Case study · 12 May 2026
How a B2B distributor cut SKU stockouts without doubling the warehouse
Demand forecasting across 18k SKUs, wired into ERP. Fewer stockouts, no second warehouse, no weekly Excel ritual.
A B2B parts distributor had the classic split: too much stock where nobody buys, and stockouts where customers call on Friday afternoon. Planning lived in Excel. The ERP knew inventory. Nobody joined the two truths.
The problem
Purchasing spent about 14 hours a week reviewing rotation by hand. Promotions and seasonality were gut feel. A-class stockouts hurt margin — not through price, but through substitutions and rush inbound.
What we deployed
A SKU × location forecast, fed by order history, lead times and a promo calendar. The output did not land in a slide. It landed as a PO proposal the buyer accepted or rejected.
- source of truth: ERP plus seasonal files, not a separate “AI warehouse”
- human-in-the-loop: the buyer sees where the number came from
- exceptions: new SKUs, clearance, no-history items — a queue, not a hallucination
After 90 days
A-class stockouts dropped 38%. Purchase planning time dropped 11 hours a week. The warehouse did not grow a metre. The team got a panel for model drift and the cost of being wrong — not a chatbot.
This was not a data-science project. It was a purchasing process with a model in the middle.
