# AI economics benchmark pack

Use this pack to make AI value measurable before and after production launch. It connects a current-state baseline, adoption evidence, unit-cost tracking, quality signals, and value-review decisions into one operating artifact.

## What to benchmark

- Baseline monthly workflow volume, average handling time, rework rate, escalation share, delay cost, and quality loss before AI scope is approved.
- Track AI-assisted usage by team, task class, reviewer acceptance, override reason, abandoned workflow, and support demand after launch.
- Attribute inference, retrieval, orchestration, observability, storage, and human review cost to completed business outcomes.
- Compare cost per resolved case, cycle time, service quality, and risk events against the approved baseline every review cycle.

## Review decision

The review should end with one of four decisions: expand, tune, hold, or retire. Expansion requires stable quality, improving adoption, visible unit economics, and named owners for scale risk. Tuning is the right answer when value is present but adoption, quality, latency, or cost drift needs correction.