A Hierarchical Time Series Forecasting Top-Down Approach for Improving Inventory Placement Demand Forecast

Inventory Placement is the process by which e-commerce retailers put items into Fulfillment Centers (FC) so that they are ready to fulfill the customer demand. Inventory Placement is crucial for shipping speed and for this motivation it aims to place the items the customers want as close to them as possible. The e-commerce retailer considered in this work doesn’t inject items directly to the FCs, but rather it uses another type of warehouse called Warehouse Router (WR) whose task is to receive and route Inventory to different Fulfillment Centers. Inventory Placement, in order to move items to different FCs, requires to have trucks scheduled. Truck Scheduling requires to know in advance what volumes needs to be shipped from each Warehouse Router to each Fulfillment Center served by it. The way to know this, is by using a Machine Learning model to predict the volume that will need to be shipped in each WR->FC lane. In this work are proposed possible approaches to improve the model predictions.