We present an inventory management solution for a non-stationary capacitated multi-echelon distribution network involving thousands of products. Assuming backlogged sales, we revisit and leverage the seminal multi-echelon inventory management results in the literature to establish the structural properties of the problem, and derive an efficient and practical solution method. In particular, we describe how the additive separability nature of the objective function can be used in a dynamic programming framework to circumvent the usual curse of dimensionality inherent to high-dimensional inventory management problems. We incorporate features such as asynchronous replenishment intervals at the different echelons, accounting of perishability, capacity constraints (both volume and flow), stochastic lead times, reverse logistics, supply risk and risk-pooling. Illustrative examples demonstrate some of these features. A version of this model was tested at a large grocery retailer in a two-layer network, resulting in a statistically significant increase of availability, sold units, and revenue, while lowering inventory volume at the hub.
SCOT, Amazon.com, New York, NY 10018, June 2022.