# AI workflow audit

By John Rice · Make AI useful.

Use this worksheet to choose and scope one practical AI experiment.

## 1. Name the work
- Workflow:
- Person responsible for the result:
- Who uses the result:
- What starts the task:
- How often it happens:
- What “finished” means:

## 2. Map one real example
1. Input arrives from:
2. Information is collected from:
3. Work is transformed by:
4. Decisions are made by:
5. Result is reviewed by:
6. Result is delivered to:

Which step creates the most friction?
Which step requires judgment or authority?
What happens when information is missing?

## 3. Choose a bounded experiment
- One step AI will support:
- Exact output it should produce:
- Actions it is allowed to take:
- Actions that require review:
- Person accountable for the final result:
- Existing process to use if the test fails:

## 4. Check the inputs
- [ ] I have permission to use these inputs.
- [ ] I understand the information they contain.
- [ ] The source of truth is identified.
- [ ] Missing and conflicting information has a defined treatment.

## 5. Define acceptance checks
- Required facts or fields:
- Important mistakes to catch:
- How a reviewer verifies the output:
- Maximum acceptable correction effort:
- Condition that stops the experiment:

## 6. Run and learn
Try a straightforward example, an incomplete example, and an ambiguous example.

| Example | Checks passed? | Review/correction effort | What needs to change? |
| --- | --- | --- | --- |
| Straightforward | | | |
| Incomplete | | | |
| Ambiguous | | | |

Decision: revise / expand carefully / stop
Why:
Next owner and action: