Module 13The six AI workflows4 min read
Client onboarding: one intake, three documents that agree
How to produce the engagement letter, the information request and the AML checklist from a single intake, so the scope in the letter matches the records you ask for and the entity in the AML file. Includes the beneficial ownership step most firms miss.
The onboarding workflow produces the engagement letter, the information request and the AML checklist from one intake conversation. The value is not speed on any single document. It is that the thirtieth client gets the same three documents as the first, instead of thirty slightly different versions accumulating in a folder. Build it before you take on the next ten clients rather than after.
Key takeaways
The three documents have to agree. Every service in the letter needs matching records in the request, and every entity needs to appear in the AML file.
Use your own templates. A firm's engagement letter has usually been through a professional body or an insurer. A generated one has not.
Never fill a gap with a plausible answer. Unanswered fields are written as NOT PROVIDED.
It records AML facts, it does not rate risk. A responsible person assigns the rating and signs it.
The intake
Ten questions, answered before anything is drafted, because each one lands in a different place downstream.
| Question | Where it lands |
|---|---|
| Legal entity name, number, registered address | Letter parties, AML record |
| Entity type: sole trader, partnership, company, trust | Scope, filing obligations, AML approach |
| Directors, partners or trustees | AML identity checks |
| Anyone owning more than 25 per cent | Beneficial ownership, the step most often missed |
| Financial year end | Deadlines, records requested |
| Which services | Scope clause, fee basis |
| Platform, and whether you get access or an export | Information request wording |
| Who signs, and their role | Signature block |
| First year, or taking over from another firm | Handover records, professional clearance |
| Anything unusual about this client | Risk rating, scope exclusions |
The cross-check that makes it worth doing
Drafting three documents separately is how they drift apart. Drafting them together lets the workflow check them against each other and say so plainly:
- Every service in the engagement letter has matching records in the information request.
- Every entity named in the letter appears in the AML checklist.
- The year end in the letter matches the period in the information request.
- Nothing appears in the information request that is outside the scope clause.
"Scope lists payroll, information request has no payroll records" is the kind of finding that costs five minutes now and an awkward conversation later.
Install it: skills/onboarding-workflow
Runs the ten-question intake, drafts all three documents together, cross-checks them against each other, and reports what it could not complete. Includes AML checklist fields and fallback document outlines.
Get the six workflows on GitHubWorks in Claude, Claude Code, Codex and Cursor. Copy the folder into .claude/skills/ and ask in plain language.
Beneficial ownership, the step people miss
If the shareholder is a company, the chain is not complete until a natural person is named. A checklist that records "Holdco Ltd, 100 per cent" and stops there has not done customer due diligence, it has recorded a company number.
The workflow asks for the chain and marks it NOT PROVIDED when it does not get one, which is the entire point of a checklist.
What it cannot do
It does not set a risk rating. AML risk is a judgment a responsible person makes and signs. The workflow records the facts the rating is made from.
It does not verify identity. No document checking, no database lookup. It produces the checklist a human completes.
It does not give legal advice. Engagement letter clauses come from your firm's template or your professional body.
It does not send anything. All three are drafts. The same boundary applies across every workflow here, and the reasoning is in AI for accountants.
Common questions
Do I have to use the generated engagement letter?
No, and you probably should not. The workflow asks for your existing template first and follows its structure and clauses. The fallback outline exists only for a firm that has nothing.
Is this AML compliant?
It records the common core of customer due diligence. Requirements differ by jurisdiction and professional body, so check yours. Compliance is the responsible person's, not the workflow's.
How does this connect to pricing the client?
It does not, deliberately. Pricing runs from the client's actual data in the fee scoping workflow rather than from an impression formed at intake.
More in The six AI workflows
The monthly query chase: one email instead of six
Find every transaction you cannot code and turn it into one client email instead of six.
The client email desk: first drafts of the replies that eat an afternoon
First drafts of the technical replies, with every position marked for a human to verify.
Management pack commentary: three reports to a one-page pack
Three reports to a one-page commentary, with every variance computed by a script.
Fee scoping: price a client from their data, not from a guess
Price a client from measured volume and measured messiness instead of from a guess.
The year-end review pass: a second set of eyes that is not yours
A second set of eyes that runs the same checklist on every file, at the same standard.