Turn a Messy Client Brief Into Better Scope Questions With AI

Capture what the client actually supplied

A scope intake may arrive as an email, a slide deck, meeting notes, and a few comments in a shared document. Before generating a statement of work, put the approved intake materials in one review set. Record their dates and authors, and distinguish client statements from your team’s notes. Ask AI to extract requested outcomes, deliverables, dependencies, deadlines, and named stakeholders with a citation to each source passage.

Keep a separate field for information that is absent or ambiguous. If the brief says “launch support” but never defines channels or duration, the correct extraction is not a plausible package of services. It is an open question. NIST’s generative AI profile warns that generated responses can diverge from input and still sound confident; an evidence-first intake makes such additions easier to catch.

Make uncertainty visible before drafting

Sort open questions into categories the client can answer: desired outcome, in-scope work, materials they will provide, acceptance criteria, review timing, and decision authority. Then prioritize questions that would change the effort or sequence. “Who provides the approved copy?” may matter more than a preference about report formatting. An AI-generated list is a starting point for a conversation, not a substitute for one.

A useful intake table has four columns: source statement, proposed interpretation, confidence or uncertainty, and owner of the answer. Avoid numerical confidence scores unless they have a real evaluation method behind them. Plain labels such as “explicit,” “inferred,” and “unanswered” are easier to challenge. If two source documents conflict, show both passages and ask the client to resolve the conflict rather than choosing the newer-looking sentence.

Draft from confirmed answers only

Once answers arrive, ask AI to draft a scope outline using only confirmed material. Give it a structure: objectives, deliverables, client inputs, milestones, review process, exclusions, and assumptions. Require every assumption to be labeled as an assumption. A sentence such as “Client supplies access within five business days” should not appear as a settled obligation merely because it makes the schedule work.

Have the delivery owner compare the outline with the intake table. Look especially for nouns that grew in meaning: “dashboard” may have become “automated analytics platform,” or “review” may have become “unlimited revisions.” NIST’s AI Risk Management Framework and Playbook emphasize adapting risk-management practices to the particular use case; here, the practical control is a human check of every proposed commitment against its source.

Close the loop with a decision record

Send a concise question list rather than a speculative full document when major inputs remain missing. After the client responds, update the intake record with the answer, respondent, and date. Mark superseded assumptions instead of deleting them from your internal working history. This gives the team a way to explain how the scope evolved without asking someone to search through messages months later.

Before circulating the service document, run one final question: can a person outside the sales conversation tell what will be delivered, what the client must provide, and what happens when information arrives late? If not, revise the document rather than asking AI to make it sound more polished. Clear scope is the product of resolved decisions; fluent prose is only its presentation.

Sources and further reading

• NIST — AI RMF Generative AI Profile
• NIST — AI Risk Management Framework overview
• NIST — AI RMF Playbook