Offering Online Recommendations to Impatient, First-Time Customers with Conjoint Based Segmentation Trees

Online consumers are known to be impatient. It is therefore critical for merchant websites to offer relevant recommendations quickly and with minimal customer input to their online visitors. Recommending pertinent items to first-time visitors, however, is particularly challenging. In this paper, we show how conjoint analysis can be leveraged to design a recommender system that elicits customers’ preferences based on simple demographics and product usage questions. We compare three algorithms, cluster-classification, Bayesian treed regression and stepwise-conjoint regression to develop an optimal sequence of questions. In an empirical study, stepwise-conjoint regression outperformed individual-based conjoint analysis to elicit individual preferences, both in term of efficiency and predictive accuracy.