Assembling the crystal ball : using demand signal repository to forecast demand

Abstract

Improving forecast accuracy has positive effects on supply chain performance. Forecast accuracy can reduce inventory levels, increase customer service levels and responsiveness, or a combination of the two. However, the further upstream in the supply chain, the more difficult it becomes to forecast accurately. Demand for consumer products might be subject to factors that are hard to identify and quantify. One way to overcome this is to observe external factors or predictors that might help explain demand. The purpose of this thesis is to explore the factors that potentially influence the demand of a fast-moving consumer product (bottled water), and build a demand signal repository for these factors to help the manufacturer generate more accurate forecasts. We identified more than 30 such factors that might affect demand, using interviews and industry research. We tested more than 200 causal models of the relationship between observed demand and the predicting factors. The resulting model explained almost 60% of demand for two out of three customers using daily buckets and over 85% using weekly buckets compared to less than 50% using time-series techniques. Using the results of this extensive analysis, we propose a new forecasting model. We also identified additional factors that could not be included this analysis due to the lack of data; adding these to the model may further improve the forecast accuracy.