Accurate demand forecasting is critical for supply chain efficiency, especially for the pharmaceutical (pharma) supply chain due to its unique characteristics. However, limited data have prevented forecasters from pursuing advanced models. Such problems exist even when long history of demand data is available because historical data in the distant past may bring little value as market situation changes. In the meantime, demands are also affected by many hidden factors that again require a large amount of data and more sophisticated models to capture. We propose to overcome these challenges by a novel demand forecasting framework which “borrows” time series data from many other products (cross-series training) and trains the data with advanced machine learning models (known for detecting patterns). We further improve performance of the cross-series models through various “grouping" schemes, and learning from non-demand features such as downstream inventory data across different products, information of supply chain structure, and relevant domain knowledge. We test our proposed framework with many modeling possibilities on two large datasets from major pharma manufacturers and our results show superior performance. Our work also provides empirical evidence of the value of downstream inventory information in the context of demand forecasting. We conduct prior and post-hoc field work to ensure the applicability of the proposed forecasting approach.
All Science Journal Classification (ASJC) codes
- Management Science and Operations Research
- Industrial and Manufacturing Engineering
- Management of Technology and Innovation