Key takeaways
What this article covers, in order:
- What it feels like in practice
- How it actually works (without the hype)
- Where it does well
- Where you still need to look
- A worked example: one month, checked
- A ten-minute weekly routine
AI transaction categorisation is very good at the repetitive stuff: the same suppliers, the same direct debits, the same kinds of purchase every month. Where it still needs you is anything that depends on context it can't see, such as mixed personal and business spending, transfers, one-off big purchases and vague descriptions. Treat it as a fast first draft you check, not a bookkeeper you never look at.
What it feels like in practice
Daniel is a self-employed photographer in Cardiff. Before, his Sunday-evening bookkeeping meant scrolling through 60-odd transactions and picking a category for each one. Fuel. Software. Fuel. Lens hire. Fuel.
With AI categorisation switched on, he opens the review list and most lines already have a sensible category against them. He scans down, agrees with the obvious ones, and stops on the handful that look off. The job hasn't vanished. It's just gone from "decide everything" to "check and correct".
That shift is the whole point, and also the whole caveat.
How it actually works (without the hype)
AI categorisation, in general, looks at what it can see about each transaction: the description the bank sends, the amount, whether it's money in or out, and patterns from how transactions like it have been categorised before. From that, it suggests the most likely category.
What it doesn't have is everything in your head. It doesn't know that the £300 at a camera shop was a gift for your brother. It doesn't know that a customer paid you via a friend's account. It can only work with the clues on the statement line.
Where it does well
| Transaction type | Why AI handles it well |
|---|---|
| Monthly subscriptions | Same description, similar amount, every month |
| Fuel and parking | Recognisable merchant names, consistent category |
| Phone and broadband direct debits | Regular, clearly labelled |
| Bank charges and interest | The bank's own wording is consistent |
| Regular suppliers | Once you've confirmed a supplier a few times, the pattern is strong |
| Customer receipts from known clients | Recurring names on incoming payments |
For most small businesses, those make up the bulk of the list. That's where the time saving comes from.
Where you still need to look
Mixed-use merchants
Supermarkets, Amazon, big DIY chains. The same shop could be office supplies one week and the weekly food shop the next. The AI can make a good guess based on your history, but only you know what was in the basket.
Transfers between your own accounts
"TFR 40-12-34 12345678" might be a payment to a supplier or a move to your own savings. If it's a transfer, it shouldn't hit profit at all. Always check anything that looks like a transfer.
Owner spending
A sole trader paying a personal bill from the business account, or a director using the company card for something personal, needs to go to drawings or the director's loan account. The description won't tell the AI that.
One-off big purchases
A £1,850 laptop could be an expense or a piece of equipment, depending on how you and your accountant treat it. Large, unusual amounts deserve a second look every time.
Refunds and reversals
Money coming in from a supplier looks a lot like income. It isn't. It's a refund that should reduce the original cost. Watch for incoming payments from names you normally pay.
Vague descriptions
"SQ JOHN" or "PAYPAL PAYMENT" could be anything. The AI will guess; you'll know.
A worked example: one month, checked
Here's how Daniel's Oct 2026 might look once the AI has made its suggestions and he's reviewed them. The numbers are illustrative, but the pattern is typical.
| Group | Transactions | Total value |
|---|---|---|
| Suggestions accepted as they were | 52 | £3,214.60 |
| Suggestions changed by Daniel | 6 | £3,410.30 |
| Total reviewed | 58 | £6,624.90 |
The six he changed, line by line:
| Description | AI suggestion | Daniel's change | Amount |
|---|---|---|---|
| CAMERA WORLD | Equipment | Drawings (gift) | £300.00 |
| AMAZON MKTPLACE | Office costs | Drawings (personal) | £46.80 |
| TFR TO 20-11-55 | Subcontractors | Transfers (own savings) | £1,000.00 |
| LENS PRO REFUND | Sales | Equipment hire (refund reduces cost) | £85.00 |
| LAPTOPS DIRECT | Office costs | Equipment | £1,849.00 |
| ESSO | Motor and travel | Drawings (family car) | £129.50 |
Add them up: £300.00 + £46.80 + £1,000.00 + £85.00 + £1,849.00 + £129.50 = £3,410.30, which is the figure in the summary.
Six changes out of 58 doesn't sound like many. But look at the value: £3,410.30 of £6,624.90, over half the money, sat in the lines that needed a human. Left unchecked, his business costs would have been overstated by £300.00 + £46.80 + £1,000.00 + £129.50 = £1,476.30 of personal spending and transfers, the refund would have shown as £85.00 of extra sales, and the laptop would have gone to the wrong place.
The lesson: AI does the volume. You do the value.
A ten-minute weekly routine
- Open the review list and sort by amount, largest first.
- Check every line over a threshold you pick (say £250) properly.
- Look at every incoming payment that isn't from a known customer.
- Look at anything with "TFR", "transfer" or a sort code in the description.
- Scan the rest quickly. If a suggestion looks right, confirm it.
- When you correct something, be consistent. Same supplier, same category every time, so your history tells one clear story.
What AI categorisation doesn't replace
It doesn't replace reconciliation. Categorisation decides what a transaction was. Reconciliation proves your books match the bank, with nothing missing or doubled. You need both.
It also doesn't replace judgement on treatment. Whether something is an asset or an expense, or how a vehicle with mixed use should be handled, is a decision for you and your accountant. Once you've made it, the AI can follow it.
How HelloBooks helps
AI auto-categorisation is part of HelloBooks Pro (£14.99 a month, or £149 a year). Transactions from your bank feed (connect your bank through Open Banking, most UK banks and cards) or from a CSV statement import land in the review list with a suggested category, and you confirm or change each one.
The Free plan comes with free AI credits to get started. If your AI credits run out, AI categorisation pauses and the books keep working; you simply categorise by hand until credits are available again. Read more on the AI bookkeeping software page, or see what's on each plan.
FAQs
Is AI categorisation accurate enough to trust without checking?
For routine, repeating transactions it's usually right. For anything large, unusual, personal or a possible transfer, check it. A quick weekly review is enough for most small businesses.
Does it learn from my corrections?
AI categorisation in general works from patterns, and how similar transactions were categorised before is one of the strongest. Whatever tool you use, consistent corrections give it cleaner patterns to work from.
What happens if I run out of AI credits?
AI categorisation pauses. Everything else carries on, and you can categorise transactions manually.
Can AI categorisation tell personal from business spending?
Not reliably. It only sees the bank description and amount. Personal spending on a business account needs a human to flag it.
Do I still need to reconcile if I use AI categorisation?
Yes. Categorisation and reconciliation do different jobs. One labels transactions; the other proves the books match the bank.
Let it take the dull, repetitive bulk off your plate, and keep your eyes on the lines where the money actually is.
Start free, no card needed. Try HelloBooks Free
