Split client deliverables into AI gathers, a person checks and a lead decides, with a three-pass evidence pack and a short method note
A consultancy founder sets out how a small advisory firm can use AI on client work: AI gathers and sorts, a named person opens every source, the lead decides what reaches the client, and the client is told what AI did.
Evidence: The source shows its work: steps, screenshots, code or data you can inspect.
The business problem
A consultant or agency wants the speed of AI on research and deliverables without sending clients fabricated citations or breaking confidentiality terms.
What was tried
Decide up front which work AI may do alone: gathering, sorting, summarising, tidying data and drafting slide text from your own notes. The test is whether a person can quickly check the output against a real source. For interview synthesis, have AI group interviews by theme, then open three quotes behind each theme and read the interviews that fit no theme. Run an evidence pack in three passes: AI gathers and sorts, a named person opens every source and checks each figure, quote and citation, and the lead decides what is sent. Add a two or three sentence method note saying where AI helped and who owns the recommendation. Check engagement letters and data terms first, use business accounts with training turned off, strip names where possible, and keep a list of which tools each client has cleared.
What was reported (not reported)
No result from the firm's own work is reported. The article cites outside studies: AI helped on tasks it handles well but users were 19 percentage points less likely to get a harder problem right, and a survey in which 70 percent of companies said they would not trust a consultancy report prepared with AI.
Limitations
Guidance from a consultancy that sells AI consulting, training and workshops, with no client case study of its own. The outside figures are quoted from third-party surveys and were not checked here. It names a documented incident of a government report with fake citations that led to a partial refund, which is the risk the checking step is meant to catch. It says its own rule of thumb about a share of work has no formal basis.
What you need
A named reviewer for each deliverable, a list of approved tools per client, and engagement letters that allow AI use.
Sources
- AI Smart Ventures (Nicole A. Donnelly; sells AI consulting) ↗ Vendor case study, published October 2, 2026
Source published: October 2, 2026. Last reviewed here: October 14, 2026. Spot a mistake? Tell us.
Related workflows
Hand over an AI workflow so a client's team can run it: named owner, prompt library, written reasoning, review cadence and a consultant-free week
A consultant's checklist for leaving an AI system the client can keep running: access held by a named employee, a dated prompt library, the reasons behind choices, a standing review and a few health metrics, over a 90-day handover.
Run a one-person company on five building blocks: a browser agent, a coding agent, scheduled routines, memory notes and approval gates
A solo founder of a physical-product business with no employees describes the five pieces of an agent setup and the rules that keep it safe: run a workflow by hand twice, write memory notes, gate money paths, then schedule it.
Plan ten Claude Code skills for a small service business, one a day, and review the first 20 to 50 runs before anything is sent or paid
An agency guide lists ten back-office tasks to turn into Claude Code skills, such as inbox triage, reply drafts, invoice processing and weekly reconciliation, with a rule to review early runs by hand before irreversible actions.
Run a solo consulting practice on Claude Cowork agents with a three-pass quality check before anything reaches a client
A former strategy consultant describes a plugin that gives a solo consultant agents for proposals, kickoff packs, research and deliverables, with source checking and two independent audits before delivery.