Demand Forecasting for Omnichannel Inventory Management

Abstract

We study the problem of fine-grained demand forecasting (product-location-channel level) for omnichannel retailers to support inventory management. We go over necessary developed forecasting components, e.g., modeling demand for different tasks and horizons, generating forecasts across product-location-channel combinations while leveraging cross-series information, and effectively evaluating forecasts, in this sparse observation setting. We present an end-to-end solution from raw transaction data to consumable forecast outputs for down-stream tasks, and demonstrate forecasting results, evaluations, and analyses using real, large-scale data from an omnichannel retailer.