01
What this scorecard tracks
This template is structured. The working copy is being finalized by our subject-matter editors — directionally accurate today, refined shortly.
- Baseline: how long the task took, and how often it needed rework, before AI.
- Time saved per run, measured against that baseline.
- Correction rate: how often the human reviewer had to materially change the AI output.
- A scale / hold / stop recommendation tied to those two numbers.
02
Why two metrics, not ten
A pilot that cannot show time saved and a falling correction rate is not ready to scale. Everything else is a vanity metric.