Using imperfect advance demand information in forecasting

In this paper we consider the demand forecasting problem of a make-to-stock system operating in a business-to-business environment where some customers provide in- formation on their future orders, which are subject to changes in time and hence consti- tuting imperfect advance demand information (ADI). The demand is highly volatile and non-stationary, not only because it is subject to seasonality and changing trends, but also because some individual client demands have significant in°uence on the total demand. In such an environment, traditional forecasting methods may result in highly inaccurate fore- casts, since they are mostly developed for the total demand based only on the demand history, not making use of demand information and ignoring the effects of individual order patterns of the customers. We propose a forecasting methodology that makes use of in- dividual ordering pattern histories of the product-customer combinations and the current build-up of orders. Moreover, we propose making use of limited judgmental updates on the statistical forecasts prior to the use of ADI. In our application at a company that produces dairy ingredients this method resulted in significant improvements in the forecasts and as a result, required safety stocks are reduced by 25% by incorporating ADI in forecasts, and we demonstrated that the reduction amounts to 37% in case judgmental updates are also utilized.