Optimizing Retail Assortments

Abstract

Retailers face the problem of finding the assortment that maximizes category profit. This is a challenging task because the number of potential assortments is very large when there are many stock-keeping units (SKUs) to choose from. Moreover, SKU sales can be cannibalized by other SKUs in the assortment, and the more similar SKUs are, the more this happens. This paper develops an implementable and scalable assortment optimization method that allows for theory-based substitution patterns and optimizes real-life, large-scale assortments at the store level. We achieve this by adopting an attribute-based approach to capture preferences, substitution patterns, and cross-marketing mix effects. To solve the optimization problem, we propose new very large neighborhood search heuristics. We apply our methodology to store-level scanner data on liquid laundry detergent. The optimal assortments are expected to enhance retailer profit considerably (37.3%), and this profit increases even more (to 43.7%) when SKU prices are optimized simultaneously.

Keywords: retail assortments, optimization, product attributes, substitution, similarity, endogeneity, heuristics, micromarketing, pricing, hierarchical Bayes, Gibbs sampling