Demand forecast + reorder

"We are out of stock on the things that sell and sitting on a warehouse of the things that do not."

Forecasting was a spreadsheet of last year plus a growth factor. Reorder points were static, so service level and stock cover were both wrong in opposite directions across the range.

A hierarchical forecast at SKU-location-week with reconciliation up to category, a reorder policy that takes lead-time variance seriously, and a weekly run that writes suggested purchase orders for the planners to approve.

Promotions were not in the feature set, so the forecast missed every uplift. The promo calendar became a first-class input and a data dependency the business had to own.

M5 Forecasting — Accuracy (Walmart, 42k series)

Hierarchical forecast with reconciliation, plus the inventory policy simulator that turns forecast error into a working-capital number.