Leveraging Comparables for New Product Sales Forecasting

Many firms regularly introduce new products. Before the launch of any new product, firms need to make various operational decisions, which are guided by the sales forecast. The new product sales forecasting problem is challenging when compared to forecasting sales of existing products. For existing products, historical sales data gives an indicator of future sales, but this data is not available for a new product. We propose a novel sales forecasting model that is estimated with data of comparable products introduced in the past. We formulate the problem of clustering products and fitting forecasting models to these clusters simultaneously. Inherently, the model has a large number of parameters, which can lead to an overly complex model. Hence, we add regularization to the model so that it can estimate sparse models. This problem is computationally hard, and as a result, we develop a scalable algorithm that produces a forecasting model with good analytical guarantees on the prediction error. In close collaboration with our industry partner Johnson & Johnson Consumer Companies Inc., a major fast moving consumer goods manufacturer, we test our approach on real datasets, after which we check the robustness of our results with data from a large fast fashion retailer. We show that, compared to several widely used forecasting methods, our approach improves MAPE and WMAPE by 20-60% across various product segments compared to several widely used forecasting methods. Additionally, for the consumer goods manufacturer, we develop a fast and easy-to-use Excel tool that aids managers with forecasting and making decisions before a new product launch