Best AI for Accountants in Australia: Tools, Privacy and ATO Checks

Use the AI inside the ledgers you already run (Xero, MYOB), add one business-tier assistant that does not train on your data by default, and keep client data out of free tiers. Nothing reaches the ATO unreviewed: the Privacy Act, the Tax Practitioners Board's AI guidance and your engagement letter set the limits.

The rules below come from the regulators' own pages as read in November 2026. They are practical guidance, not legal advice, and they change, so confirm anything that affects a decision with your professional body or adviser. Prices are vendor list prices, in USD unless marked AUD, and your invoice may arrive in Australian dollars at a different figure.

Follow me on Instagram@sagnikteaches

Three layers of AI a practice actually pays for

Most practices end up with three layers, and the mistake is buying the third before using the first. The first is AI inside the accounting software your clients already use, which costs little extra and sees the data without anyone copying it. The second is a general assistant for writing, research and summarising. The third is point tools such as meeting note-takers and document capture.

Connect on LinkedInSagnik Bhattacharya

Free and personal plans sit outside this table on purpose. They are where most accidental breaches start, as the section on free tiers below explains. If your practice is still on them, free ChatGPT for accountants and when client work needs a paid plan sets out the limits.

Subscribe on YouTube@sagnikteaches

Xero and MYOB fit: which ledger AI matters for your client list

A practice rarely chooses between Xero and MYOB for its own sake. It follows its clients, so the question is which ledger AI helps with the clients you already have. Both vendors market AI features on their Australian pages, and both change them often.

Xero's Australian pages describe JAX as "your AI finance partner". The Australian plans are Ignite, Grow, Comprehensive and Ultimate (sized for 10, 20 or 50 people on payroll), priced in AUD including GST; on the plans I could read, usual prices ran from AUD 107 a month for Comprehensive to AUD 250 for Ultimate 50, and new-customer discounts come and go. The Comprehensive and Ultimate cards list smart document capture and auto-reconcile transactions, both marked with Xero's AI icon. Xero launched Smart Document Capture in July 2026, free on all plans according to its announcement, and renamed Files to Documents on 18 September 2026. I could not confirm which JAX functions come with Ignite and Grow, so check that before you promise a client anything.

MYOB's pricing page marks AI BAS "Currently in beta" and describes it as an AI agent that continuously helps prepare your BAS throughout the quarter, sold as an additional purchase on the Lite and Pro plans. Beta means features and pricing can change, and a practice should treat the output as a draft to review, not a result to lodge. The same page lists Ask Iris, a plain-language chat inside MYOB Business (Lite, Pro and Plus), and Smart Reconciliation, also in beta (Pro and Plus). Read both vendors' current pages, because these labels are the first things to change.

The wider comparison of agent features across ledger products is in the tutorial on AI accounting agents in QuickBooks, Xero and Sage, and BAS-specific checks for bookkeepers are in AI for Australian bookkeepers: BAS prep, GST checks and Xero. Here the focus stays on the practice as a whole: which tools, what privacy and professional rules, and what to review.

A pilot that respects your client mix

Say a six-person practice has 140 clients: 70 on Xero, 40 on MYOB and 30 on other software or spreadsheets. Do not pilot everywhere. Pick five Xero clients and five MYOB clients with clean books and a quarterly BAS, ask them in writing, and run the ledger AI on those ten for one quarter. Record, for each BAS, minutes of preparation, number of AI-coded transactions changed by a person, and any error found at review.

The Privacy Act duties that bite when you paste client data into AI

The Office of the Australian Information Commissioner published guidance on privacy and the use of commercially available AI products. Its message is direct: treat putting personal information into an AI tool as a privacy decision, and as best practice do not put personal information, especially sensitive information, into publicly available generative AI tools.

These are the principles that matter for a practice, in plain words.

• Use and disclosure (APP 6). Entering client information into an AI system may count as using or disclosing it. Use it only for the purpose you collected it for, or one the client would reasonably expect, and put in the minimum needed.
• Overseas disclosure (APP 8). Check where the provider's servers are, and whether personal information could go outside Australia. Many AI tools process data offshore, so your engagement letter and privacy policy need to match what really happens.
• Security (APP 11). Take reasonable steps to protect the information. In practice that means business-tier accounts with single sign-on or multi-factor authentication, admin control of who can use the tool, and no personal accounts for client work.
• Accuracy (APP 10). Treat AI output as probabilistic, note which system produced it, and keep human oversight. This is the same point as the ATO-facing review below.
• Transparency (APPs 1 and 5). Keep your privacy policy and notices in step with how you actually use AI.

Does the Act apply to a small practice at all? The OAIC treats a business with annual turnover of $3 million or less as a small business, generally outside the Australian Privacy Principles, but lists exceptions, such as a business related to a larger one the Act covers or one that has opted in. Separately, it notes that the Privacy (Tax File Number) Rule 2015 regulates how tax file number information is handled, and a tax practice handles TFNs every week. Whatever your turnover, assume you are in scope and ask your adviser to confirm it.

One date is worth putting in the diary. From 10 December 2026, organisations covered by the Act have to describe in their privacy policy where a computer program makes, or is substantially involved in making, decisions that significantly affect people's rights or interests. The OAIC's APP 1 guidelines say a program used only to document a human decision, such as a word processor, is not captured, so drafting an email or summarising a meeting is unlikely to be the target. Automated onboarding screens or client-risk scoring deserve a look from your adviser before then.

What the Tax Practitioners Board guidance expects

On 22 July 2026 the Tax Practitioners Board issued TPB(GS) 55/2026, The use of Artificial Intelligence and the Code of Professional Conduct, after an exposure draft in March. It creates no new obligations. It explains how the existing Code and the 2024 Determination apply when AI is used: competence, reasonable care, confidentiality, record keeping and supervision. The message running through it is that practitioners remain ultimately responsible for their services, must apply their own professional judgement, and should document how they verified AI output.

Two obligations catch practices out. Confidentiality: the Code prevents disclosing information about a client's affairs to a third party without the client's permission, and an AI provider that stores or processes the data is a third party. The guidance says that, depending on how a tool is configured and used, entering client information into it can be such a disclosure, so get permission first, and it recommends telling the client to whom and where the information goes, where it will be stored and whether AI tools may be used. A signed engagement letter, a signed consent or a fact-find with consent can carry that permission, and tax file numbers bring the extra obligations of the TFN Rule. Supervision: if a junior uses an assistant to draft a position, the reviewer is responsible for what leaves the practice.

An engagement-letter paragraph you can adapt

This is illustrative wording for a practice to discuss with its own adviser, not a template to paste.

Use of AI tools. We may use artificial intelligence tools, including
business-grade assistants and features built into your accounting software,
to help prepare drafts, summarise documents and check data. We only use tools
that do not use your information to train their models, we limit the
information we enter to what the task needs, and a qualified member of our
team reviews all output before it is relied on or lodged. Some providers
process data outside Australia. If you do not agree to the use of AI on your
affairs, tell us in writing and we will not use it. [Practice name]

The letter is the formal part. What to tell clients about your AI use covers the conversation, and AI engagement letters for accountants covers drafting and scope.

ATO-facing review steps: what a person checks before lodging

Everything the practice lodges is the practitioner's responsibility, however the draft was made. A review routine that is written down and followed every time is also your best evidence of reasonable care. Here is one for a quarterly BAS prepared with AI-coded transactions.

1. Trace the totals. Check that the AI-assisted totals agree to the ledger reports, and that the ledger agrees to the bank statement for the period.
2. Sample the coding. Open a sample of AI-coded transactions, say 20 or 10% (whichever is larger), and check the account and the GST code against the source document. Open every transaction above a materiality line you set, such as $1,000.
3. Test the exceptions. Look at uncategorised items, new suppliers and anything the tool flagged as uncertain; these are where errors cluster.
4. Compare with last period. Investigate GST collected, GST paid and total sales if they moved by more than your threshold, for example 15%.
5. Check any research against source. If an assistant explained a tax treatment, find the ATO guidance, ruling or legislation yourself. Do not accept a summary or a citation you have not opened.
6. Record it. Note who reviewed, what AI feature produced the draft, what was changed and the date. The TPB's guidance asks practitioners to document each verification step, and it ties this to the record-keeping and quality-management duties.
7. Get approval. The client signs off the figures as you normally would before lodgement.

Ninety BAS a quarter: what ten saved minutes are worth

Illustration: a six-person practice prepares 90 quarterly BAS. Before AI, preparation averaged 75 minutes. With AI-coded bank feeds and an exceptions list, say preparation falls to 50 minutes, and a new review step adds 15 minutes, so the net saving is 10 minutes a BAS. Across 90 clients and four quarters that is 360 BAS and 60 hours a year. At an internal cost of $70 an hour (an assumption) it is worth $4,200 a year.

The costs: six assistant seats at $21 a month on Microsoft 365 Copilot Business billed annually (an add-on to a Microsoft 365 Business plan) come to $126 a month, or $1,512 a year, plus the time to set up the pilot, say 12 hours ($840). Net benefit in year one is about $1,850, before any gain from faster responses. The sum is thin on purpose. The real value is usually in the other work an assistant helps with, such as drafting client letters and summarising correspondence, and the discipline of the review step protects against the cost of one bad lodgement.

A filled-in review record and a realistic miss

Here is how the review record might look for one client after the steps above. The figures are illustrative.

Client: Café (sole trader), BAS quarter ended 30 September
Draft prepared by: Staff A using ledger AI coding on 214 bank lines
AI-coded lines opened: 22 (sample) + 6 over $1,000 = 28
Changes made: 3 (a refrigeration repair coded to cost of sales, a supplier
  invoice with GST missing, a personal purchase coded to supplies)
GST collected vs last quarter: +6% (within 15% threshold)
GST paid vs last quarter: +21% (explained: new freezer, invoice on file)
Research used: none
Reviewed by: Partner B, [date]. Client approval received: [date]

The miss worth learning from is the third change. The tool coded a personal purchase paid from the business account as supplies, because the description looked like stock. Nothing about the transaction was unusual to the software. It surfaced only because the reviewer sampled lines and asked the client. Sampling by value alone would have missed it, which is why the routine mixes a random sample with a value threshold.

After a few quarters, use the records to tune the routine. If the tool's changes cluster in one account, such as repairs and maintenance, add a standing check for that account and tell the tool's rules the correct treatment, so the same error does not return every quarter.

Jobs an assistant can do with little or no client data

Each of these can be done with a business-tier assistant and minimal data.

1. A client chaser with no client data. Prompt: "Write a friendly email asking a small-business client to send missing bank statements for the quarter, due in five working days. Placeholders for name and period." The output is usable after you cut the second paragraph, which will be longer than it needs to be.

2. A plain-English explainer. Paste a public ATO guidance page, not client facts, and ask for a one-page summary for a non-accountant. Always open the page afterwards and check the figures.

3. Exception triage on anonymised data. Strip names, account numbers and addresses from a short list of uncertain transactions, then ask the assistant to suggest likely categories and say what evidence would confirm each. The before-and-after shows what anonymising means.

BEFORE (do not paste):
12/09 EFTPOS J SMITH PLUMBING PTY LTD BSB 062-000 ACC 1234 5678  $1,430.00

AFTER (safe to paste):
12/09  Card payment, plumbing contractor  $1,430.00
Question: likely expense category and GST treatment, and what document
would confirm it?

How to anonymise client data before AI goes through the technique for batches.

4. A meeting note-taker, with consent. Ask the client before recording, state retention in the engagement letter, and set the tool to delete recordings after the notes are approved.

5. A workpaper review checklist. Ask the assistant to turn your written review steps into a checklist for each client type. It works from your steps, so no client data is needed.

The job that does not. Uploading a client's full tax return PDF, with tax file number and bank details, into a free chatbot to "check it". That combines sensitive information, an unsuitable tool and an unreviewed answer, and it is what the OAIC's guidance warns against.

Why free tiers stay closed to client data

Business plans from the major vendors do not train on your content by default. Free and personal plans differ and their settings move, so a practice cannot promise a client that data entered there stays private. Add that no one can check which staff have used a personal account, and the policy becomes simple: client information goes into business-tier tools or ledger AI only.

Write it down as one line in your staff policy: "No client information in any AI tool that is not on the approved list." Review the list each quarter, because vendors rename and re-plan their products often, and keep a record of when each tool was approved and what settings were chosen.

Which option for which practice

Four weeks from approved-tool list to standard review sheet

In week one, write the approved-tool list and the one-line staff rule, and check each current client engagement letter for AI wording. In week two, choose the ten pilot clients, send the consent request and set up the assistant with training off and retention set. In week three, run the pilot on one cycle of BAS using the review steps above and log time and errors. In week four, review the log with the whole team, decide what to keep, and put the review sheet into the workpaper file as standard.

If you would like help working out which jobs in your practice are worth automating first, that is what my AI implementation consultation covers.

Questions Australian practices ask about AI

Does the Privacy Act apply if my practice turns over under $3 million?

Possibly, yes. The OAIC treats businesses with turnover of $3 million or less as generally outside the Australian Privacy Principles, with listed exceptions, and the Privacy (Tax File Number) Rule 2015 still governs how a practice handles tax file numbers. The Tax Practitioners Board's confidentiality duty applies regardless of turnover. Ask your adviser to confirm your position.

Can I paste client bank transactions into ChatGPT?

Not on a free or personal plan, and not unless the data is anonymised first. The OAIC advises against entering personal information into public generative AI tools. A business-tier plan with training off by default, a clause in your engagement letter and minimal data is the safer route.

Do I have to tell clients I use AI?

If client information goes into an AI tool in a way that discloses it to a third party, the TPB's AI guidance says you need each client's permission first, and recommends telling them where the information goes, where it is stored and whether AI is used. A signed engagement letter can carry that permission, and your privacy policy should match it.

Can AI lodge a BAS or tax return for me?

AI can draft, code and flag, but the registered practitioner remains responsible for what is lodged. Treat every AI figure as unverified until traced to a source document, and record who reviewed it. The TPB says practitioners remain ultimately responsible for the services they provide, whatever tools they use.

Further reads

• AI Readiness Checklist for Accountants, Solicitors, Consultants — A readiness checklist before any tool goes live in the practice.
• Dext Pricing for Accountants: Essentials, Advanced and AI Assist — Plan and price detail for a common document-capture tool.
• Best AI Note Taker for Accountants: Which Fits a Small Practice? — Which meeting note-taker suits a small practice, with retention settings.
• Is Xero Smart Document Capture Enough Without Dext? — Whether Xero's built-in capture is enough without a separate tool.
• 7 AI Mistakes That Put an Accounting Firm's Client Trust at Risk — Seven ways AI can damage client trust, with fixes.
• AI Tax Research Tools Compared: Citations, Cost and Checks — Tax research tools compared on citations and cost.
• Best AI Tools for Chartered Accountants: Seven for a Small Practice — Seven tools for a small practice, outside the Australian rules.
• AI for Accountants in Australia: Tax Agent Duties and Client Data — The Code of Professional Conduct, client consent and Privacy Act steps for tax and BAS agents using AI, plus three safe first jobs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OAIC guidance on privacy and the use of commercially available AI products; OAIC small business page; OAIC APP 1 guidelines on automated decisions; Tax Practitioners Board guidance statement TPB(GS) 55/2026 (issued 22 July 2026); Xero Australia and MYOB Australia pricing pages (read November 2026); vendor pricing for Claude Team, ChatGPT Business, Microsoft 365 Copilot Business and Otter.