Field notes on forecasting, inventory, and putting predictions to work — written for the teams who have to act on the number.
Ordering more can mask the demand, lead-time and phasing errors behind an empty shelf. Here are four measurements that help separate those causes.
Showing 1–7 of 7 articles
MAPE breaks on intermittent demand, accuracy without bias is misleading, and a headline number at the wrong grain tells you nothing. A practical guide to measuring forecasts you intend to act on.
Service-level targets and safety stock are linked through explicit assumptions and approximations. Understanding that relationship helps quantify the inventory effect of a higher availability target.
Allocation decides where the first units go. Replenishment keeps them there. Confusing the two produces stores drowning in size 8 while three miles away it is sold out.
A churn model can rank customers well and still have no measurable business impact when no suitable intervention follows. Common failure modes include leakage, a mismatched horizon and an unusable scoring cadence.
Margin rarely disappears in one place. It drains through discount authority, freight recovery, returns, and price-cost lag — each too small to notice on its own, together large enough to matter.
The features that win offline competitions are frequently the ones that break in deployment. A practical set of rules for building features that behave the same on Tuesday as they did in the backtest.
Walk through the decision, the data available and the evidence a useful evaluation would need.