An efficient algorithm for large-scale dynamic assortment planning problems

Abstract

Single-period dynamic assortment planning involves the retailer’s selection of a set of products to offer and the determination of their initial inventory levels, considering stochastic demand and dynamic substitution. The objective is to maximize the expected revenue, subject to a capacity constraint. While existing heuristics are better suited to brick-and-mortar retailers with limited capacity, we introduce a novel heuristic designed to efficiently address the large-scale problems encountered by online retailers with high customer arrivals, a capacity of thousands of units, and extensive product variety. Through extensive simulation experiments across a range of customer types and demand scenarios, our method consistently delivers high-quality solutions while being significantly faster than existing approaches. We further validate our approach with a numerical example calibrated with real-world data from Wayfair, a major online home goods retailer. In this setting, our algorithm captures 90.16% of the expected revenue upper bound and delivers solutions in under 80 s. In contrast, existing approaches are unable to return solutions within a reasonable amount of time, highlighting the scalability and practical relevance of our method for large dynamic assortment planning problems.