Dynamic Pricing and Inventory Management in the Presence of Online Reviews

We study the joint pricing and inventory management problem in the presence of online customer reviews. Customers who purchase the product may post reviews that would influence future customers' purchasing behaviors. Under the common practice of customer-generated reviews on e-commerce platforms, rigorous investigation of their operational implications offers valuable insights and guidance for both the research community and practitioners. We develop a stochastic joint pricing and inventory management model to characterize the optimal policy in the presence of online reviews. We show that a rating-dependent base-stock/list-price policy is optimal. Interestingly, the inventory dynamics of the firm do not influence the optimal policy as long as the initial inventory is below the initial base-stock level. Hence, we can reduce the dynamic program that characterizes the optimal policy to one with a single-dimensional state-space (the aggregate net rating). The presence of online reviews gives rise to the trade-off between generating current profits and inducing future demands, thus having several important implications upon the firm's operations decisions. First, online reviews drive the firm to deliver a better service and attract more customers to post a review. Hence, the safety-stock and base-stock levels are higher in the presence of online reviews. Second, the evolution of the aggregate net rating process follows a mean-reverting pattern: When the current rating is low (resp. high), it has an increasing (resp. decreasing) trend in expectation. Third, although myopic profit optimization leads to significant optimality losses in the presence of online reviews, balancing the current profits and near-future demands suffices to exploit the network effect induced by online reviews. We propose a dynamic look-ahead heuristic policy that well leverages this idea and achieves small optimality gaps which decay exponentially in the length of the look-ahead time-window