Hanmi Science

Hanmi Science, a Korean pharmaceutical enterprise, faced the core challenge of OTC drug demand forecasting in a sector where miscalculation carries a dual penalty: stockouts deny patients access to treatment, while overstock triggers mandatory disposal costs due to pharmaceutical expiration requirements. Demand for OTC drugs is shaped by interlocking variables — disease incidence rates, seasonal patterns, competitor supply disruptions, and prescription behavior shifts — none of which traditional methods could model in combination. Forecasting relied on historical sales data and planners' institutional knowledge, manually assembled in Excel. When COVID-19 disrupted baseline demand patterns, the fragility of experience-dependent planning became acute, exposing the full operational and patient-care cost of the status quo.

Hanmi Science deployed Deepflow Forecast, an AI-based demand forecasting platform developed by Impactive AI and purpose-built for the pharmaceutical industry. The implementation covered more than 60 OTC drug SKUs and ingested over 6 million data points in real time. The platform runs 224 disease-specific prediction models, linking patient-number forecasts directly to downstream drug demand — achieving 80.1% sales volume forecast accuracy and 96.5% patient-number prediction accuracy across those disease models. Machine learning algorithms identify correlations across disease occurrence patterns, seasonality, weather variables, and prescription trends — relationships too complex for conventional statistical approaches. Critically, the system incorporates explainable AI (XAI), surfacing the specific variables and their weighted influence behind each forecast, so procurement and supply chain managers can interpret predictions with confidence rather than treating AI output as a black box.

The headline outcome was a 55.1% reduction in monthly inventory costs, achieved by aligning stock levels to AI-generated demand signals rather than conservative manual estimates. The out-of-stock rate fell 22.6%, directly reducing missed treatment opportunities across the OTC range, while excess inventory declined 32.5%, sharpening expiration-date management and minimizing pharmaceutical disposal write-offs. On the operational side, time spent on routine forecasting and order-calculation tasks dropped 80%, replacing manual Excel workflows with automated outputs. Key metrics:

• Monthly inventory cost reduction: 55.1%
• Out-of-stock rate reduction: 22.6%
• Excess inventory rate reduction: 32.5%
• Routine forecasting task time reduction: 80%

• Scope the initial deployment to a defined, measurable product set — Hanmi Science's 60+ OTC SKU pilot produced verifiable results before broader rollout, lowering organizational risk in a regulated industry.
• Disease occurrence modeling amplifies forecast accuracy: linking patient-number predictions to drug demand (96.5% across 224 diseases) produces precision that single-variable historical models cannot approach.
• Explainable AI is non-negotiable in pharma supply chain — procurement managers require variable-level attribution before acting on forecasts that affect patient access and regulatory compliance.
• Automating routine forecasting frees planners to focus on exception management rather than spreadsheet maintenance, compounding the operational value of the system over time.