Jiani Xue, PhD Candidate, The Wharton School; Stefano Putoni, Marketing, The Wharton School; Barbara Mellers, Psychology, University of Pennsylvania
Abstract: Ensemble models, a class of machine learning algorithms that combine the predictions of multiple algorithms to form more accurate predictions, are widely used in marketing applications. This research explores ways to enhance the adoption and perceived accuracy of these models. Thirteen experimental studies (ten in the main paper and three in the Web Appendix) collectively demonstrate that consumers have an intuitive grasp of the “wisdom of the crowd,” and they perceive machine ensembles as more accurate than single algorithms. Framing algorithms as a collection of models rather than a single algorithm boosts perceived accuracy, purchase intent, and consumer preferences across different domains. Results also indicate that preferences are driven by a mistaken belief that ensembles primarily reduce bias rather than noise. Consumers seem to recognize that the diversity and relevance of the datasets on which ensembles are trained influence bias reduction. Taken together, these experiments shed light on a novel psychological process shaping AI adoption and reveal practical ways to promote the adoption of ensemble methods.