Demand Forecasting Engine

Overview

Built a demand forecasting system that combines historical sales, seasonality, promotions, and external signals to predict demand at SKU-location level....

The Problem

A retail chain was either overstocked or out of stock. Millions locked up in dead inventory. Missed sales from stockouts. No visibility into what would sell where and when.

The Solution

Built a demand forecasting system that combines historical sales, seasonality, promotions, and external signals to predict demand at SKU-location level.

Architecture

• ETL pipeline for multi-source data integration
• Feature store for time-series signals
• Ensemble model with multiple horizons
• Automated retraining and model monitoring
• Planning interface for buyers
• Integration with inventory management system

Business Impact

• 28% reduction in inventory holding costs
• Stockout rate dropped from 8% to 2%
• Improved cash flow by freeing working capital
• Better vendor negotiations with accurate forecasts

Lessons Learned

• Simple baselines are harder to beat than you think
• Promotion effects need special handling
• Buyers need to trust and understand the forecasts
• Start with high-volume SKUs for quick wins