This article considers the supermarket manager's problem of forecasting demand for a product as a function of the product's attributes and of market control variables.To forecast sales on the stock keeping unit (SKU) level, a good model should account for product attributes, historical sales levels, and store specifics, and to control for marketing mix. One of the challenges here is that many variables which describe product, store, or promotion conditions are categorical with hundreds or thousands of levels in a single attribute. Identifying the right product attributes and incorporating them correctly into a prediction model is a very difficult statistical problem. This article proposes an analytical engine that combines techniques from statistical market response modeling, datamining, and combinatorial optimization to produce a small, efficient rule set that predicts sales volume levels. © 2006 Wiley Periodicals, Inc. and Direct Marketing Educational Foundation, Inc.