The leading CPG company supplied their products to a large number of retail stores across Columbia. However, the stores lacked a common methodology for sell-out forecasting. In addition, the sell-in order quantity estimation process was quite basic and did not account for any short-term demand sensing components. Resultantly, stores were either understocked generating poor customer experience or overstocked with products expiring on the shelves. Another concern was the lack of a scalable method to identify whitespaces, i.e. the absence of a product at a store that has high sell-out potential.
We developed an ML-based order recommendation engine for the customer. The solution pulled data from SAP and CEN (3rd Party Syndicated data) for preprocessing and extracting features such as sales, products, POS, demographics, and others using Python creating a master pipeline. We then identified whitespaces and the current portfolio by generating a list of active products from the last six months of POS data for each store. Based on the collected data, the system used two primary ML techniques: a hybrid recommendation model for every store to suggest new products and a sell-out volume forecasting model to estimate optimal product quantity.