Scalable Demand Transfer Estimation Improves Assortment Optimization.

Retailers often face the challenge of optimizing product assortments, which traditionally involves complex demand forecasting for every possible item combination. This new research proposes an alternative, more efficient approach. It focuses on estimating "Demand Transfer" (DT) coefficients, which quantify how much demand for a specific item shifts to other products if the original item is unavailable. The core innovation is a scalable method to compute these DT coefficients across vast product universes, potentially involving over a million items. By combining independent item demand forecasts with these transfer adjustments, the system can more accurately predict demand while considering customer substitution patterns. Experimental results, using both simulated and historical transaction data, indicate that this procedure accurately estimates underlying DT coefficients and significantly enhances demand forecasting, provided certain assumptions about customer substitution behavior hold true.