What’s actually changed for Australian small businesses

Five years ago, “AI in accounting” mostly meant clever bank rules and a decent OCR scanner. Today, the same phrase covers large language models that can read your inbox, draft a chase email, explain a BAS line in plain English, and reconcile three months of bank feeds while you make a coffee.

For an Australian small or medium business running on Xero or MYOB, that shift matters. The big platforms have added their own AI features. A wave of third party tools plugs into the same data. And on top of that, owners are using general AI tools like ChatGPT, Claude, and Gemini to write policy docs, summarise board packs, and draft responses to ATO letters.

None of this replaces a good bookkeeper or accountant. What it does replace is a stack of repetitive tasks that used to eat half a Tuesday. The question for most owners I speak with is no longer “should I look at AI” but “where do I start, and what do I need to check before I let it near the books?”

This guide is meant for owners and operators, not software buyers at a 500-seat firm. I’ll keep it grounded.

Where AI already lives inside Xero and MYOB

If you’re already paying for Xero or MYOB, you’ve got AI features switched on whether you realise it or not. Worth knowing what they are.

In Xero, the built-in tools include automated bank reconciliation predictions, Hubdoc for receipt capture, Analytics Plus for cashflow forecasting, and the newer Xero AI assistant that can answer natural language questions about your numbers. Short queries like “what did I spend on subcontractors last quarter” are exactly the sort of thing it’s meant for.

MYOB has moved in a similar direction. Their AI features include smart bank rules, Auto-super for the SuperStream side of payroll, Smart Bills for invoice extraction, and an in-product assistant called Mia that walks you through tasks. Both platforms are also leaning into predictive cashflow and anomaly detection, flagging transactions that look unusual.

The upside of staying inside your accounting platform is obvious. Your data stays in one place, the features are covered by your existing subscription (mostly), and the vendor is responsible for the security and privacy setup. The downside is that the AI is only ever as good as the data you’ve already put in. Garbage in, garbage out, as the saying goes. If your chart of accounts is a mess, AI won’t fix that. It will simply speed up the mess.

Where third party AI tools fit alongside your stack

Beyond Xero and MYOB, there’s a thicket of add-ons. Some sit on top of your existing data through official APIs. Others operate as standalone tools that you copy and paste into.

The most useful categories we typically see adopted by AU small businesses are:

Receipt and bill capture tools like Dext (formerly Receipt Bank) that push straight into Xero or MYOB. AI category coding and approval workflows. Cashflow and forecasting tools such as Fathom, Spotlight Reporting, and Float. Practice management tools for accounting firms including Karbon, Tidy, and Botkeeper style platforms that use AI to draft client responses and prep workpapers.

Then there’s the general purpose layer. Many of the owners in our network are now using ChatGPT, Claude, or Gemini to draft everything from staff handbooks to ATO correspondence. A Sydney based tradie operator I spoke with recently said he gets Claude to write his progress claims and then has his bookkeeper check the math. That pattern, AI drafts, human reviews, is the right one in my view.

The catch is that these third party tools often sit outside the Xero or MYOB security perimeter. When you connect them, you are moving financial data somewhere else. That single fact triggers most of the questions that follow.

The data and privacy questions you can’t skip

Australia’s privacy framework is the Privacy Act 1988 (Cth) and the 13 Australian Privacy Principles (APPs). For most small businesses with turnover under $3 million, the Act does not apply directly, but if you handle health information, provide services to APRA regulated entities, or hold certain kinds of records, you can still be caught. Don’t assume you’re exempt. Have a quick chat with your accountant or lawyer on this one.

If the Act does apply to you, here are the bits that matter most for AI tools.

APP 8 limits cross border disclosure of personal information. If you send customer or staff data to a tool hosted overseas, you generally need to take reasonable steps to ensure the overseas recipient handles it in line with the APPs. Many AI vendors will sign a contract that addresses this, but you need to actually ask. A lot of owners in our network didn’t realise their free ChatGPT account was sending prompt content to US servers until they read the fine print.

The Notifiable Data Breaches scheme applies to businesses covered by the Privacy Act. If a breach is likely to cause serious harm, you need to notify the Office of the Australian Information Commissioner and affected individuals. Letting an AI tool train on your customer list, then discovering that data is sitting in a shared model, can be a problem.

If you’re using AI for HR, marketing, or anything that touches personal information, the same rules apply. A useful rule of thumb is that anything you wouldn’t put on a postcard to the US needs a hard look before it goes into a general AI tool.

ASIC, APRA, and AHPRA: what changes for your industry

For most small businesses, the Privacy Act is the headline framework. But three other regulators come into play depending on what you do.

ASIC Regulatory Guide 265 applies if you hold an Australian Financial Services Licence or provide financial advisory services. It sets expectations for information security, including how you handle data going to third party providers. The short version is that if you use AI to process client data, you need to be able to show that you assessed the risk and the vendor can meet the standard.

APRA CPS 234 is the information security standard for APRA regulated entities such as banks, insurers, and super funds. Most of you reading this aren’t APRA regulated, but if you supply services to a bank or super fund, you may have CPS 234 obligations written into your contract. AI tools that touch client data often need to be on an approved supplier list.

AHPRA registered practitioners and healthcare practices have their own considerations. The AHPRA codes of conduct and the Australian Privacy Principles both apply. If you’re using AI scribes, transcription tools, or anything that handles patient notes, the privacy bar is high. Again, verify the specific obligations with your lawyer or a specialist health sector advisor. The rules evolve quickly.

The reason I’m flagging all three is the same reason I’d flag the NZ Privacy Act 2020 and its 13 Information Privacy Principles to a New Zealand owner. The accounting work itself is rarely the hard part. It’s the data going into the AI and coming out the other side that needs a proper risk check.

What this typically costs an Australian small business

Pricing moves around, so treat these as rough guides based on what we typically see in 2026. Always check the vendor’s site for current rates.

Xero plans in Australia run from about $32 a month for the Starter tier, around $62 for Standard, and roughly $79 for Premium, plus per-user fees. MYOB sits in a similar range, often between $50 and $90 a month depending on the plan. Add ons like Hubdoc, Analytics Plus, or payroll modules cost extra.

On the AI side, expect to pay around $20 to $60 a month for a single user ChatGPT Team or Claude Team style subscription. Specialist accounting AI tools often sit in the $50 to $200 a month range per user, with firm wide pricing for accounting practices running much higher. The cost is rarely the issue. The issue is whether you’ll actually use it.

For a business with a few transactions a month, the ROI on a $300 a month AI accounting stack is questionable. For a business doing 5,000 transactions a month across multiple entities, the same stack can pay for itself in the first week.

How to roll it out without breaking the books

The pattern that works in almost every AU small business I’ve worked with is the same. Start small, keep a human in the loop, and document what you did.

Pick one workflow. Bank reconciliation is usually a good starter because the data is already inside Xero or MYOB, the rules are clear, and the worst case is a wrong code that you can fix. AI tools that draft chase emails to overdue customers are also low risk, because you can read the email before it goes.

Don’t give the AI everything on day one. The temptation is to plug in the API key, point the tool at your whole history, and let it loose. That’s how you end up with miscoded journals, weird forecast charts, and a slightly worried feeling. Stage the data, test on a slice, expand once you’re comfortable.

Keep a clear audit trail. If you’re a regulated business, your reviewer needs to be able to see what the AI was asked, what it answered, and what the human then decided. This is the kind of thing ASIC and APRA examiners look for. It’s also just good practice. Six months from now, you won’t remember why that journal was posted.

Train the people who’ll actually use the tool. The single biggest reason AI accounting projects stall is that the bookkeeper or ops manager wasn’t brought along. They see the tool as a threat, or they don’t trust the outputs, or they keep doing things the old way. Spend real time on this.

Common traps with Xero and MYOB integrations

A few things we see go wrong, often enough that they’re worth calling out.

Auto posting journals without a review queue. Some tools will let you set “post automatically” as the default. Don’t, at least not at first. The default should always be “draft, then review.”

Giving the AI access to payroll and HR data before you’ve stress tested the privacy side. Payroll contains tax file numbers, super details, and bank account information. Treat it as sensitive until proven otherwise.

Letting the AI draft client facing communication without your tone of voice. Customers can tell when an email sounds like it was written by a machine. Use the AI for the first 80 percent, then rewrite the last 20.

Assuming the platform handles everything. Xero and MYOB publish their own security and privacy commitments, but once you connect a third party tool, you inherit that vendor’s practices. Read their privacy policy. If it doesn’t have one, that’s your answer.

Forgetting that customers and staff have rights. If you’re using AI on data that includes personal information, there are notification and consent considerations. The Privacy Act and the APPs set the floor. Your industry code of conduct may set a higher bar.

Questions to ask any AI accounting vendor

Here’s a short list I give to owners before they sign anything. If a vendor can’t answer these clearly, walk away.

Where is the data stored, and where is it processed? Australia, US, Singapore, or somewhere else matters for APP 8 purposes.

Is our data used to train your models? The answer should be no, and it should be in writing. Many consumer tier products will say yes unless you opt out.

What happens to our data if we cancel? Can we export everything in a usable format, or are we stuck?

Who’s liable if the tool produces a wrong BAS figure, a wrong payroll calculation, or a wrong tax position? Read the contract.

Do you comply with the Privacy Act 1988 and the Australian Privacy Principles? A vendor that handles AU financial data should have a clear answer.

Can we run a small pilot, with a defined scope, before we commit to a 12 month contract? Any vendor worth working with will say yes.

Have you been through an independent security review? SOC 2, ISO 27001, or an equivalent audit gives you something to point at.

If the vendor is vague, evasive, or sends you to a marketing page, treat that as the answer.

When AI accounting is overkill

Not every business needs any of this. If you’re a sole trader with 30 transactions a month, paying $100 a month for an AI accounting layer is bad arithmetic. A good bookkeeper and your existing Xero or MYOB plan is plenty.

If your accounts are simple because your business is simple, the biggest wins usually come from cleaning up your chart of accounts, getting your bank feeds right, and having a real human look at the numbers once a month. AI is a multiplier on quality, not a substitute for it.

It’s also the wrong tool if the people using it don’t want to. Forcing AI on a long standing bookkeeper who’s about to retire is a fine way to lose them. The best tooling in the world fails when the team doesn’t trust it.

Bringing it back to your team and your customers

The thing I keep coming back to with the AU owners in our network is that AI accounting isn’t a software decision. It’s a trust decision. You’re trusting a tool to read your numbers, your customers, and your staff records. Your customers are trusting you to handle their data well. Your staff are trusting you to bring them along, not replace them with a chatbot.

The good news is that the technology has caught up with what most owners actually want. The boring, repetitive work is the part that’s easiest to automate. The judgment heavy work, the conversations with customers, the tricky tax calls, the strategy for next year, all of that still belongs to you. AI just gives you more of your week back.

Start with one workflow. Keep a human reviewer. Document what you did. Ask the vendor the hard questions. And treat your accountant as a partner, not a vendor, because the regulatory landscape in Australia is shifting faster than most of us expected.

If you want a structured way to think through where AI fits in your own business, including a simple rollout framework, the team at Enterprise DNA has put together a free field guide. Enterprise DNA works with NZ and AU businesses on this challenge. Get the free Working With Claude field guide at https://enterprisedna.co/resources/working-with-claude?utm_source=edna-landing&utm_medium=blog&utm_campaign=nzau

Whatever you do, don’t let the marketing convince you that AI accounting is a finished product you can buy off the shelf. It’s a set of tools you’ll spend the next few years learning to use well. That’s not a bad place to be. It’s just not a magic one.