Has AI actually automated anything real in accounting?

Yes, but a narrower set of things than the marketing suggests. The genuine, working automation today is data capture and OCR from invoices and receipts, transaction categorisation, reconciliation matching, and first-draft reports and summaries. These are real, in production, and save real hours. The claims that are still hype are fully autonomous audits, zero-review books, and software that replaces your whole team. So it is neither snake oil nor magic. It is genuine automation of the repetitive middle of the job, with a human still owning the judgement.

Is AI in accounting just macros and Power BI with better marketing?

The critique has a fair point. A lot of what gets sold as AI is automation that rules, macros, or a well-built Power BI model could already do. What is genuinely new is handling unstructured and messy input: reading a crumpled photographed receipt, matching transactions that do not tie out exactly, or drafting a plain-English summary of a variance. Macros need clean, predictable structure. The new tools tolerate mess and ambiguity, which is where most real bookkeeping actually lives.

How do I tell real AI automation from vendor hype?

Ask four questions. First, what specific task does it remove from my week, named precisely. Second, what is the accuracy rate and does it still need review. Third, can I see it run on my own messy data, not a clean demo file. Fourth, what happens when it is wrong and who is accountable. Real tools answer all four plainly. Hype dodges into vague language about transformation, intelligence, and the future. If a vendor cannot name the task and the review step, assume there is not much underneath.

How accurate is AI transaction categorisation?

Transaction categorisation accuracy typically exceeds 90 percent after the tool has learned your coding patterns, but that remaining margin is exactly why a human still reviews. Ninety-plus percent is genuinely useful because it removes most of the manual sorting, yet it is not a number you sign off on unreviewed. The reliable pattern is AI drafts the coding, a human approves and corrects the exceptions. That review step is now part of the role, not an optional extra.

If most of it is hype, is AI in accounting worth adopting?

Yes, if you adopt it for the narrow things it genuinely does. Intuit reported roughly 98 percent of accountants and bookkeepers used AI in some form in the past year, which tells you the working tools are already embedded in everyday software. The mistake is buying the transformation story. The win is deploying automation on the repetitive 40 to 60 percent of manual work and keeping humans on the judgement. Adopt the boring, proven parts and stay sceptical of the rest.

Has AI Actually Automated Anything in Accounting, or Is It Just Hype?

Q: Will AI fully automate audits or bookkeeping without review?

Not today, and not credibly on the near horizon. Fully autonomous audits, zero-review books, and replace-your-team platforms are the overclaims. The Institute of Certified Bookkeepers has noted that platforms built to fully replace bookkeepers have launched and collapsed within weeks. Accountability, professional sign-off, and the messy judgement calls do not automate away. What automates is the repetitive drafting underneath them.

A bright modern Australian accounting office with an accountant reviewing reconciled ledgers on screen while a soft teal AI hologram sorts transactions in the background, no text in the image — Some of the automation is real and boring and useful. Some of it is a slide deck. This is how to tell them apart.

Short Answer

Yes, AI has automated real things in accounting, but fewer than the marketing claims. The genuine wins are data capture and OCR, transaction categorisation, reconciliation matching, and first-draft reports. The overclaims are fully autonomous audits, zero-review books, and replace-your-team software. It is neither snake oil nor magic. The trick is telling the two apart, and this article gives you a test for exactly that, plus an honest answer to the "it's just macros" critique.

Spend an hour in an accounting forum and you will find the sceptics out in force, and rightly so. "Has AI actually automated anything in your workflow, or has it just been snake oil fluff so far?" "Hot take: the AI fearmongering in accounting is way overhyped." "Everything they demo could already be done with Power BI or a macro." These are good, sharp questions, and the people asking them have watched a decade of software promising the moon and delivering a dashboard.

So let us do the honest audit that the hype cycle never bothers with. What has AI genuinely automated in accounting today, what is still a slide deck, and how do you cut through a vendor pitch in under two minutes? We will treat the sceptic's case seriously, including the macros critique, because it deserves it. For the wider question of whether the profession itself is at risk, we covered whether AI will replace bookkeepers and accountants separately.

What AI Genuinely Automates Well Today

Start with the good news for the optimists and the bad news for the pure sceptics: some of this is real, in production, and saving real hours right now. These are not demos. They are the boring, repetitive tasks that AI has quietly become reliable at.

Data Capture and OCR

Reading a photographed receipt or a PDF invoice and turning it into a coded transaction. The "pile of paper to type up" problem is largely solved, including messy, low-quality scans that older tools choked on.

Transaction Categorisation

Learning your coding patterns and sorting bank transactions automatically, getting the large majority right without being asked twice, and improving as you correct it.

Reconciliation Matching

Matching transactions across feeds and ledgers, including the near-misses that do not tie out to the cent, then surfacing only the genuine exceptions for a human.

First-Draft Reports and Summaries

Producing a month-end pack or a plain-English variance summary from data that is already clean, giving you a draft to edit rather than a blank page.

The Number That Matters

Intuit reported that roughly 98 percent of accountants and bookkeepers used AI in some form over the past year. That figure is not a prediction about the future. It is a measure of how far the working tools have already spread into everyday software. The question is no longer whether AI is used, but which uses are real and which are decoration.

What Is Still Marketing Vapour

Now the part the sceptics are right about. For every genuine automation above, there is a bigger, shinier claim that does not survive contact with a real client's books. If a pitch leans on any of these, hold on to your wallet.

"Fully autonomous audits"

AI can flag anomalies and sample transactions faster than any human. It cannot own the professional judgement, the risk assessment, or the sign-off that an audit is. Anomaly detection is real. An unsupervised audit is not.

"Zero-review, hands-off books"

The tools are confident even when they are wrong. The mis-coded intercompany transfer, the duplicated bill, the vendor detail quietly changed to the wrong value, these are precisely what a "no human needed" system misses. Review is not the old world's leftover. It is the new world's core skill.

"Replace your whole team"

This is the claim that keeps collapsing. The Institute of Certified Bookkeepers has noted that platforms built to fully replace bookkeepers have launched and folded within weeks. The durable tools assist a professional. The ones that promise to remove the professional entirely tend not to be around long enough to disappoint you twice.

"But Couldn't Power BI or a Macro Already Do This?"

This is the sharpest critique in the room, and it is partly correct. A great deal of what gets rebadged as AI is automation that a well-built macro, a rules engine, or a Power BI model could already handle. If your data is clean and structured, and the logic is fixed, you often do not need AI at all, and paying an AI premium for it is a waste. Give the sceptics that point fully.

What is genuinely new is the tolerance for mess. A macro needs predictable structure: the same columns, the same format, the same wording, every time. The moment reality intrudes, a photographed receipt at an angle, an invoice from a new supplier with a layout nobody mapped, a bank narration that varies by transaction, the macro breaks and a human takes over. The new tools handle that unstructured, ambiguous input, which is where most actual bookkeeping lives. That is the real dividing line, and it is worth understanding what an AI agent actually is before you judge whether you need one.

Worth Remembering

The right question is not "is this AI or a macro?" It is "does this reliably remove a task I currently do by hand?" If a macro does it, use the macro and keep your money. If the task involves messy, varied, unstructured input that a macro cannot cope with, that is exactly where the newer tools earn their place. Pick the tool by the job, not by the label on the box.

A Two-Minute Test for Cutting Through the Hype

Next time a vendor, a colleague, or a LinkedIn post tells you AI has changed everything, run it through these four questions. Real tools answer all four plainly. Hype retreats into vague talk of transformation and intelligence.

1. Name the exact task it removes

Not "streamlines your close" but "reads supplier invoices and codes them". If they cannot name a specific task you do today, there is usually nothing underneath.

2. Ask for the accuracy rate and the review step

A credible tool tells you its accuracy and admits a human still reviews the exceptions. A tool that claims perfection or "no review needed" is selling the overclaim.

3. Run it on your own messy data

Demos use clean files. Insist on watching it work on your real, ugly month, the odd suppliers, the split payments, the corrections. That is where hype quietly falls apart.

4. Ask what happens when it is wrong

Who catches the error, who is accountable, and how is it corrected? A real tool has an answer. Vapour treats being wrong as a possibility it never planned for.

This is roughly the pattern the honest tools already follow. When we ran dedicated AI over a real client's books, it did not replace anyone and it did not claim to. It caught the things a rushed human had missed, named them, and handed them back for a decision. That is what working automation looks like: specific, reviewable, and honest about its limits.

The Bottom Line

So has AI actually automated anything in accounting? Yes, and the sceptics are still right to be sceptical. The real wins are narrow and boring: data capture, categorisation, reconciliation matching, and first-draft reports. These genuinely automate the repetitive 40 to 60 percent of manual work. The rest, the autonomous audits, the zero-review books, the replace-your-team pitches, is marketing that has not met a real ledger. Both things are true at once, and pretending otherwise is how firms end up disappointed or left behind.

The smart move is neither to swallow the hype nor to dismiss the lot. It is to adopt the proven, boring automation, keep humans firmly on the judgement, and apply the four-question test to everything else. If you want help sorting the real from the vapour for your own workflow, that is exactly what we do. Book a free consultation and we will show you what dedicated AI genuinely automates, and what we would tell you to skip.

Skip the Hype, Automate the Boring Parts

Agentive helps accounting and finance teams deploy dedicated AI on the repetitive work that genuinely automates well, with humans firmly on the judgement.

Book a Free Consultation See Our AI Solutions

Has AI Actually Automated Anything, or Is It Just Hype?