Assortment Planning for Recommendations at Checkout under Inventory Constraints
29 Pages Posted: 17 Oct 2016 Last revised: 20 Aug 2018
Date Written: October 16, 2016
In this paper, we consider a dynamic personalized assortment planning problem under inventory constraints. In particular, each arriving customer is defined by a primary item of interest (which can be interpreted as her type). As long as the item is available, the decision-maker can then offer the customer a personalized assortment of add-on items based on her type, at potentially discounted prices. This problem, which is motivated by a new checkout recommendation system at Walmart’s online grocery, not only finds practical applications but also serves as a general framework that incorporates many operations problems as special cases (e.g., personalized assortment planning , single-leg booking [2, 15], and online matching with stochastic rewards [21, 23]). In our personalized assortment planning problem, multiple item prices (discounted or not), combined with customer withdrawal when their initially selected items stock out, pose additional challenges for the development of an online policy with performance guarantees. We overcome these challenges by introducing the notion of an inventory protection level in expectation, and derive an algorithm with a 1/4−ε competitive ratio guarantee under adversarial arrivals, unifying the existing results.
Keywords: revenue management, personalized recommendation, assortment planning, online algorithms, competitive ratio, protection level
Suggested Citation: Suggested Citation