Retail AI Demand Forecasting & Replenishment

The Challenge

A 15-store retail group was using rule-based reordering and gut feel from store managers — leading to chronic stockouts on bestsellers and dead stock on slow movers. £1.2M was sitting in inventory that hadn't moved in 90 days, while top-selling SKUs were out of stock 18% of the time. They wanted an autonomous AI layer that learned per-store demand, generated reorder suggestions, and (with permission) executed orders against approved supplier catalogues.

Our Approach

1. Ingested 4 years of POS, weather, local events and promotional data into a unified feature store on Snowflake.
2. Trained per-store, per-SKU demand models — capturing localised patterns (e.g., one branch sells 4× the gluten-free range due to a nearby clinic).
3. AI agent generates daily reorder proposals with explanation ("ordering 60% more stock due to forecast 18°C weekend") for store-manager review.
4. Auto-execute mode for trusted SKUs — orders dispatched directly to suppliers via EDI / supplier APIs.
5. Operator dashboard surfaces shrinkage, dead-stock and margin opportunities in plain English summaries generated by GPT.

The Results

• Stockouts reduced by 23% across the bestselling 200 SKUs.
• £412,000 of dead stock cleared via targeted markdowns surfaced by the AI.
• Reorder cycle reduced from weekly manual review to daily automated dispatch — 7× faster.
• Margin uplift of 2.4 percentage points within the first quarter.
• Store managers report ~5 hours/week recovered from manual stock review.

Stack & Disciplines