Optimal Feature-Based Market Segmentation and Pricing

In this work, we study semi-personalized pricing strategies where a seller uses features about their customers to segment the market, and customers are offered segment-specific prices. In general, finding jointly optimal market segmentation and pricing policies is computationally intractable, with practitioners often resorting to heuristic segment-then-price strategies. In response, we study how to optimize and analyze feature-based market segmentation and pricing under the assumption that the seller has a trained (noisy) regression model mapping features to valuations. First, we establish novel hardness and approximation results in the case when model noise is independent. Second, in the common cases when the noise in the model is log-concave, we show the joint segmentation and pricing problem can be efficiently solved, and characterize a number of attractive structural properties of the optimal feature-based market segmentation and pricing. Finally, we conduct a case study using home mortgage data, and show that compared to heuristic approaches, our optimal feature-based market segmentation and pricing policies can achieve nearly all of the available revenue with only a few segments