AI-Enabled Demand Forecasting for Over-the-Counter Drug Sales in Urban Pharmacies

Abstract

Urban pharmacies operate in a high-velocity retail–health environment where short product life cycles, seasonal illness waves, and localized socioeconomic patterns make demand volatile—especially for over-the-counter (OTC) medicines such as analgesics, antipyretics, cough and cold preparations, antacids, and nutritional supplements. Traditional rule-of-thumb and spreadsheet forecasts often underperform during abrupt shifts (e.g., weather swings, school reopenings, pollution spikes, or festival seasons), causing stockouts, wastage from expiries, and revenue loss. This manuscript proposes a practical, end-to-end framework for AI-enabled demand forecasting tailored to urban pharmacies. The approach integrates point-of-sale (POS) histories with exogenous signals—calendar/holiday effects, promotions, meteorological readings, epidemiological proxies, mobility indicators, and neighborhood demographics—within a hierarchy of models ranging from seasonal ARIMA and gradient-boosted trees to deep learning architectures such as LSTMs and Temporal Fusion Transformers (TFT). We outline design choices for feature engineering, hierarchical reconciliation across store → category → SKU, backtesting with rolling origins, and business-centric metrics (WAPE, sMAPE, service level, stockout rate, and waste).

To ground the method in practice, we detail a clinical research–style operational study situated in real pharmacy workflows, considering data governance, patient safety, and ethical use of retail health data. An illustrative results section (based on a realistic pilot evaluation design) shows how a multi-model stack can reduce weighted absolute percentage error (WAPE) by 25–40%, increase on-shelf availability, and lower expiries. We conclude with deployment guidelines—MLOps, model monitoring, exception handling—and a candid scope and limitations discussion relevant to urban markets in India and comparable settings.