How OgenTech partnered with one of world's largest Health Management Organization (HMO) to deploy an AI-powered foundation model across logistical centers and hundreds of pharmacies, transforming drug inventory planning for routine operations and crisis scenarios alike.
The world healthcare systems faces supply chain disruptions at every scale — from global pandemics, raw material shortagas, supply chain disruptions and regional conflicts to localized facility constraints. Over the past decade, events have repeatedly destabilized drug supply equilibrium: extreme demand surges, supplier shortages, delivery delays, and facility closures that reduce pharmacy capacity to protected-space-only operations.
Traditional inventory management policies responded with blunt instruments — across-the-board safety stock increases of one month — without the granularity to differentiate between locations, item categories, or the specific nature of each disruption.
Global pandemics, regional conflicts, and local logistics failures — all impacting the same supply chain simultaneously.
Traditional systems took weeks to recalibrate after disruptions, using one-size-fits-all policies that couldn't differentiate between locations.
Each of 1,000 pharmacy endpoints carries thousands of products with unique demand patterns — many slow-moving or intermittent.
OgenTech and the HMO's Logistics & Infrastructure Division developed and refined state-of-the-art statistical and analytical models for drug consumption forecasting — among the most advanced in the market. These models served as the foundation for understanding demand patterns across the pharmacy network.
With increased computing power and data richness, OgenTech developed and deployed an AI model using consumption data and associated features. The model learns pharmacist adjustments to understand real-world constraints. Initial pilot across 10 pharmacies expanded to 114, then 600+ — demonstrating significantly higher forecast accuracy, better trend detection, and reduced pharmacist workload.
The next evolution — a foundation model trained on 6 billion data points across 10 years, incorporating demographics, geography, weather, prescriptions, and crisis events. This "logistics brain" develops awareness of macro events and generates forecasts that remain valid in both routine and emergency scenarios, expanding to all 1,000 pharmacy endpoints.
The system combines multiple AI approaches, each selected for specific forecasting challenges within the pharmacy supply chain.
Recurrent networks connected via Temporal Fusion Transformer architecture, leveraging both static features (pharmacy characteristics) and dynamic features (events, seasons, policies) with built-in interpretability.
For slow-moving items consumed at non-continuous intervals, ensemble algorithms including gradient boosting and SVM handle the discrete, irregular patterns that challenge continuous forecasting methods.
Trained on ~6 billion data points using 8 parallel A100/H100 GPUs on AWS, with 50-100 features per observation. The model undergoes periodic retraining as new data accumulates.
The trained model connects to an NLQ interface enabling stakeholders to ask questions in natural language about the model's predictions, investigate scenarios, and explore "what-if" situations.
The AI model achieves significantly higher prediction accuracy compared to previous statistical approaches. It better understands consumption changes throughout the year and more precisely forecasts demand trends — including for challenging slow-moving items.
By learning from pharmacist corrections and real-world constraints, the system delivers practical plans that reduce the time pharmacists spend on inventory management while improving item availability for patients.
The foundation model introduces "macro-awareness" — understanding how national and global events cascade through the supply chain. When a crisis is declared, the model generates context-aware predictions accounting for facility restrictions, behavioral changes, and supply disruptions.
This capability also serves as a powerful investigation tool, enabling analysis of drug shortage impacts, prescription fulfillment patterns, and vaccination rates without requiring separate data collection and analysis efforts.
"The AI model's ability to learn from pharmacist adjustments and real-world constraints represents a fundamental shift — from reactive inventory policies to proactive, intelligent supply chain management."
Joint Innovation Partnership