- *What it measures:*
- _Utilization:_ What % of eligible tasks are actually processed by the AI?
- _Override:_ How often do humans reject or ignore the AI's suggestion?
- *Why it matters:* High override rates usually mean the model is untrustworthy or the workflow is clunky. This is the fastest indicator of a failing project.
- *Lean Goal*: Low friction, high trust.