Built an ML-powered demand forecasting system that reduced stockouts by 38% and excess inventory by 22% across 450 retail locations.
The retailer managed inventory across 450 stores using spreadsheet-based forecasting updated weekly. Seasonal demand shifts, regional preferences, and supply chain disruptions created chronic stockouts in high-demand stores while surplus accumulated in others. Manual adjustments by store managers introduced compounding errors.
We developed a demand forecasting pipeline using time-series models trained on 4 years of POS data, incorporating weather, holiday calendars, and regional demographic signals. A RAG-based interface lets store managers query forecasts and explain anomalies. Weekly automated replenishment recommendations integrate directly with the ERP system.
38%
Stockouts reduced
22%
Excess inventory reduced
94.2% at SKU-store-week level
Forecast accuracy
450
Stores served