Abstract: In this thesis, we study new applications of revenue management in three different online platforms and develop models and techniques in these settings. In the first study, we design a robust inventory-sharing policy for an omnichannel platform, selling their products through online and physical outlets. We build a two-stage stochastic programming model based on the proposed policy, coordinating fulfillment and transshipment decisions to minimize the retailer's total expected cost. We compare our policy with another standard inventory-sharing policy using real data from a well-known retailer.
In the second study, we investigate an online marketplace platform (OM), which determines the assortment of sellers displayed on its platform and charges commission fees on the sales of selected sellers to maximize its expected revenue. Each selected seller, to maximize its profit, then decides the price of its product in response to other sellers' behavior on the platform. To model the interactions between the OM and the sellers, we develop a Stackelberg game and consider two types of competition among the sellers: Bertrand and Cournot. We provide characterizations and insights of the OM's optimal assortments and commission fee decisions for both types of competitions. We also introduce a monopoly model, in which the OM makes pricing decisions while guaranteeing that the selected sellers' profits are no less than their target profits. We study how the proposed oligopoly and monopoly models affect the OM's profit and total market share.
In the third study, we integrate dynamic pricing and assortment with a dynamic routing problem allowing for deadline decisions in a unified framework for an online ride-sharing problem. This framework considers a ride-sharing platform that dynamically makes routing, pricing, and assortment decisions to offer vehicle options to different requests over a finite horizon time in a rural area. At most one request stochastically enters the platform at each period, and the request then selects one option or leaves the platform with no selection. The platform's decisions will impact both the present and future requests' decisions in the current period. To this end, we develop an approximation dynamic programming algorithm and derive a policy that is able to make all the decisions in real-time. In a comprehensive numerical study, we compare our policy with four benchmark policies and show that it outperforms the benchmark policies significantly.
