Defect-detection rollout on the line

Context

A manufacturer builds the case for AI defect detection on a production line. The value is scrap-rate reduction — but the assumed baseline scrap rate and the improvement are estimates, and the case lives or dies on them.

The decision

The tornado shows the annual-value (scrap-rate) assumption dominates; fund on the band and instrument the baseline before scaling.

What most miss

Teams present a crisp NPV built on an un-measured baseline scrap rate. The honest move is to widen the band on that one assumption and measure it in the pilot.

Stakes

A defect-detection case built on an optimistic scrap baseline can miss its number by half once the real rate is measured.