AI client onboarding for accounting firms: what to hand off

AI is useful in client onboarding for the writing and checking work: turning the first conversation into a scope note, tailoring the document request to the client’s structure, checking what the client sends back, drafting the welcome emails and setting up the first jobs. It should not verify identity, rate a client’s risk or decide whether you take them on. And before any of it, the client needs to have agreed to their information going into an AI tool, which onboarding is the right moment to ask.

This guide takes the onboarding sequence most firms already run and marks which steps to hand to AI, which to keep, and what to settle first. It is part of our wider coverage of AI for accountants.

Most of the effort in onboarding is not the engagement letter. It is the back and forth around it: the call notes nobody wrote up, the generic document list that asks a sole trader for trust deeds, the three emails chasing a missing bank statement, the prior accountant’s file that arrives as 40 unlabelled PDFs, the jobs set up by hand in the practice management system.

None of that is professional judgement. It is reading, sorting and writing, which is what current AI tools do well. That is why onboarding is a good first project for a firm starting out with AI: the gain is visible to the client in their first week, and the risky decisions are easy to fence off.

1. The discovery call becomes a scope note

Record or take notes on the first call, then ask an assistant to turn them into a structured scope note: entities, services requested, deadlines mentioned, software the client uses, open questions. The person who ran the call checks it the same day while it is fresh. This note feeds every step after it, so it is worth getting right.

2. A document request that fits the client

Give the assistant the scope note and your master document list, and ask for a request tailored to this client’s structure: a company with a trust gets a different list from a sole trader. Ask it to explain in plain English why each item is needed. Clients send things faster when they understand the reason.

3. Checking what comes back

When documents arrive, AI can do the first pass: is every requested item there, is each one legible, does each cover the right period, which names or dates do not match the scope note. The output is a short gap list, which becomes the one chase email instead of three.

4. The welcome and “how we work” sequence

Draft a short series of emails from your own templates: who the client’s contacts are, how to send documents, what happens in the first month, what you need from them and by when. AI personalises the template from the scope note. A person reads each one before it goes.

5. Setting up the practice management record

Whether you run Karbon, FYI, Kloud Connect, AccountKit or Xero Practice Manager, the scope note can be turned into the client record, the recurring jobs and the first deadlines. Some of these systems now have AI features that do part of this inside the product; otherwise, ask a general assistant to produce the structured list and paste it in. Our piece on building a tax intake tool in forty minutes shows how far a small team can take this.

6. The first-year question list

Once the prior accountant’s file and last year’s returns arrive, ask an assistant to read them and list questions for the first meeting: balances that look odd, elections that may need revisiting, items that recur and need a process. These are questions for a qualified person to raise, not conclusions.

Identity and due diligence. From 1 July 2026, Australia’s anti-money laundering and counter-terrorism financing laws apply to designated services commonly provided by accountants. Not every accounting service is designated: AUSTRAC’s guidance on professional designated services gives examples, such as an accounting practice that receives and pays out a client’s money being likely to provide a designated service. Where you are caught, AUSTRAC’s obligations guidance covers initial customer due diligence, customer risk ratings and ongoing due diligence. UK firms already apply customer due diligence under regulation 27 of the Money Laundering Regulations 2017. AI can collect and organise the evidence. The verification and the risk rating are yours. Our piece on getting ahead of Tranche 2 covers the Australian changes in more depth, and using AI for AML risk assessments shows where AI can help draft the assessment a person then signs.

Accepting the client. Whether you take on a client is a judgement about fit, risk and capacity. An assistant can summarise what you know. It should not make the call.

The scope and terms you sign. AI can draft the engagement letter from your template and the scope note. What the work includes, what it excludes and what it is worth are decisions a partner makes and signs.

Onboarding is where client information first enters your systems, so it is where permission for AI use belongs. The rules differ by market.

• Australia. The Tax Practitioners Board’s TPB(GS) 55/2026, issued 22 July 2026, says registered tax and BAS agents must get permission from each client before disclosing client information to a third party, which can include entering it into AI tools depending on how they are configured and used. It recommends telling the client to whom and where the disclosure will be made, where data will be stored and whether AI tools may be used, and says permission can come through a signed engagement letter or signed consent. Where tax file numbers are involved, the Privacy (Tax File Number) Rule 2015 adds obligations, and the TPB suggests practitioners get their own advice on how it applies.
• United States. Section 7216 restricts how tax return preparers use and disclose return information, and some uses need the client’s written consent in a prescribed form. Our guide to AI and the IRS rules explains when. Separately, the IRS reminds tax and accounting professionals that they must keep a written information security plan, and the FTC expects firms to choose service providers that maintain appropriate safeguards. An AI vendor is a service provider.
• United Kingdom. UK GDPR applies to the personal data you send to any AI processor, and ICAEW and ACCA guidance applies the usual confidentiality principle. Our guide for UK practices covers both.

For the wording, see our guide to client consent for AI, which includes an engagement letter clause. This is a summary of published guidance, not legal advice.

1. Pick your next three new clients as the pilot. Do not retrofit existing clients yet.
2. Add the AI permission wording to the engagement letter those clients will sign.
3. After each discovery call, have AI draft the scope note; the person who ran the call corrects it.
4. Generate the tailored document request and the welcome email from the scope note.
5. Run the completeness check on everything that comes back, and send one consolidated chase.
6. Build the practice management record and first jobs from the corrected scope note.
7. Keep identity checks, risk rating and acceptance exactly where they were, with a person.
8. After the third client, write down what you changed in each prompt and make it the firm’s standard.

Put the approved tools and the data rules in a one-page AI policy so the next person to onboard a client uses the same setup.

Do not rely on the feeling that onboarding got faster. Before the pilot, pull three numbers from your last few new clients: days from signed engagement to first job started, chase emails sent per client, and the share of documents complete at the first request. Measure the same three for the pilot clients. If the numbers do not move, the bottleneck is somewhere AI is not touching, often the engagement letter sitting unsigned, and the fix is a process change rather than another tool.

Can AI do KYC or identity verification for an accounting firm?

It can gather and organise the evidence, and flag documents that look incomplete or inconsistent. The obligation to verify, and the judgement about what the evidence shows, stay with the firm. In Australia, firms providing designated services have AUSTRAC due diligence obligations from 1 July 2026, and those are not delegated by using a tool.

Do I need client consent before using AI in onboarding?

For Australian registered tax and BAS agents, TPB(GS) 55/2026 says permission is needed before client information is disclosed to a third party, which can include an AI tool. US tax return preparers may need written section 7216 consent for some uses. Since the engagement letter is signed at onboarding, that is the simplest place to collect it.

What is the easiest onboarding task to start with?

The tailored document request. It uses information you already have from the first call, a person can check it in a minute, and a request that fits the client usually cuts the number of chase emails that follow.

Should existing clients go through an AI onboarding process too?

Not straight away. Start with new clients, where permission can be built into the engagement letter from day one. For existing clients, update permission at the next engagement renewal before their information goes into any new AI tool.

Will AI onboarding feel impersonal to new clients?

Only if you let AI send things unread. Used well, it removes the generic parts: the irrelevant document list and the repeated chasing. That leaves more time for the first conversation, which is where the relationship is actually set.