AI / Operations
AI-Powered Inventory Forecasting for F&B Chain
Built a demand forecasting system that cut food waste by 30% across 12 restaurant locations using historical sales and weather data.
Context
A multi-location F&B group in Dubai was losing 15-20% of perishable inventory weekly. Each branch manager ordered based on gut feeling, leading to inconsistent stock levels and significant waste.
Problem
No centralized view of demand patterns. Branch managers over-ordered to avoid stockouts, creating a culture of waste. Seasonal fluctuations and event-driven spikes were entirely unaccounted for.
Approach
Aggregated 18 months of POS data across all locations. Enriched with weather, holidays, and local events. Trained per-category XGBoost models with rolling 7-day forecasts. Built a dashboard for managers to review and adjust orders.
Build
FastAPI backend serving predictions via REST. PostgreSQL for historical data and forecast storage. Next.js dashboard with per-branch and per-category views. Automated daily model retraining pipeline.
Result
30% reduction in food waste within the first quarter. Managers reported spending 60% less time on ordering decisions. The system paid for itself in under 2 months.
What This Proves
AI doesn't need to be exotic to deliver massive ROI. The right model on clean data, paired with a usable interface, beats complex solutions that nobody adopts.