Reference: Klein, D. A. & Shortliffe, E. H. Integrating Artificial Intelligence and Decision Theory to Forecast New Products. Milano, Italy, 1990.
Abstract: Established forecasting techniques generally are unsuitable for forecasting sales of new products, because most such techniques require the availability of directly pertinent historical data (e.g., previous sales of the product) to produce a forecast. In this paper, we present FORECASTER, a methodology and a supporting computer program that forecasts sales of new products by predicting the purchasing behaviour of individual customers. FORECASTER employs a novel integration of production rules and decision-theoretic models to provide a customer-specific forecast for all the products in a particular market simulataneously. Although motivated by the requirements of forecasting sales of new products, FORECASTER also can be employed in the context of forecasting sales of mature products to confirm forecasts produced by established techniques, and to increase the resolution of such forecasts. Our methodology suggests the feasibility of managing large collections of loose assumptions in forecasting new products, and, more generally, that systhesis of techniques from artificial intelligence and from decision theory potentially provides a basis for increasing the capabilities of current forecasting tools.