AI can do your Australian tax. Should you trust it?

A finance director in London, Auckland or Austin can now upload their Australian subsidiary's trial balance to an AI tool and ask it to draft the company tax return. They can ask whether their sales team creates a permanent establishment, draft an intercompany services agreement, or produce a first cut of transfer pricing documentation. 

Much of what comes back is good. Some of it is very good. So, the question is fair: if AI can do this, why pay an Australian accountant? 

We look after around 700 foreign-owned Australian companies from more than 50 countries, and we hear some version of that question more every month. We don't think the right answer is to argue that AI can't do the work. It can do a lot of it, and we use it heavily ourselves. 

The better answer is a clear way to decide three things: what you can sensibly do yourself, what you should hand over, and what deserves someone prepared to stand behind the answer.

What AI is taking off our plate (and should)

A good share of traditional accounting work is becoming a commodity. Pulling data out of documents, coding transactions, reconciling accounts, researching a rule and writing a first draft are all faster with AI, and they will keep getting faster. 

We've built that into how we work. Our team across Australia, the UK and the Philippines uses AI in BAS preparation, workpaper review and client file reviews. The result is that our people spend less time assembling information and more time checking it. 

If a task used to take ten hours and now takes two to the same standard, that should matter to clients. We'd rather say so plainly than defend hours that no longer need to be worked. 

But producing something is not the same as knowing it's right. AI can produce a tax return. The harder questions are whether it found the right issues, had all the facts, applied current law, and reached a position that someone will stand behind if the ATO asks about it three years from now.

Why foreign-owned subsidiaries are a special case

For a purely domestic small business, AI plus a good bookkeeper may cover a lot of ground. A foreign-owned subsidiary sits where two tax systems meet, and that Australian layer is where generic AI answers are most likely to miss something. 

The issues we see most often:

• Related party dealings. Service fees, cost recharges, royalties and loans with the parent all need an arm's length position, and often an International Dealings Schedule with the return.

• Withholding tax. Whether a payment offshore is a royalty, interest or a service fee changes the outcome, and treaty relief depends on the specific treaty.

• Permanent establishment. What your people actually do in Australia, and who signs contracts, can create a taxable presence for the parent.

• GST. Imports, cross-border services and registration, where the answer turns on who the customer is and where they are.

• Group-level rules. If the global group is a significant global entity, higher penalties and extra reporting can apply, even when the Australian company is small.

• Directors and ASIC. The company needs a resident director, and carries ASIC and record-keeping obligations that sit outside tax entirely.

AI often handles each rule reasonably well on its own. Where it struggles is knowing which of them apply to you. That depends on facts held in two countries, by people who may never have spoken to each other.

Four questions before you rely on AI

The lower your answers to these, the stronger the case for doing it yourself. As they rise, independent professional involvement becomes more valuable.

1. How complex is it?

Some questions are settled law applied to simple facts. AI can help an informed finance team deal with them well. 

Others depend on legislation, a tax treaty, ATO guidance, contracts and detailed facts all at once. 'Do we have a permanent establishment in Australia?' sounds like one question. In practice it turns on what staff do here, where deals are negotiated, what authority people hold, whether there are premises, and which treaty applies. AI can speed up that analysis enormously. As ambiguity grows, so does the value of experienced judgement. 

2. How much judgement does it need?

AI is extraordinarily good at producing plausible answers. Plausible is not the same as right.

Take transfer pricing. Your team may describe the business better than any adviser, because they run it. 

AI can turn that into a polished document. The hard part is not the writing. It's whether each entity is characterised correctly, who really controls the risks, whether the contracts match what happens, and whether a recharge with no markup holds up. The document is getting cheaper. The judgement behind it isn't getting less important. 

3. What's the exposure?

The same GST question can affect $50,000 of sales for one company and $20 million for another. Same law, very different decision. 

Now suppose there's a 10% chance an AI answer is materially wrong. That's fine if the downside is $500. It isn't if the downside is $2 million, several years of amended returns and a dispute with the ATO. 

For subsidiaries, exposure also travels upstream. An Australian problem can surface in the group audit, an investor's due diligence or the sale of the business, often years later when the people who made the call have moved on. 

4. Who carries the responsibility?

There is a big difference between 'can you look over what we've prepared?' and 'can you tell us whether this position is supportable, so we can rely on it?' The first is a review within a set scope. The second means we do enough work to reach our own professional conclusion. 

For many of our clients, responsibility isn't abstract. When we lodge a return, we do it as a registered tax agent bound by a professional code of conduct. When we provide a resident director, an Australian individual takes on personal duties as a director of your company. No AI tool sits on your board. AI can shift a lot of preparation from us to you. It doesn't shift the responsibility for a conclusion you want us to endorse. 

The fifth question: whose time is this using?

Being able to do something yourself doesn't mean you should. AI has lowered the barrier to attempting professional work. It hasn't added hours to your week. 

For our clients there's a twist. The person doing the AI work usually isn't in Australia. It's a group CFO, a financial controller or a founder overseas, juggling several countries in a different time zone, with Australia as one subsidiary among many. 

So the real comparison isn't $0 with AI against $5,000 with an adviser. It's this:

The cost of professional advice, against the value of your time, plus what else you could have done with it, plus the risk that's left if you get it wrong.

Say your controller spends six hours building an Australian transfer pricing position with AI and saves $3,000 in fees. That can be entirely rational. But if those six hours belonged to the group month-end close, a board pack or your next market entry, the saving may not be a saving at all. 

Preferences matter too. Some founders enjoy the detail and want to own it. Others want to understand the commercial impact and leave the legislation to someone else. Neither is wrong. The useful question is where your involvement creates the most value. 

Three ways to work with us

AI makes the service model far more flexible, and we think that's a good thing. Most clients end up somewhere on this spectrum, and different parts of the same engagement can sit in different places. 

Where a question falls outside our expertise, we say so and bring in specialists we trust. Knowing where your own edge ends is part of good judgement, for us as much as for AI.

Two examples from our world

Transfer pricing

A group needs Australian transfer pricing documentation. Its finance team knows the business well and uses AI to map the group structure, describe the intercompany dealings, explain how costs are allocated and draft a functions, assets and risks analysis. 

That can remove a lot of traditional professional work, and we won't recreate good work just to generate fees. Instead, our tax team concentrates on testing it:

• Does the functional analysis reflect what actually happens?

• Is the method right, and is the pricing arm's length?

• Is benchmarking needed?

• Does the Australian result line up with the International Dealings Schedule and the parent's own records?

• Where would an ATO review find gaps? 

A pattern we see often: an Australian distributor described as low risk that reports losses year after year. AI will document that arrangement neatly. It won't necessarily tell you the two facts don't sit comfortably together. 

The tax return

Sooner or later a client will tell us, 'We've already prepared the return with AI.' 

Our answer isn't that AI can't prepare tax returns. It's a question: what would you like us to do? If the affairs are simple and you're happy to lodge it yourselves, that's a legitimate choice.

If you want us to lodge it as your tax agent, we need to be professionally comfortable with it. That means reconciling to the accounts, reviewing international dealings and withholding, testing related party transactions and looking hard at anything unusual. The cost of preparing a return has fallen a long way. The work needed to stand behind one hasn't fallen nearly as far. 

What this means for our fees

If technology genuinely reduces the work, clients should share in that. Charging for historical effort while doing far less work isn't a sustainable way to price. 

Nor do we think advice should be priced on the minutes it takes to produce an output. Our thinking is increasingly: 

Professional fee=Efficient cost of delivery×Engagement factor

The delivery cost reflects what the work genuinely needs, including what AI and automation save. The engagement factor reflects what time doesn't capture: the same things we weigh whenever we price advisory work, namely scope, complexity, exposure and the responsibility we're taking on. 

In practice, routine compliance stays on fixed monthly fees. As AI lowers what that work costs us to deliver, we'd rather put the gain into faster turnaround and more proactive review than into bigger invoices. Advisory work is scoped and priced on its own merits, so a simple question stays affordable and a high-stakes one gets the attention it needs. 

A quick way to decide

Most decisions land somewhere in between, which is why we expect AI and advisers to work together rather than one replacing the other. 

Where ABN Australia fits

We don't think our future lies in charging people to fill in forms. Technology should do more of that, and basic setup and compliance will keep getting cheaper across the market. 

Our value is concentrated in five things:

• Understanding. Knowing your business, your group and why things are set up the way they are, not just the numbers.

• Judgement. Applying Australian law where there isn't an obvious answer.

• Challenge. Testing assumptions, including the ones AI produces, rather than documenting what management already believes.

• Responsibility. Putting our professional name behind the work, within an agreed scope, and providing resident directors who take on real duties.

• One team for Australia. Tax, accounting, payroll, ASIC and resident director services under one roof, so the facts don't fall between providers in different countries. 

That changes what an accountant does. Less time producing information, more time evaluating it. Less time filling in returns, more time spotting the positions buried inside them. And more time helping you answer the questions that matter: what does this mean, what are the risks, what should we do, and is this worth your own time? 

The question has changed

It's no longer 'can AI do this?' Increasingly, the answer is yes. 

The better question is: what's the right mix of technology, your own time and professional expertise for this particular decision? Sometimes it'll be AI. Sometimes an adviser. More and more often, both. 

We use AI ourselves, we're happy for clients to use it where it makes sense, and we focus our people on the places where expertise and responsibility genuinely change the outcome. That's what good delegation has always been about: not whether you can do it yourself, but whether doing it yourself is the best use of your time, attention and appetite for risk.