# AI Foundation Check

Use this before turning an AI idea into a value claim.

The point is simple: do not measure AI as if the surrounding work is already clean. Check the work, data, ownership, risk, and review path first.

## 1. Work being measured

- Work unit:
- Who performs it today:
- Who owns the process:
- Monthly volume:
- Current pain:
- What success would look like:

## 2. Baseline

- Current median time per task:
- Current p90 time per task:
- Current quality score:
- Current error or rework rate:
- Current escalation rate:
- Current cost per task:
- Source of baseline:
- Confidence in baseline:

## 3. Data and knowledge

- Data sources AI will use:
- Source owner:
- Last reviewed:
- Duplicates or conflicting sources:
- Missing metadata:
- Known quality issues:
- What must be cleaned before scaling:

## 4. Access and policy

- Who should have access:
- Who should not have access:
- Sensitive data involved:
- Retention requirement:
- Sensitivity label or classification:
- Policy owner:
- Approval needed before pilot:

## 5. Review work

- Who reviews AI output:
- What they check:
- Average review time:
- Common corrections:
- Failure examples:
- Decision rule for accepting output:

## 6. Measurement decision

Choose one before the pilot starts.

- Scale if:
- Change if:
- Retest if:
- Stop if:

## 7. Honest status

Use one of these labels.

- Ready to measure:
- Needs foundation work:
- Useful experiment, not an ROI case yet:
- Too risky or unclear:

## Notes

What did this check reveal that was not really an AI problem?