Improving Forecasting of Censored Demand for Bread using Footfall Data in a Norwegian Retail Chain
Abstract
The retail business for fresh food products is characterized by short shelf life and low-to-moderate profit margin, leading to frequent stockouts. Forecasting demand for such products faces difficulty due to the presence of censored demand, where customers find empty shelves and the demand fails to register as sales. Footfall data, or the number of visitors coming to a store can be used to estimate censored demand, and this thesis presents such an estimation procedure, and evaluates the potential benefits of this approach mathematically. The proposed model exploits footfall as additional data using a maximum likelihood estimator (MLE) principle that requires historical sales and on-hand inventory data. The findings show that the footfall-based model does not significantly improve profits, forecasting error (MSE) and fraction of optimal order, in steady state for a single product. However, it provides qualitative benefits of establishing upper limit for demand estimation, a restart mechanism for discontinued products and applicability to multiple stores and new launched products.