Predicting "Design Gaps" in the Market: Deep Consumer Choice Models under Probabilistic Design Constraints

Abstract

Predicting future successful designs and corresponding market opportunity is a fundamental goal of product design firms. There is accordingly a long history of quantitative approaches that aim to capture diverse consumer preferences, and then translate those preferences to corresponding "design gaps" in the market. We extend this work by developing a deep learning approach to predict design gaps in the market. These design gaps represent clusters of designs that do not yet exist, but are predicted to be both (1) highly preferred by consumers, and (2) feasible to build under engineering and manufacturing constraints. This approach is tested on the entire U.S. automotive market using of millions of real purchase data. We retroactively predict design gaps in the market, and compare predicted design gaps with actual known successful designs. Our preliminary results give evidence it may be possible to predict design gaps, suggesting this approach has promise for early identification of market opportunity.