Product sales forecasts are used throughout at a major retailer such as Target Corporation, supporting inventory replenishment, labor scheduling, merchandise assortment planning and so on. This talk describes an automated statistical forecasting system developed at Target that seeks comprehensively to address the problem of product sales forecasting, preparing daily forecasts with horizons up to 1 year for hundreds of thousands of products available in over 1,800 stores and on-line. The forecasting system relies on a multiplicity of models, and a forecast ensemble mechanism is used to combine component forecasts. The ensemble mechanism operates at fine-grained level, furnishing a "mixture of experts" in which component forecast models may be specialized for the demand patterns associated with particular parts of the product range or times of the year. In addition to an overview of the component models and the ensemble mechanism, we will also discuss the software required to deploy them at scale in a mission-critical commercial application, as well as pragmatic issues such as forecast performance measurement and the provision of actionable forecasts to users throughout the business.