# AI Measurement Operating Model

Use this when AI measurement needs to move from a one-off pilot report to a repeatable portfolio practice.

## 1. Intake

Every AI use case should enter with:

- Value hypothesis:
- Business owner:
- Product or solution owner:
- Risk owner:
- Target role:
- Work unit:
- Monthly volume:
- Baseline source:
- Decision date:

Reject or pause intake when:

- The work unit is unclear.
- No baseline can be collected.
- The team only wants adoption reporting.
- No one owns the decision after the pilot.

## 2. Instrumentation

Collect signals from real work where possible.

- Usage events:
- Task completion:
- Review time:
- Correction or defect rate:
- Escalations:
- Accepted output count:
- Cost signals:
- User feedback:
- Risk or policy exceptions:

Mark each signal as:

- Directly measured
- Sampled
- Estimated
- Unknown

## 3. Evidence Review

Run a fixed review every two to four weeks during the pilot.

Review with:

- Business owner
- Product or solution owner
- IT owner
- Security or compliance owner
- Finance partner when cost or scale is material

Classify every claim:

- Proven
- Likely
- Unknown
- Risky

## 4. Portfolio Decision

Score each use case from 1 to 5.

| Criterion | Score | Notes |
| --- | --- | --- |
| Proven value | | |
| Confidence in evidence | | |
| Scale potential | | |
| Risk level | | |
| Maintenance burden | | |
| Strategic relevance | | |

Decision:

- Fund and scale
- Improve and retest
- Keep local
- Stop

Next action: