Assessment

Could you answer a client’s question about AI today?

The level your record supports across the six things attribution has to get right, the dimension holding it down, and the two moves worth making next.

Capture

Whether AI involvement is recorded at all, and by everyone.

When someone uses AI on client work, where does that get recorded?
How consistent is that recording across the team?

Linkage

Whether the record attaches to a deliverable and to human hours.

Can you see AI contribution for a single deliverable or workstream?
Where does AI work sit relative to logged human hours?

Review

Whether a named person checked the work, and what they checked.

For client-facing work with material AI contribution, is the review recorded?
Could you say what a review checked, six months later?

Reporting

Whether the record can answer a question someone actually asks.

If a client asked how AI was used on their project, how long would the answer take?
Which of these could you answer for last quarter?

Privacy

Whether measurement stays about work rather than about people.

What does your AI record contain about individuals?
Do people know what is recorded about their work, and why?

Client disclosure

Whether clients know how AI is used on work they pay for.

What do clients know about AI use on their work?
Is there an agreed position on AI in your commercial terms?

Your result

12 questions left.

About the levels

The chain is only as good as its weakest link.

Five levels: ad hoc, structured, connected, reconciled, governed. A firm with automatic capture and no recorded review will still fail the first question a client asks, so the result is held to one level above the weakest dimension.