What AI for accountants means in practice is narrower than the marketing suggests, and that is good news. The tools that earn their place at a small accounting firm are not trying to replace the accountant. They are taking the high-volume, low-judgement work off the desk so the qualified person spends their hours on the parts that need a brain.
This note is the free read on where that line sits. It draws on the same testing behind our accounting reports, condensed to the version you can act on this afternoon.
The work AI is genuinely good at
Transaction categorisation is the strongest case. Once a tool has watched a few months of a client's bank feed, it sorts the recurring items reliably and the accountant reviews the exceptions instead of every line. The card-processing fee lands in the right place without a human touching it.
Reconciliation matching is the other one that holds up. Matching a bank line to an invoice is pattern work, and the tools are good at pattern work. Used well, the accountant looks only at the unmatched residue rather than the whole ledger.
Document handling has quietly become useful too. Pulling figures off a supplier invoice, reading a receipt photo a client took in a car park, sorting a year of paperwork into the right periods. None of it is glamorous. All of it used to eat junior hours.
The work it still gets wrong
Anything that depends on knowing the client's intent is where the tool guesses, and a guess here is not a rounding error. Was that payment a capital purchase or a repair? The software reads the description, and the description is whatever someone typed in a hurry. The treatment it picks changes the tax position.
VAT on mixed or unusual supplies is the second soft spot. The edge cases are exactly the ones a general model handles by reaching for the most common answer, which is wrong in the cases that matter.
The dangerous part is that the tool files these with the same calm confidence as the easy items. It does not flag them as uncertain. So the error is invisible unless a person goes looking, which is the whole argument for keeping review in the loop.
What this means for how a firm staffs
The firms getting value are not chasing a smaller payroll. They moved their best bookkeeper off data entry and onto review and client work, kept the headcount, and put more billable judgement through the same door.
A practical test before you buy: if your firm spends more than a few hours a week on manual categorisation and chasing paperwork, a tool that integrates with the ledger you already run will pay for itself quickly. If you are below that, the overhead of running the software can cost more than it saves.
The report behind this note tests the specific tools, names which ones integrate cleanly with the common ledgers, and shows the pricing each vendor actually quotes a small firm.