New Challenges in Revenue Management for Modern Marketplaces
A core challenge in revenue management is to determine the optimal assortment to offer customers while accounting for various constraints, such as limited capacities. The rapid evolution of marketplaces has introduced new complexities to this problem. Sellers now have flexible products, allowing multiple alternatives to fulfill a single request. Revenue streams have also become more diverse, extending beyond direct sales to include other monetization strategies. Additionally, online marketplaces facilitate collaboration between platforms and multiple merchants but pose new challenges in allocating market share. In this dissertation, we introduce new models to address revenue management challenges in each of these areas and develop efficient algorithms with performance guarantees. First, we study network revenue management problems with flexible products, where each product request can be fulfilled using multiple possible combinations of resources. Under limited resource capacities, the objective is to find a policy to determine which assortment to offer to each customer that maximizes the total expected revenue over the selling horizon, while deferring the decision on which resource combination to use to fulfill the customer purchases until the end of the selling horizon. We develop an approximate policy based on approximate dynamic programming that gives a constant-factor performance guarantee. Second, we explore dynamic assortment optimization problems with multiple revenue streams, where platforms generate income from both direct sales and customer clicks on sponsored products. Under limited product capacities, the goal is to find a policy to determine which assortments to offer and sponsor to each customer that maximizes total expected revenue from both sales and clicks over the selling horizon. Based on the fluid approximation of the problem, we develop approximate policies with constant-factor guarantees and asymptotic optimality when customer choices follow the multinomial logit choice model. Lastly, we study assortment optimization under market share constraints, where the platform must select an assortment from a set of product classes while ensuring that each product class represented in the offered assortment meets a minimum market share requirement. When customer choices follow the multinomial logit model, we analyze both the randomized and deterministic variants of this problem. We develop a unified approximation framework that provides a fully polynomial-time approximation scheme (FPTAS) for the randomized case and a $1/2$-approximate pseudo-polynomial algorithm for the deterministic case.