AI for accountants and bookkeepers: what is actually working

What is genuinely working now, what works only with supervision, and what should never go to a client or the ATO without a qualified human reading it.

A lot of what’s pitched to accounting and bookkeeping practices this year claims to think like a professional. Most of it doesn’t, and pretending otherwise is how practices end up with tools nobody trusts and nobody uses past the second week. The more useful question isn’t “can AI do accounting” — it’s “which parts of a normal week are genuinely mechanical, and which parts need a qualified person’s judgement.” Once you draw that line, a surprising amount becomes worth automating, and a smaller amount than the marketing suggests.

The shape of the work matters

Practice work isn’t evenly spread across the year. There’s a quiet stretch, then a compliance season where everything lands at once, then client meetings clustered around EOFY and BAS deadlines. A lot of what makes those peaks painful isn’t the technical work — it’s the chasing. Chasing a client for their bank statements. Chasing a signature. Chasing the same missing receipt for the third time. That chasing is repetitive, low-judgement, and exactly the kind of task that tools handle well without putting anything sensitive at risk, because nobody’s relying on the tool to get an answer right — they’re relying on it to keep asking.

That’s worth more than it sounds. Automating the thinking — the technical judgement that makes you a professional — is harder, riskier, and frankly not where the time goes most weeks. Automating the chasing gives staff back hours during the exact weeks they have none to spare, without touching anything that needs a qualified eye.

Genuinely working now

Some uses are dull, low-risk, and already reliable enough for daily use in a small practice.

• Client onboarding and document chasing. Automated reminders and checklists that follow up missing information without a staff member manually tracking who owes what.
• Turning source documents into structured data for review. Invoices, receipts and statements converted into a format a bookkeeper can check quickly, rather than typed in line by line.
• Drafting routine client correspondence. A first draft of a standard email — a reminder, a request, a thank-you — that a staff member edits and sends under their own name.
• Summarising a client’s year into talking points before a meeting. Pulling together what changed, what’s due, and what needs discussing, so the meeting starts from a summary instead of a blank page.
• Searching your own precedent and advice library. Finding the memo you wrote for a similar situation two years ago, instead of digging through folders.
• Meeting notes into file notes and action items. Turning a recorded or typed conversation into a clean file note and a list of who’s doing what.

None of these involve a tool deciding anything. They involve a tool doing the clerical half of a task so a person can do the professional half faster.

Working with supervision

A second tier is useful, but only with a human checking the output before it goes anywhere.

• First-pass classification and coding. A tool suggests how a transaction should be coded; a person reviews and confirms it before it’s final. This can speed up reconciliation noticeably, but the review step isn’t optional — it’s where the actual accounting happens.
• Drafting advice that a qualified person rewrites and signs. A first attempt at an explanation or a piece of correspondence that touches on advice, which a qualified staff member then checks, corrects and puts their name to. The draft saves time getting started. It is not the advice.

The test for this tier is simple: would you be comfortable if the client found out a machine produced the first draft, provided a qualified person genuinely reviewed and changed it before it went out? If yes, it’s supervision. If the honest answer is that nobody really checked it, it’s not supervision — it’s just unsupervised work with an extra step.

Not appropriate

Some things don’t belong in this workflow at all, at least not yet, and probably not ever without a person in the loop.

• Anything that goes to a client or the ATO without a qualified human reading it first.
• Anything that constitutes advice generated and sent without a qualified person taking responsibility for it.

This isn’t caution for its own sake. A practice’s value is the judgement it takes responsibility for. Handing that off, even quietly, even just once, undermines the actual product a client is paying for.

Professional obligations don’t disappear

None of this changes who’s responsible. You remain responsible for work that goes out under your name or your practice’s name, regardless of what produced the first draft. Confidentiality obligations don’t relax because a tool was involved in preparing something — client information is still client information, and where it goes and how it’s stored still matters. And whatever your professional body and the ATO expect of practitioners generally, those expectations don’t pause for a new tool. If you’re unsure whether a particular use fits within them, that’s a conversation with your professional body or your own compliance processes, not something to guess at.

A sensible first project

If your practice has never automated anything, don’t start with anything client-facing and don’t start with anything involving advice. Start with document chasing during onboarding — a simple, automated sequence that follows up new clients for the standard documents you need, tracks what’s arrived, and tells a staff member when someone’s gone quiet. It touches no judgement, it’s low-risk if something goes wrong, and it fixes a problem every practice already has. Get that working well, see how it feels to trust it, and only then look further up the list.