Guide

AI for Australian accountants and bookkeepers using Xero

How Australian accountants, bookkeepers and BAS agents can use AI with Xero: what JAX does at no extra cost, safe workflows, limits and Claude vs ChatGPT.

Want this done for you? Book a free 30-minute AI audit

In short

Australian accountants, bookkeepers and BAS agents can use AI with Xero for transaction coding suggestions, month-end prep, anomaly checks before a BAS is lodged, and first drafts of client reports and emails, while the registered agent still reviews, signs and lodges. Xero’s own AI, JAX, is currently available to Australian subscribers at no extra charge. On Autopilot, a Melbourne AI consultancy, builds AI Bookkeeping Assist for businesses on Xero: $1,500 setup plus $399 a month.

How can accountants and bookkeepers use AI in Australia?

Use AI for the preparation and drafting around compliance work, not for the compliance decisions. Sensible uses today are suggesting account codes and GST treatment, preparing month-end lists, flagging anomalies before a BAS goes to the agent, drafting client reports and emails, and summarising ATO guidance for you to verify. The registered tax or BAS agent stays responsible for every figure, opinion and lodgement.

JobWhat AI doesWhat stays with a person
Transaction codingSuggests a code and GST treatment, with a reasonAccepting or changing each code
Month-end prepLists uncoded items, suspense, duplicates, unusual GSTFixing them and closing the month
BAS-ready checksFlags what moved against prior quartersInvestigating; the agent reviews and lodges
Client reportsDrafts a plain-English summary from the P&LChecking every figure and any advice
Client emailsDrafts chasers and explanationsAnything that is advice
ATO researchSummarises guidance, with referencesChecking every reference on ato.gov.au

What is Xero JAX, and does it cost extra?

JAX is Xero’s built-in AI assistant, which Xero calls “your AI finance partner”. As at October 2026 it is available to all Australian Xero subscribers at no additional charge. It auto-categorises and matches bank transactions, answers questions about the financial data in the file, and drafts invoices and quotes. Because it works where the data already lives, it is the first AI tool most Xero users should switch on.

Xero plan (Australia)Price per month, including GST
Grow$78
Comprehensive$107
Ultimate$143

Prices from Xero’s Australian pricing page and JAX details from Xero’s JAX page, as at October 2026; Xero describes the no-charge access as current, so check before relying on it. JAX handles the work inside Xero; Claude or ChatGPT suits the work around it: client reports, emails, document review and research.

How do I use AI with Xero?

There are three practical ways. Use JAX inside Xero for coding suggestions, bank matching and questions about the file. Export data such as unreconciled transactions or a P&L and paste it into Claude or ChatGPT for analysis and drafting, with identifying details removed or on a business plan. Or have a connected agent read Xero through its API each day and queue suggestions for your bookkeeper to approve.

A safe first hour with Claude or ChatGPT:

  1. Pick the plan for the data. Client-identifying data belongs on a business plan or the API (see the privacy section below). For practice admin and anonymised analysis, a paid personal plan with training switched off is a reasonable floor.
  2. Switch off training on personal plans. Claude: Settings, Privacy. ChatGPT: Settings, Data controls, “Improve the model for everyone”.
  3. Load your context once. Create a Claude Project or ChatGPT custom GPT with your house style, standard chart of accounts and one anonymised sample report. Our beginner’s guide to Claude walks through Projects.
  4. Run a coding batch. Export 20 unreconciled transactions (date, payee, amount, reference) with client identifiers stripped, and ask for a code, GST treatment and one-line reason for each, flagging uncertain ones. Compare with your own coding.
  5. Draft one client report from a month’s P&L (prompt below), then edit it.

Measure the time and corrections on one client before rolling out further. A job that runs daily becomes a candidate for an agent; our guide to building AI agents safely explains when that is worth it.

Which AI workflows work for accountants and bookkeepers?

Six workflows reliably earn their place: coding suggestions, month-end prep, BAS-ready anomaly checks, client report drafts, client email drafts and ATO research summaries. In each one AI produces a suggestion or draft that a person reviews, and none of them lets AI change the ledger or send anything on its own. Start with the one that takes the most hours in your practice.

Transaction coding suggestions

Ask for the code, the GST treatment and a one-line reason. Mistakes tend to repeat for the same supplier, so turn each correction into a standing rule in your Project instructions, for example “subscriptions from [supplier] go to [account], GST on expenses”.

Month-end prep

Ask for uncoded transactions, suspense items, likely duplicates, unusual GST treatment (overseas purchases, capital items, mixed use) and lines that moved sharply against last month. The bookkeeper works from that list instead of hunting.

BAS-ready checks before the agent lodges

Compare the draft figures with prior quarters before the BAS goes to the registered agent. The ATO’s BAS guidance explains each label.

Check these BAS figures for anything that looks wrong before my BAS
agent reviews them. This quarter: G1 total sales $[ ], 1A GST on sales
$[ ], G11 non-capital purchases $[ ], 1B GST on purchases $[ ],
W1 total wages $[ ], W2 PAYG withheld $[ ].
Previous four quarters: [same fields].
Flag: 1A out of line with taxable sales, W2 out of line with W1, any
figure that moved sharply without an obvious reason, and any zero
where a number is expected. List anomalies to investigate. Do not
change any figures.

Client report drafts

Draft a monthly report for [client], an Australian [industry] business.
Data: P&L for [month], compared with last month and the same month
last year (pasted below).
Reader: the owner, who is not an accountant. Plain English; explain
any term you use.
Sections: one-line headline; revenue; the three biggest expense
movements; profit and cash; what to watch next month; two questions
for our next catch-up.
Under 600 words. AUD. Australian English.

Check every figure against Xero before the report goes out under your name.

Client email drafts

Document chasers before BAS time, plain-English explanations of a higher GST bill, and payment reminders. AI drafts; you decide anything that amounts to advice.

ATO research, verified

Ask for a plain-English summary with the ruling or page reference, then open each reference on ato.gov.au yourself. Both Claude and ChatGPT can misstate thresholds and rates or cite superseded guidance, so treat the summary as a lead, not an answer.

Can AI do my bookkeeping or BAS?

Not on its own. AI can suggest codes, prepare month-end lists and flag anomalies, but it cannot take responsibility for the books or the BAS. Providing BAS services for a fee is reserved for registered BAS and tax agents, and the Tax Practitioners Board warns of significant civil penalties for anyone doing so while unregistered. AI makes a registered agent faster; it does not replace one.

What AI must not do:

  • Lodge or sign. That stays with the registered agent.
  • Give tax advice without the agent. AI can summarise rules; applying them to a client’s affairs is the agent’s job.
  • Change the ledger unreviewed. Suggestions only, each approved by a person.
  • See tax file numbers. Never paste a TFN, bank details or identity documents into any AI tool.

The TPB’s Code of Professional Conduct applies whatever tools you use. Among other things, you must not disclose information about a client’s affairs to a third party without the client’s permission, must provide services competently and must take reasonable care. Weigh any AI provider against that confidentiality duty; it is why your plan choice and engagement letter matter.

Claude or ChatGPT for accounting work?

Both handle accounting drafting and analysis well, and the plan tier and your review step matter more than the brand. In general use, Claude’s long-form drafts often need less editing and it readily explains its coding suggestions, while ChatGPT is just as capable on structured tasks when you ask for its reasoning. Inside Word, Excel and Outlook, Microsoft 365 Copilot is the integrated option. Test both on one real job.

ClaudeChatGPT
Personal paid planPro, US$20/month (about A$33 including GST at checkout)Plus, US$20/month (about $33 incl GST)
Business planTeam, US$25 per seat monthly or US$20 annuallyBusiness, US$25 per user monthly or US$20 annually
Business plan data used for trainingNot by defaultNot by default
Keeping practice contextProjectsCustom GPTs
Client reports and long emailsDrafts tend to read naturallyGood drafts; default tone can be more corporate
Coding suggestionsUsually gives the reasoning unpromptedAsk explicitly for the reasoning

Prices as at October 2026 from Claude pricing and OpenAI’s ChatGPT Business page. Microsoft 365 Copilot Business is an add-on at A$31.40 per user a month paid yearly, ex GST (Microsoft). The quality rows are general observations, not a controlled test. For the wider picture, see our Claude vs ChatGPT guide for Australian small business.

Which AI plan is safe for client data?

Choose the plan by the data, not the price. Personal plans (Claude Free, Pro and Max; ChatGPT Free, Plus and Pro) may use your chats for training unless you switch it off, so keep them for practice admin and anonymised work. Client-identifying data belongs on a business plan (Claude Team or Enterprise, ChatGPT Business or Enterprise) or the API, which are not used for training by default (Anthropic, OpenAI).

  • Where data is processed. Anthropic’s own API has no Australian data residency (Anthropic). Claude on Amazon Bedrock with the Australian cross-region profile keeps processing in Australian regions for supported models (AWS). Sending personal information overseas engages APP 8 for practices covered by the Privacy Act.
  • Small practices. Turnover of $3 million or less generally means a Privacy Act exemption (OAIC), but the TPB confidentiality duty still applies.

The OAIC recommends, as best practice, not entering personal information into publicly available AI tools. Our AI privacy guide for Australian businesses sets out the plan tiers in full.

What should an engagement letter say about AI?

Say what you actually do: which tools and plans you use, for what kinds of work, that a person reviews the output, that responsibility stays with your registered agents, and that clients can ask you not to use AI on their file. Do not promise anything you have not checked, such as data never leaving Australia or never being stored by the provider.

A paragraph to adapt (delete any bracketed claim that is not true of your setup):

We use AI tools, currently [tool and plan, for example Anthropic’s Claude on a Team plan], to help with drafting, summarising, categorising transactions and reviewing documents. [Under the terms of these plans, your information is not used to train the providers’ models.] [These providers may process your information outside Australia.] Our staff review all AI output, and all professional judgement, advice, signatures and lodgements remain the responsibility of [practice name] and its registered agents. You can ask us how we use AI, or ask us not to use AI tools on your file.

Update your privacy policy to match, and revisit both when you change tools or plans.

How On Autopilot can help

On Autopilot builds AI Bookkeeping Assist for Australian businesses on Xero: $1,500 setup plus $399 a month (ex GST). It helps your existing bookkeeper rather than replacing them, with daily Xero categorisation suggestions, month-end draft prep, BAS-ready checks for your bookkeeper or BAS agent, and a plain-English monthly digest by Slack or email. It reads Xero and suggests; a person approves every change, and it never lodges anything. The Claude API cost is included, the build takes 10–14 working days, and it runs month to month. For an accounting or bookkeeping practice, the usual first build is a $497 Quick Start that chases client documents and signatures at tax time; AI for accounting and bookkeeping practices sets out the options. For several agents run for you, see managed AI, from $1,500 a month. Start with a free 30-minute AI audit with Jenn Yang.

Common questions

Will AI replace my bookkeeper?

Unlikely. Coding transactions is only part of bookkeeping: payroll, Single Touch Payroll, super, GST questions, client contact and year-end work with the accountant still need a person. What AI changes is the routine, with suggested codes, month-end lists and anomaly flags arriving ready for review, so the bookkeeper spends more time on judgement and less on data entry.

Do I still need Hubdoc or Dext if I use AI?

For most practices, yes. Receipt capture tools are built for volume: they extract the details, keep the source document and publish the bill into Xero. Claude and ChatGPT can read a clear receipt photo, but they do not run that pipeline. Keep your capture tool for the bulk and use AI for the odd ones, such as unusual invoices, supplier queries and statement reconciliations.

Can AI get the numbers wrong?

Yes. AI tools sometimes give a confident but wrong total, especially when adding long lists of figures, and can misread a column in pasted data. Check every total, percentage and date against Xero, ask the AI to show its working, and prefer workflows where figures are read directly from Xero rather than retyped. A person verifies before anything reaches a client or the ATO.

Does this work with MYOB or QuickBooks?

The chat workflows in this guide work with any accounting software, because they start from an export or report you paste in, either with identifying details removed or on a business plan. JAX is specific to Xero. On Autopilot's AI Bookkeeping Assist is built for Xero; if you run MYOB or QuickBooks, a free audit call can scope whether a similar build makes sense.

Want this built for your business?

Book a free 30-minute AI audit call with Jenn Yang. We map where your time goes, name the first systems worth building and give you fixed AUD prices. No obligation.

Book a free AI audit

Or have us run it for you, end to end: On Autopilot is Australia's outsourced AI department.