Brightfield's 12 stores now run on AI demand forecasts that hit 94.7% accuracy — up from ~78% manual. Stockouts dropped, capital tied up in inventory fell 18%, and buyers got their evenings back.
The challenge
Brightfield's 12 specialty stores each had their own demand rhythm — weather, footfall, local events. The central buying team forecast all of them by hand. They were always late, often wrong, and burning out.
A previous SaaS attempt failed because it couldn't adapt to each store.
"Our buyers used to forecast on gut and spreadsheets. Now they spend their week on the 5% the AI gets wrong — which is exactly where they're most valuable."
What we did
- Per-store models — one forecast model per store, trained on its own three years of history.
- Explainable recommendations — every "restock" comes with the reasoning a buyer can sanity-check.
- Buyer dashboard — a single screen replaces a folder of spreadsheets.
- Continuous learning — models retrain weekly as new sales come in.
The outcome
Forecast accuracy lifted from 78% to 94.7%. Stockouts fell, overstock fell, and the buying team's week shrank — without losing the buyer judgement that made Brightfield what it is.
