AI for Accountants in Australia: A Playbook

AI for accountants works best on the unglamorous parts of a practice: chasing client documents, coding bank transactions, drafting routine emails and finding answers in your own files. It works worst when asked to prepare a tax return unsupervised. For a 5 to 30 person Australian practice, start with document collection and client communications, keep a qualified human on every output, and follow the Tax Practitioners Board’s July 2026 guidance on client confidentiality.

TL;DR: Adopt AI by workflow, not by tool. Document collection and client comms return time fastest with the least professional risk, while tax positions and lodgments always need human sign-off. First, get client permission for AI use in your engagement letters and keep client data (especially TFNs) out of consumer AI tools.

AI for accountants starts with workflows

Most “best AI for accountants” lists rank products, which is the wrong unit. A practice has a queue problem, not a tool problem: the client who has not sent their bank statements, the 400 uncoded transactions, the BAS waiting on one missing invoice. Tools change every quarter. The queues stay put.

Here is our map of the main workflows. The tools are examples, not endorsements, and several features below are still in beta.

The risk ratings are our judgement. The rule behind them: the closer an output gets to a lodged number or a client’s tax position, the less autonomy the AI gets.

Where the time comes back first

Our stance: the biggest win is document collection and client communications, not “AI doing the tax return.”

Look at where a principal’s week actually leaks. Rarely the analysis. It is the fourth reminder email, the client who sends a photo of a bank statement upside down, the Monday triage of who owes what. That work is repetitive, low-stakes and text-heavy: what language models handle well. A bad draft chase-up email costs you a rewrite. A bad tax position costs you a client, and possibly a TPB investigation.

Karbon’s AI can compose and refine emails, summarise long threads and build a client summary from emails, notes, work and billing. Karbon says this is currently included at no additional cost, with advance notice promised if that changes. Its Kai AI coworker is in early access and meeting intelligence is marked coming soon: roadmap, not capability.

On the ledger side, Xero says JAX chat is available to all Xero subscribers at no additional charge for now, and that it does not use your data to train the models. MYOB’s AI BAS, Smart Reconciliation and AI Business Insights are all in beta. AI BAS, open to Lite and Pro customers, flags missing receipts and unusual GST, and MYOB says the customer owns decisions made on its suggestions. Dext’s partner plans are priced per client per month, excluding GST, with a 10-client minimum. AI Assist appears in both tiers, but the comparison lists a per-client charge on that row, so get a written quote.

For the accounts-payable side, see our guide to invoice automation in Australia.

Pick one workflow and test it properly

The failure we see most in professional-services rollouts is tool sprawl: four subscriptions, no owner, no measurement. Pick your highest-volume workflow, time it as it runs today, and decide what “good” looks like before switching anything on.

That is the job of our AI Quick-Win Audit: a fixed $2,950 (AUD) paid audit that maps one workflow, identifies the automation opportunities and shows a live working demonstration. The fee is credited toward any build of $10,000 or more commissioned within 3 months, and it starts with a free 30-minute fit call. In a practice that workflow is usually document collection or month-end coding. A build that follows sits under our process automation service.

Your professional obligations do not change

AI does not come with a new rulebook. Regulators have been clear the existing one applies.

Tax practitioners. On 22 July 2026 the Tax Practitioners Board issued TPB(GS) 55/2026, The use of Artificial Intelligence and the Code of Professional Conduct. It creates no new obligations, applying the existing Code to competence, reasonable care, confidentiality, record-keeping, professional judgement and supervision. The guidance says practitioners “are still ultimately responsible for the tax agent services they provide to their clients,” and that AI output must pass professional judgement before it is relied on. Practitioners must get each client’s permission before divulging client information to a third party, which the TPB says can include entering it into AI tools, depending on how they are configured and used. A signed engagement letter can record it. Where TFNs are involved, additional obligations under the Privacy (Tax File Number) Rule 2015 apply, and the TPB tells practitioners to seek their own advice.

Accountants generally. APES 110, the Code of Ethics for Professional Accountants, was amended on 20 June 2024 to add technology-related considerations to professional competence and due care and to confidentiality, effective 1 January 2025. CPA Australia’s August 2025 commentary was blunt: do not upload client data into a public-facing AI tool like ChatGPT, be upfront with clients about AI use, and consider spelling it out in engagement letters. Chartered Accountants ANZ publishes an AI ethics and governance resource and made a submission on the TPB’s draft guidance.

Privacy. The OAIC’s guidance on commercially available AI products recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools. Client TFNs, dates of birth and bank details all qualify, and TFNs carry extra rules.

The practical rules:

• No client names, TFNs or financials in consumer chatbots, on personal logins, ever.
• Use business-tier tools with contractual data protections. Our AI vendor data privacy checklist lists the questions to ask.
• Put AI use, and where data is processed, in engagement letters.
• Write these rules down. Our AI policy template for Australian businesses is a starting point, and staff quietly using personal accounts is the shadow AI problem in its most expensive form.

For the wider advisory-firm view, see AI for professional services in Australia.

A 90-day adoption plan for a 5 to 30 person practice

This adapts our 90-day AI pilot approach for a practice.

Days 1 to 30: policy and baseline. Name one owner, a partner or practice manager, not “everyone”. Inventory the AI features already inside Xero, MYOB, Karbon or Microsoft 365. Update engagement letters and publish a one-page AI policy. Record one workflow’s baseline: hours per week, turnaround, error rate.

Days 31 to 60: pilot. Two or three people run it with AI. For document collection, that means AI-drafted requests and reminders sent from your practice management tool. For coding, AI suggestions with mandatory review of exceptions and a sample of accepted lines. Log every error the reviewer catches. That log supports the record-keeping the TPB expects and doubles as training material.

Days 61 to 90: decide and extend. Compare against the baseline. If the numbers hold, roll it out firm-wide and pick the next workflow. If not, stop. A pilot that ends in “not worth it” is cheaper than a firm-wide subscription nobody uses.

For a team-wide licence decision, our guide to Microsoft 365 Copilot for business in Australia covers cost and payback. If you want an outside view on where to start, get in touch.

What we would not do

Do not let AI prepare and lodge returns or BAS without a registered practitioner reading every figure. Do not treat a confident answer on tax law as a source; check the ATO and the legislation.

AI for accountants is a workflow discipline before it is a technology purchase. Fix one queue with a registered practitioner still signing off, measure it honestly, and only then pick the next.

Frequently Asked Questions