Empirical methods for two-echelon inventory management with service level constraints based on simulation-regression

We present a simulation-regression based method for obtaining inventory policies for a two-echelon distribution system with service level constraints. Our motivation comes from a wholesale distributor in the consumer products industry with thousands of products that have different cost, demand, and lead time characteristics. We need to obtain good inventory policies quickly so that supply chain managers can run and analyze multiple scenarios effectively in reasonable amount of time. While simulation-based optimization approaches can be used, the time required to solve the inventory problem for a large number of products is prohibitive. On the other hand, available quick approximations are not guaranteed to provide satisfactory solutions. Our approach involves sampling the universe of products with different problem parameters, obtaining their optimal inventory policies via simulation-based optimization and then using regression methods to characterize the inventory policy for similar products. We show that our method obtains near-optimal policies and is quite robust.