What a monthly client report should say about AI
Clients should not learn about AI from the invoice or from a mistake. A short, steady section in the monthly report tells them first, in facts they can check.
Most clients now assume AI is somewhere in the work they pay for. What they do not know is where, how much it costs, and who checked the result. If the monthly report says nothing, they fill the gap themselves, usually at the worst moment: when an invoice looks lower or higher than expected, or when something goes wrong and the first question is whether a model was involved.
A short, steady section in the report you already send answers those questions before they are asked. It does not need to be long. It needs to be factual, the same shape every month, and drawn from a record rather than written from memory.
Why the report, and not the invoice
The invoice is a commercial document. Whatever appears there reads as a justification for the amount, and a client who first learns about AI from an invoice line will read it as one. The monthly report is where the client already looks to see what was done, which is where AI belongs: as part of how the work was done, next to the work itself.
This does not replace the invoice detail. If you pass model spend through, the invoice still carries it. The report is what makes that line unsurprising, because the client has seen the same figure, with its context, every month.
The five lines that belong in the AI section
Each line is something the record observed, not something a person estimated after the fact. Each one also has a limit on what it claims, and saying that limit out loud is part of what makes the section credible.
| Line | What it shows | What it does not claim |
|---|---|---|
| Workstreams | Which workstreams had AI sessions this month | How much of the work the AI did |
| AI sessions | How many, and which tools and models ran | What was asked or answered |
| Model spend | What AI cost on this client, and whether it is passed through or absorbed | What the work was worth |
| Review | Which deliverables were reviewed, and by which step | A grade of anyone's work |
| What changed | A new tool, model or workflow since last month, or none | That the change was better |
Hours stay where they already are in the report, by workstream. The AI section sits beside them and refers to the same workstreams, so the client can see both on one line of work without anyone converting one into the other.
What to leave out
The temptation in a client report is to make AI look like a benefit. Most of the ways to do that are claims the record cannot support, and a client who checks one and finds it soft will doubt the rest.
- Hours saved. Nobody observed the version of the work that was not done. Any figure is a guess about a counterfactual.
- A share of the work. "The AI did most of this" is not something any tool measures. It is a judgement dressed as a number.
- Per-person AI figures. The client is not managing your team. Report AI by workstream. The people who led the work stay named as the people who led it.
- Prompts and transcripts. They contain the client's own confidential material and your team's working notes. Structured facts, such as tool, model, sessions and cost, are enough.
- Tool marketing. Which model you use matters to the client only if it changes how their work is handled, and then it goes under what changed.
Keep the same shape every month
The value of the section is mostly in its consistency. When the same five lines appear in the same order every month, a change stands out without anyone having to point at it, and a month where nothing changed is information too. Write "No change" rather than dropping the line.
When a figure moves, add one sentence that says why. Model spend doubled because a new workstream started. A workstream shows no AI sessions because the client asked for that work to be done without AI, and the record shows that request was kept. If your AI disclosure policy promises the client a particular kind of review or notice, this section is where they see each month that the promise has a record behind it.
An illustrative example
The numbers here are illustrative. A consultancy runs three workstreams for a client: research synthesis, a monthly performance pack, and a data migration. Its October report has the usual sections on progress, hours and next steps. The AI section reads like this.
- Workstreams. Research synthesis and data migration had AI sessions. The performance pack did not.
- AI sessions. 18 sessions across one assistant and one coding agent, on two models.
- Model spend. $46 this month, absorbed in the fixed fee as agreed.
- Review. All four research summaries were checked against their sources before delivery. Every migration script was reviewed and run against a copy of the data before release.
- What changed. The coding agent is new this month, on the migration workstream only.
That is five lines. The client's operations lead asks one question in reply, about how the migration scripts are tested, and the answer is already in the review line. The next month, when the model spend line shows the agent's first full month, nobody is surprised.
Start with next month's report
You do not need new tooling to write this section once. Pull the five lines for one client from whatever record you have. If the record cannot answer a line, that is the first thing to fix, and an AI work register kept by hand for a month is enough to fill it.
Send it to one client, unprompted, and keep the shape. The first month is a disclosure. By the third it is simply part of how you report the work, which is what it should have been all along.