AI accounting: what the software can do, and what the law keeps human
The capability is real and moving fast. So is the boundary: in Australia, tax agent services provided for a fee must come from a registered practitioner — and that line is where trust in AI accounting is actually built.
The short answer
AI can now run much of the machinery of business accounting — transaction processing, reconciliation, draft reports, variance flags and payables workflows. What it cannot legally or sensibly do is the other half: providing tax agent services for a fee requires registration with the Tax Practitioners Board, and applying tax law to your specific circumstances remains professional judgement, not pattern matching. The businesses getting real value treat AI as the processing layer underneath a registered, accountable human team — not as a replacement for one.
THE BASICS
What is AI accounting?
AI accounting is the use of artificial intelligence across the accounting function — transaction processing, reconciliation, reporting and first-draft analysis — with qualified people responsible for judgement, tax positions and advice. It extends AI bookkeeping upward: not just recording transactions correctly, but assembling them into reports, spotting movements worth explaining and preparing the groundwork that accountants and advisers then apply judgement to.
The capability arrives through the same channels as AI bookkeeping. Accounting platforms — Xero, MYOB, QuickBooks — embed machine learning in coding, matching and increasingly in report generation and query tools. Specialist tools layer on top for payables processing, forecasting inputs and management reporting. And accounting firms and back-office providers use AI internally, so even a business that never buys an AI product is often receiving AI-assisted accounting already.
The core tension in AI accounting is not technical, it is professional. Software can calculate anything; it cannot be accountable for anything. Australian law draws the line explicitly: tax agent services provided for a fee must come from a practitioner registered with the Tax Practitioners Board. AI can prepare, suggest, draft and flag — but the tax position taken, the advice given and the return lodged carry a human professional's responsibility, and the TPB's 2026 guidance makes clear that responsibility cannot be delegated to a model.
Read that boundary as a feature, not a limitation. It means an Australian business adopting AI accounting is not choosing between automation and accountability — the regulatory settings force the combination. The practical question is simply whether your setup honours it: AI on processing, registered professionals on judgement, and a clean handover between the two.
CAPABILITY
What AI does well today
Across the accounting function, these are the jobs AI already does at production quality — as an input to human review, not a substitute for it.
Transaction processing at scale
Coding, matching and reconciling across high transaction volumes and multiple accounts — the foundation layer of accounting — is now largely automatable, with consistency a manual process cannot match. This is the same engine that powers AI bookkeeping, running underneath everything else.
Accounts payable and receivable workflows
AI reads incoming supplier invoices, extracts the data, matches them against orders and payments, and flags mismatches and likely duplicates before money moves. On the receivables side it tracks who owes what and drafts the follow-up. People approve; AI prepares.
Draft management reports
Monthly profit and loss, cash summaries and management packs can be assembled by AI from a live ledger — including a first-draft written commentary on what moved. A qualified person still decides what the numbers mean, but the assembly work stops consuming their hours.
Variance and anomaly detection
AI compares this month against the pattern of previous months and flags what a human should look at: a cost line that jumped, revenue that landed differently from the trend, a margin quietly drifting. It turns 'review the accounts' from a blank page into a short list.
Preparation for compliance work
Assembling the figures behind a BAS, organising the records that support a tax return, reconciling clearing accounts before period close — AI does the gathering and first-pass checking that used to be the slowest part of every compliance job, so the registered professional's time goes into review and the position taken.
Answering questions about your own numbers
Modern tools can answer plain-language questions — what did we spend with this supplier last quarter, which clients pay late — directly from the ledger. Treated as a way to explore your own verified data, this is genuinely useful and low-risk.
LIMITS
Where AI falls short
The gaps are not edge trivia — they are the parts of accounting where the consequences concentrate.
Applying tax law to your facts
Tax outcomes turn on specifics: your structure, your intentions, the character of a payment, timing. AI generates plausible general answers; it does not ascertain your state of affairs — which is exactly the analysis the TPB says a practitioner must bring themselves.
Structuring and advice
Whether to restructure, how to treat a loan between you and your company, when to bring forward or defer — these are judgement calls with long tails, made against your whole position. The TPB is explicit that AI output is not a substitute for a practitioner's own analysis of a client's circumstances.
Hallucinated confidence
Language models produce fluent, authoritative-sounding answers that are sometimes wrong — including citing rules that do not exist. The TPB's guidance names this directly: AI models may hallucinate or provide inaccurate information and cannot replace tax knowledge, experience or expertise.
Related-party and unusual transactions
Transactions between connected entities, one-off events, disposals, settlements — the ledger shows a movement, but the correct treatment depends on documents, intent and structure that AI does not hold. These are reliably the transactions where automated treatment goes wrong.
Knowing what it doesn't know
A good accountant escalates: 'this one I need to check'. AI does not reliably distinguish the routine from the exceptional — it processes both with equal confidence. The review layer exists precisely because the system cannot be trusted to flag its own blind spots.
Accountability to the ATO
When the ATO reviews a position, someone must explain it, produce the records — kept for five years — and stand behind the judgement. No software vendor does that for you. Accountability is the one deliverable AI structurally cannot provide.
COMPLIANCE
The Australian rules
Australia regulates who may do paid tax work, and the rule is blunt: under the Tax Agent Services Act 2009, providing tax agent services for a fee or other reward requires registration with the Tax Practitioners Board, and the same applies to BAS services. Registration comes with qualification requirements, professional conduct obligations and consequences for getting it wrong. No software product is, or can be, a registered agent. For a business owner this is protection, not red tape — it guarantees that whoever takes tax positions on your behalf is qualified, supervised and answerable.
In July 2026 the TPB issued TPB(GS) 55/2026, its guidance statement on the use of artificial intelligence and the Code of Professional Conduct. It sets the terms on which registered practitioners may use AI: they remain ultimately responsible for the services they provide; AI outputs must be assessed and supplemented by professional judgement before being relied on; AI-generated content should be verified and reviewed for accuracy throughout the workflow, with the process documented; and client information must not be entered into AI tools that disclose it to third parties without the client's permission. If your accountant uses AI — and increasingly they will — this is the standard their use is held to.
Underneath all of it sits the ATO's record-keeping baseline: most business records must be kept for five years, generally from when the record was prepared or obtained, or the transaction completed, whichever is later. AI-assembled accounts do not change the obligation — the records behind every figure must be complete, retrievable and explainable.
The compliance calendar has also tightened in ways that favour AI-assisted accounting done well. From 1 July 2026, payday super requires superannuation guarantee — now 12% — to reach employees' funds within 7 business days of each payday. Accounting that reconciles continuously, rather than quarterly, is becoming the operating requirement rather than the premium option.
PRICING
What it costs
AI accounting is not a single line item, so compare structures rather than products. At the base is software: accounting platform subscriptions with AI features generally bundled in, plus optional specialist tools for payables, reporting or forecasting, each adding its own monthly fee. This layer is cheap relative to what it automates — but it produces drafts, not accountability.
Above that sits professional work, priced either hourly or fixed-fee. Traditional hourly engagement means AI efficiency shows up as fewer billed hours for the same compliance output — worth asking your accountant about directly. Fixed-fee services price the outcome instead: books kept, BAS prepared and lodged, reports delivered, questions answered, with AI efficiency absorbed into the fee rather than itemised.
For the bookkeeping layer specifically, our published pricing-page comparison — indicative, not a quote — puts a separate bookkeeping provider at $500–800 a month and a DIY part-time hire at $25–40k a year. The wider pattern across the accounting function is consistent: AI keeps compressing the cost of processing, so the share of your spend that buys judgement, review and advice keeps rising. That is the correct direction — judgement is the part worth paying for.
HOW VALONT RUNS IT
Inside a connected back office
How the AI-and-human split works when accounting is delivered as part of one connected system rather than a stack of separate providers.
AI runs the processing layer
Transactions code themselves, payables are extracted and matched, reconciliations stay current and draft reports assemble from a live ledger — continuously, not at month-end. The machinery of accounting runs without consuming anyone's week.
Qualified people own the judgement layer
Tax positions, unusual transactions, structuring questions and anything the AI flags as an exception go to qualified professionals — with the work the law reserves for registered agents routed to a registered agent, never left to the software. The handover point is explicit, not accidental.
One ledger feeds everything
Bookkeeping, BAS, payroll and management reporting draw on the same live data, so your accountant is never reconciling three providers' versions of your business before they can start. The picture is shared, current and singular.
One team is accountable
Questions — yours or the ATO's — land with one team that holds the records, the reasoning and the registrations. Review standards like those in TPB(GS) 55/2026 are applied once, consistently, rather than assumed to be someone else's job.
The coordination overhead disappears
The hidden cost of fragmented accounting is the owner relaying context between a bookkeeper, an accountant and a stack of software that do not talk to each other. Connect the function into one system and that job — the coordination tax — disappears from your calendar.
Find the right split between AI and judgement
If you are trying to work out what AI should take over in your accounting and what should stay with qualified people, a 30-minute review will map it against your actual setup — where the drafts can run themselves, and where the accountability needs to sit. No pressure attached.