, retailer attributes, and the position of the alternative in the table shown to the consumer. In addition, we include covariates for learning over time, the coefficient of variation of the prices, and the list price of the product to explain consumerspecific variation in the size of the evoked consideration set. Our model is calibrated on clickstream data collected at a major shopbot over a 23-month period. A weakness of the traditional multivariate probit model is that the marginal distribution of the number of items in the evoked consideration set is a function of the latent utility which predicts which offers will be chosen. Unfortunately, the standard model provides a poor approximation to this distribution, so we propose a model for the number of items that will be selected and then conditional upon this value we predict which offers will be chosen. Our findings indicate this model better captures consumer choice. Additionally we find that positioning and advertising have substantial impact on which items a consumer will choose. The size of the evoked consideration decreases with increasing consumer experience, larger offer sets leads to less search, and more expensive products and offer sets with more variation in prices result in larger evoked consideration sets. These results suggest that shopbot owners can influence the composition of a consumer’s evoked consideration set — and therefore the shopbot’s revenue — by changing the order of offers, the number of offers shown, and the display of logos next to a particular retailer’s name.