For a small accounting, tax or bookkeeping practice, the AI that pays first is the admin after a client conversation. That means the file note, the summary email and the follow-up tasks, all drafted by AI and checked by a person before anything is filed or sent. It doesn’t pay where a client would rely on an answer no practitioner has checked. For registered tax and BAS agents, the Tax Practitioners Board’s July 2026 guidance sets two ground rules. Putting client information into an AI tool can count as disclosing it, depending on how the tool is set up, and that needs the client’s permission. And whatever the tool produces needs checking.
A sole practitioner I know ranks AI behind AML, engagement letters and the Privacy Act, with no time left to keep up with it. That’s understandable: from 1 July 2026, practices that provide certain services have anti-money laundering obligations for the first time. So the useful question isn’t which AI to buy. It’s which hour of admin AI can take off you without adding to that list.
A partner at a practice I spoke with recently records most of his client calls with an AI note-taker. The note-taker does its job. Everything after it is manual: filing the transcript in the client’s folder, emailing the client a summary, turning the action items into dated tasks. The AI part works. The hand-offs around it are where the time goes.
What can AI do for an accounting practice today?
Today AI is best at reading and drafting: summarising a client call, drafting the file note and the follow-up email, turning action items into tasks, explaining a document in plain English and speeding up tax research. In its guidance on AI and the Code of Professional Conduct, TPB(GS) 55/2026, issued on 22 July 2026, the TPB says AI tools, “when used appropriately, provide a significant opportunity to increase productivity and deliver efficiencies for tax practitioners”.
The sole practitioner I mentioned sees providers building AI search over ATO material and their own content, with no platform yet on top. Tax Nuggets’ Ask Nugget is one. As at October 2026, it says it “generates AI answers exclusively from verified Australian tax sources” and that “Every answer cites what it relied on”. Those are the vendor’s claims, and checking the sources is still your job.
What admin should a practice hand to AI first?
Hand AI the admin after a client conversation first. It happens every day, the steps are the same each time, and each draft is quick for the person who ran the call to check.
- Tell the client you’re recording At the start of the call, and get their agreement. Recording laws differ between states.
- The note-taker transcribes Use one your practice has approved for client information.
- AI drafts the file note What was discussed, what you advised and what the client will send.
- AI drafts the client email A plain-English summary and the next steps.
- AI drafts the follow-up tasks Each with an owner and a due date, in your task manager.
- You check and approve Nothing is filed or sent until a person has read it.
- It’s filed in the client’s folder In your own document system, so the record stays with the practice.
Write the steps down before you automate them. Without a clear process, you’re only going to amplify what is missing. Then decide where AI fits by asking how much each step matters to get right.
Is it safe to put client information into AI tools?
It can be, with the client’s permission and a tool set up for client information. For registered tax practitioners, the TPB’s July 2026 guidance spells it out:
- Under the Code of Professional Conduct, you “must not disclose any information relating to a client’s affairs to a third party without your client’s permission”.
- That disclosure “can include entering client information into AI models and tools, depending on how these tools are configured and used”.
- The TPB recommends telling clients “where data will be stored and whether AI tools may be used”, and says “The client permission may be by way of a signed letter of engagement”.
- Where the information includes tax file numbers, “additional obligations under the Privacy (Tax File Number) Rule 2015 will apply”.
Where a tool stores your data is part of that answer, and a vendor’s security page may not say. As at October 2026, Otter’s privacy and security page says it “uses AWS S3 storage” but doesn’t name a country. If a vendor’s page doesn’t tell you, ask before client calls go into the tool.
My rule for a practice: no tax file numbers or client files in any AI tool the practice hasn’t approved for client information. Know where each approved tool stores data, and get consent before recording a call. The plan matters too, so start by checking which plan your staff are on.
What AI is already in Microsoft 365 and Xero?
Xero has an AI companion, JAX, and Microsoft 365 has Copilot, so start with what you already pay for. I always present the native option first. Switching tools costs time and money, so why switch just because something’s 1% better?
Xero says JAX chat is “available to all Xero subscribers and users”, and as at October 2026 “There’s currently no additional charge to Xero subscribers to use JAX chat experience”. Xero also lists a JAX feature that “automatically categorises and matches high-confidence bank transactions”, so check which plans include it. The TPB’s guidance covers “AI features which may be embedded within software or third-party platforms”, so the permission and checking rules apply here as well.
If your practice runs on Microsoft 365, Copilot does a lot and has moved fast. On 4 November 2025, Microsoft announced in-country data processing for Microsoft 365 Copilot, with Australia in the first four countries. That means “Copilot interactions are processed, under normal operations, in data centers located within a nation’s borders”. In an April 2026 update, Microsoft said it now expects local processing for Microsoft 365 Copilot to become available in Australia “by the end of 2026”. Check the status before you tell clients their Copilot data is processed here. My guide to Microsoft 365 Copilot pricing covers what’s free with Microsoft 365 and what a paid licence adds. If your team already has it, Copilot training gets everyone using it the same way.
How do AML/CTF, engagement letters and the Privacy Act affect AI use?
AML/CTF, engagement letters and the Privacy Act shape what client information goes into AI tools and what your engagement letter needs to say.
AML/CTF. AUSTRAC says “AML/CTF obligations apply if you provide a designated service in table 6 from 1 July 2026”. Its list of professional designated services covers work such as helping with the sale or purchase of real estate, a company or a trust, creating or restructuring companies and trusts, and holding or managing client property for a transaction. It also says “Merely influencing how the customer proceeds, providing general advice or ancillary services isn’t sufficient.” A practice that only does tax returns and BAS may not be caught, so check the list against the work you do. A practice that is caught must enrol with AUSTRAC, have an AML/CTF program and know its customer before it starts the service. AUSTRAC has program starter kits for accountants.
Privacy Act. The OAIC’s small business guidance says “A small business is one with an annual turnover of $3 million or less”, and most are exempt. But “Regardless of turnover”, the Act covers any business that is “a reporting entity for the purposes of the Anti-Money Laundering and Counter-Terrorism Financing Act 2006”. So a practice that would otherwise be exempt is covered for its AML/CTF-connected activities once it’s a reporting entity, and the rest of its work stays exempt unless another exception applies. Check where you stand, including the automated decision-making rule that starts on 10 December 2026. Korbai’s free AI and privacy check shows whether that rule is likely to reach you.
Engagement letters. APES 305 Terms of Engagement (2024), effective from 1 January 2025, requires members in public practice “to document and communicate the terms of engagement with a client”. With the TPB’s guidance, the engagement letter is the natural place to ask permission for AI use and to say where client data will be stored.
A one-page AI use policy sets what staff may put into AI and who checks what comes out.
Where does AI fall short, and who checks the output?
AI falls short wherever a confident wrong answer costs more than the time it saved. The TPB says tax practitioners “should verify and review AI generated content for accuracy throughout each step of the workflow”, and should “ensure AI outputs are assessed and supplemented by professional judgement, before being relied on”.
| Task | Use AI? | On one condition |
|---|---|---|
| File notes, summaries and tasks after a client call | Now | A person approves each draft |
| First drafts of client emails and letters | Now | The practitioner reads it before it’s sent |
| Tax research | With care | In a tool that cites its sources, and you check them |
| Coding bank transactions | With care | Reviewed before anything is lodged |
| Advice a client will rely on | Not yet | Not without a practitioner’s check |
| Tax file numbers in a tool not approved for client data | No | Never |
Where should a practice start?
Pick one job you do every week, and measure it.
- Pick the job. The admin after client calls is a good first candidate.
- Write the steps down. If they live only in your head, AI will copy the gaps.
- Check what you already pay for. Microsoft 365 or Xero may already do part of it.
- Sort out permission and storage. Add AI use to your engagement letter, and confirm where each tool keeps client data.
- Measure before and after. Time the next five by hand and the five after that with AI. The time savings calculator does the arithmetic.
Don’t wait for a perfect process, either. Waiting has a cost too: the time you don’t get back.
If you run a practice in Sydney and you’d like a second view on your first AI job, tell me about it on a discovery call. Afterwards, I’ll come back to you with whether it’s a good first candidate and what I’d do first.
FAQ
Do I have to tell clients I use AI?
If client information goes into an AI tool, probably. The TPB’s July 2026 guidance says entering client information into AI tools can count as disclosing it to a third party, depending on how the tool is set up, and that needs the client’s permission. It recommends telling clients where data will be stored and whether AI tools may be used. A signed engagement letter can carry the permission.
Can I record client calls for AI notes?
Tell everyone on the call at the start and get their agreement before you record or use an AI note-taker. Recording laws differ between states, so check the rules that apply to you.
Can AI do tax research?
It can speed it up. Use a tool that cites its sources, such as one built on the ATO Legal Database, and check each source before you rely on the answer. The TPB expects professional judgement before any AI output is relied on.
Is there a free AI for accounting?
Some of it comes with software you already pay for. As at October 2026, Xero says there’s no additional charge for JAX chat, and Microsoft’s Copilot Chat FAQ says Copilot Chat is “automatically included” for organisations with Microsoft 365. The Copilot pricing guide above explains what a paid licence adds.
Will AI replace accountants?
AI can take on routine drafting, and the TPB sees it, used appropriately, as a way to increase productivity. The TPB also keeps practitioners ultimately responsible for their services, so the judgement and the sign-off stay with you.
General information only, not legal advice, and not tax or accounting advice. Current as at October 2026: the rules and the tools in this area are changing quickly. Xero, JAX, Microsoft, Microsoft 365, Copilot, Otter, Tax Nuggets and Ask Nugget are trademarks of their respective owners. Korbai is independent and not affiliated with or endorsed by any of them.
