Your proposal says the team will use AI to make client reporting faster. The client hears “the report will arrive earlier.” Your team means “we hope drafting takes less time.” Resolve that gap before the first delivery.
Promise the outcome and review process your team can perform. If AI changes the client's data handling, approval role, delivery conditions, or another material term, address that change in the agreement and operating plan.
Two sample scope clauses
The examples below are practical drafting language for a hypothetical reporting service. They require adaptation and appropriate contract review. They do not define legal obligations or The Ops Guide's commercial commitments.
Client-facing monthly report: “We will prepare a monthly report from the agreed source systems covering the stated reporting period. Our team will review reported totals against the available source records, assess the recommendations, and approve the final report before delivery. We will identify missing or delayed source data in the report and coordinate with your named contact when it prevents an agreed section from being completed.”
Working analysis draft: “We will prepare a working analysis for your internal review, with source references and stated assumptions. Your designated reviewer will confirm the business facts and approve any external use. The draft will identify unresolved questions that could change the recommendation; preparing the draft does not authorize publishing it or acting on those unresolved assumptions.”
The first clause places final review with the provider. The second gives the client a defined review role. Your team should select the arrangement it can support and price, rather than leave responsibility implicit.
Translate the promise into a fulfillment checklist
| Agreed commitment | Operating evidence | Owner |
|---|---|---|
| Report uses the agreed source systems. | Source list, covered dates, and access status. | Analyst checks the inputs. |
| Reported totals receive review. | Comparison with source records and disposition of differences. | Reviewer approves the numeric claims. |
| Recommendations receive review. | Named reviewer records accepted changes and unresolved assumptions. | Engagement lead approves the advice. |
| Missing information reaches the client. | Report note and agreed follow-up when an omission affects the deliverable. | Client contact coordinates the next action. |
| Final delivery receives approval. | Exact final version, approver, recipient, and delivery record. | Engagement lead authorizes delivery. |
If your team cannot produce the evidence, revise the promise or the process before selling it. “Human reviewed” needs an actual reviewer with time, access, and criteria for the parts they approve.
Explain material AI use in context
A client may care that its records enter a third-party system, that an output remains a draft, or that the service depends on a client approval. Address those facts in the appropriate place. Follow applicable requirements and the client's agreement.
Do not substitute a broad “AI may be involved” sentence for the information needed to approve data handling or understand the service. Conversely, a long description of every drafting tool can obscure the deliverable. Decide which facts affect the client's choice and obtain appropriate advice where requirements are unclear.
For example, if the agreement permits only approved systems for customer records, the team must resolve that requirement before uploading the records to another tool. An employee's intention to review the output does not settle the data-handling decision.
Test the promise before the first live delivery
Run a rehearsal with an approved sample. Give the analyst a missing source, introduce a conflicting total, and have the reviewer identify an unsupported recommendation. Confirm the team can produce a coherent report while showing the limitation that matters to the client.
Check the calendar too. If client approval is necessary before delivery, give that person a defined review window and explain what happens when they do not respond. Avoid promising a delivery date that depends on an unstated client action.
NIST's voluntary generative AI guidance supplies broader risk context. The Ops Guide's fractional CAIO service can help align your AI-assisted work with clear responsibility, acceptance, and client commitments.

