Explainer
AI bookkeeping: what it does well, and where a person still decides
AI can now do much of the daily bookkeeping: reading bills, matching payments, recording receipts. What it handles well, where it should stop and ask, and how to judge a tool before you trust it with your books.
Key takeaways
- AI bookkeeping reads documents the way a person would, so it copes with bills that don't look the same twice.
- It's strongest on daily, high-volume work: bills, receipts, matching payments and keeping invoice statuses current.
- It should stop and ask on anything new, large, unclear or that moves money.
- Judge a tool on where it works, what it logs, what it can undo and who supervises it.
What's changed
Bookkeeping software has been automating for years: bank feeds, rules that match a reference, templates that read one supplier's invoice layout. It worked until something looked different, and then a person took over.
What's new is that software can now read a document the way a person does. A bill from a new layout, a receipt photographed at an angle, a payment reference with a typo: modern AI, built on LLMs, can read these, work out what they are and decide the next step against your rules. That moves a lot of daily bookkeeping from "someone has to look at this" to "done, and here's the record".
What it does well
The best bookkeeping work to hand to AI is recurring, rule-based, high in volume and spread across more than one system.
- Recording supplier bills the day they arrive, against the right supplier, account and tax rate
- Reading receipts from email and chat, and matching them to card payments
- Reconciling the bank daily, including part and combined payments
- Checking payments against invoices, and keeping invoice statuses current
- Spotting duplicates, price changes on renewals and charges nobody recognises
- Preparing payments with the invoice, approval and bank details checked
Where a person still decides
Good AI bookkeeping knows where it stops. It doesn't guess on the cases that carry risk; it brings a person in with the evidence attached, and waits.
- A supplier or customer it hasn't seen before
- An amount above the threshold you set
- Bank details that have changed
- A document that doesn't match the order or the contract
- Treatment questions, such as how to account for an unusual item
- Sending money, which should always stay with your own people
What it doesn't replace
AI bookkeeping keeps the books current. It doesn't replace your accountant's judgement on year-end accounts, tax planning or advice, and it shouldn't try to. In practice it makes that work faster and cheaper, because the accountant starts from clean, current records instead of a backlog.
What it looks like in practice
We put the OFFHOURS Accountant on our own books first. Every day it records new supplier bills in QuickBooks, reconciles the bank statement to the cent and updates the status of open invoices. On a typical day it registered 44 supplier bills, reconciled the bank in full and updated 10 invoice statuses, finishing by 10:30am.
Anything outside its rules, such as an unknown supplier or an amount over the threshold, comes back to a person with the evidence attached. Our finance team reviews the work; it doesn't redo it.
How to judge a tool
Most AI bookkeeping tools demo well. The questions that separate them are about control, not features.
- Does it work inside our accounting software and bank, or do we move our data to it?
- Where exactly does it stop and ask, and can we set those points ourselves?
- Can it send money on its own? It shouldn't.
- Is every action logged with the rule behind it, and can it be undone?
- Who reviews its work, and how often?
- Is it GDPR compliant, and who can see our data?
How to start
Start with one task where volume is high and the rules are clear. For most companies that's supplier bills or daily bank matching. Run it on probation, with every output reviewed, until you trust it, then widen the scope.
That's how the Accountant starts: working within a week, on probation at first, then supervised around the clock by our engineers, inside the systems you already use.
