Every accounting software vendor seems to be slapping “AI-powered” on their marketing pages this year. Xero’s doing it. MYOB’s doing it. Even that invoicing app you forgot you signed up for is doing it. But when you sit down at your desk on a Monday morning with a stack of bank reconciliations, how much of this AI stuff actually matters?

The short answer: some of it is genuinely helpful. Most of it isn’t worth switching providers for.

Where AI Is Already Doing Real Work

Let’s start with the wins. Transaction categorisation is the most obvious one. If you’ve used Xero or QuickBooks Online in the past year, you’ve probably noticed that bank feeds are getting smarter about guessing which account a transaction belongs to. It’s not magic — it’s pattern matching on historical data — but it saves time. For a bookkeeper processing hundreds of transactions a week, that adds up fast.

Receipt scanning and data extraction is another area where AI has made a genuine difference. Tools like Dext (formerly Receipt Bank) and Hubdoc can pull supplier names, dates, totals, and GST amounts from photos of receipts with decent accuracy. Not perfect. You still need to check. But it’s miles better than manual entry.

Anomaly detection is quietly becoming useful too. Some platforms now flag unusual transactions — a supplier invoice that’s 40% higher than normal, or a duplicate payment. For small firms without dedicated internal controls, this is a real benefit.

Where the Hype Outpaces Reality

Now for the less impressive stuff. “AI-generated financial insights” is a phrase that sounds incredible in a press release and means almost nothing in practice. Most of what gets labelled as insight is just a chart with a sentence underneath it. “Your expenses increased 12% this quarter.” Thanks, I can read a graph.

Automated tax advice is another area where vendors are overreaching. AI can help organise data for tax time, but the idea that it can replace a qualified tax agent for anything beyond the simplest individual return is premature. Australian tax law is complex. The ATO’s own guidance changes constantly. You still need a human who understands the nuances of Division 7A or the small business CGT concessions.

Cash flow forecasting tools have improved, but they’re still largely extrapolating from past trends. They can’t account for the client who’s about to churn, the supplier who’s about to raise prices, or the fact that your biggest customer always pays 15 days late in December.

What Small Practices Should Actually Do

If you’re running a small accounting practice or handling the books for an SMB, here’s what I’d recommend:

Use the AI features built into your existing software. Don’t go hunting for separate AI tools. The bank feed categorisation in Xero and QBO is good enough for most use cases. Turn on the smart suggestions and let them learn from your corrections.

Invest in good receipt capture. This is where the time savings are most tangible. Get your clients using Dext or a similar tool. The less manual data entry you’re doing, the more time you have for advisory work.

Be sceptical of “AI advisory” products. If a tool claims it can generate management reports with AI commentary, test it thoroughly before paying for it. Most of the time, the output reads like a first-year university student summarising a spreadsheet.

Don’t ignore the basics. The Australian Taxation Office keeps updating its digital record-keeping requirements. Making sure your systems meet those standards matters more than having the flashiest AI feature.

The Bigger Picture

AI in accounting isn’t a revolution. It’s incremental improvement. The firms getting the most value from it aren’t chasing the latest shiny tool — they’re applying simple automation to repetitive tasks and reinvesting the saved time into client relationships.

The profession’s future isn’t about whether AI will replace accountants. It won’t, at least not the good ones. It’s about whether accountants will get comfortable enough with technology to stop treating every new feature as either a threat or a gimmick.

The CPA Australia Digital Transformation report from earlier this year found that practices using automation for data entry saved an average of 8 hours per week. That’s a full working day. Not from some exotic AI system, but from basic automation that’s been available for years.

Start with what works. Ignore the noise. And keep checking the categorisation — the AI still puts Uber Eats under “Travel” half the time.