Real-time demand forecasting across 4,000 stores.

A weekly forecasting batch decided stock for 4,000 stores — and was stale before the first customer walked in. Stockouts and markdowns ate margin, and planners spent their week reconciling numbers that never agreed across systems.

We consolidated demand signals into a governed data foundation with a shared semantic layer, then put forecasting agents on top. The agents re-forecast continuously, rebalanced stock across stores and channels, and flagged exceptions for a planner — all grounded in one source of truth.

Forecasts moved from weekly to near real time across the estate. Stockouts and markdowns fell, and planners shifted from reconciling spreadsheets to handling the exceptions the agents surfaced.