Forecasting Model for Sporadic Distributor Based Markets

With the evolution of technology, more data became available to observe consumer purchase patterns. Traditional forecast methods used to rely on only shipment history. Nonetheless, due to  the accessibility of  consumer  data, a forecast process that integrates downstream flow has shown good results in improving the forecast accuracy in supply chains. In this research we investigate the benefits and validity of linking downstream distributor data in a fast-moving consumer goods company to improve forecast accuracy for intermittent demand. We used multi-tier regression analysis to link distributor sellout data to a retailer in order to predict shipment volume, and then performed a comparison analysis using the Croston method. We concluded that using multi-tier regression analysis has made a slight improvement on an aggregated level; however, the success of this method is subject to data availability that could be a constraint in certain situations. The Croston method has shown significant improvements at the item level and helped to better stabilize the forecast, yet it doesn't consider downstream data. We show a comparison between the two methods, and how to primarily link distributor data in the company's forecast to improve forecast accuracy.