Pharmaceutical demand forecasting is often constrained by limited visibility into true market demand at the country level and other phenomena, which reduces the effectiveness of planning, increases costs, and potentially impacts companies’ ability to meet patient health needs. This project addresses the question of how to extract meaningful demand signals and identify the most appropriate forecasting models across different product portfolios and countries. A scalable, data-driven forecasting framework was developed integrating demand decomposition, statistical methods, and machine learning models to process up to 36 months of historical data and benchmark performance of internal forecasts. Results show that while overall forecast accuracy is generally strong, significant improvements can be achieved for low-accuracy products, with alternative models delivering substantial gains and reducing mis-forecasted volumes. The findings support a targeted deployment approach, recommending that our sponsor company apply the framework to identify high-impact opportunities, combine model outputs with local expertise, and tailor forecasting strategies based on product maturity and data availability.
