AI bookkeeping for UK SMEs: what it does today, and what still needs a person
Manoj Tutika · Published 16 November 2026 · Last checked 16 November 2026
Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.
What does AI bookkeeping actually do today?
AI bookkeeping has moved past reading receipts. In 2026 an agent can take a supplier bill from an inbox, read it line by line, code it to the account and VAT rate the supplier was coded to last time, attach the document, create the bill in your accounting software and match the payment when it leaves the bank. It can do the same for sales receipts, card payouts and recurring journals.
For a UK SME with a few hundred to a few thousand transactions a month, that is most of the bookkeeping. What remains for people is the part that needs judgement or authority: deciding how a new type of cost should be treated, approving a payment, explaining a variance to the board and signing off the VAT return.
Which tasks can you hand over, and which stay with people?
| Task | AI agent today | Person |
|---|---|---|
| Collect bills and receipts from email and apps | Yes | No |
| Read and code bills, including VAT rate | Yes, learns from past coding | Sets rules for new suppliers |
| Code bank transactions | Yes | Reviews exceptions |
| Match payments to invoices and bills | Yes, including part payments | Reviews uncertain matches |
| Recurring journals, prepayments and accruals | Proposes and books within rules | Approves the policy |
| Credit control reminders | Drafts and sends within your rules | Handles disputes |
| VAT return | Prepares and ties out | Reviews and submits |
| Month-end close checklist | Runs checks with evidence | Signs off |
| Annual accounts and corporation tax | Prepares the trial balance and schedules | Accountant prepares and files |
How is an AI agent different from bank rules and OCR?
Bank rules in Xero or QuickBooks are if-then statements. They work for the same direct debit every month and break when a supplier changes its reference. OCR reads the text on a document and leaves the decision to you. An agent combines the two with context. It reads the bill, checks how this supplier was booked before, notices that the amount is three times the usual, and asks before it books. The difference shows on the hard cases, not the easy ones.
An agent also learns from corrections. If your controller recodes a software subscription from general expenses to IT costs once, the next bill from that supplier is coded the new way, and the agent can say why. Bank rules only change when someone edits them, and OCR does not learn at all. Over a few months that difference adds up: fewer exceptions every month, and a record of the reasoning behind each entry that an auditor or accountant can follow.
What does it look like in numbers?
Take a UK wholesaler with £6m turnover, 1,100 bank lines and 450 supplier bills a month on Xero. Before automation, a bookkeeper spends most of the month on entry and matching, and the close takes eight working days. With an agent doing the coding, matching and payout splits, the bookkeeper reviews perhaps 30 to 60 exceptions a month and the close moves to the first days of the next month. The skill shifts from typing to reviewing.
How do you judge an AI bookkeeping tool?
- Ask for a measured number. The share of work completed without a person changing it, with the period, the sample size and what counts as a unit.
- Test it on your own books. A month of your real transactions, not a demo company.
- Check what it writes. Does it create bills and matches in your software, or only suggest? Can you approve each write, or set rules for what it may do alone?
- Look at the audit trail. Every entry should link to its source document and show who or what made it.
- Check data residency. Where are your data and the AI models hosted, and is there a data processing agreement?
- Look at the exceptions. Ask how the tool handles a payout, a part payment and an unknown receipt. That is where the hours are.
Thred's own measurement method is public: read how we measure bookkeeping without a human.
What do UK rules mean for AI bookkeeping?
Making Tax Digital for VAT already requires VAT-registered businesses to keep digital records and file through compatible software, and MTD for Income Tax started in April 2026 for sole traders and landlords above the income threshold. From 1 April 2029 VAT-registered businesses must also be able to issue and receive e-invoices, according to the HMRC roadmap. Each of these makes digital, structured bookkeeping the default, which is exactly what agents work on. None of them changes who is responsible: the business, and its accountant, still own the return.
What do Thred's agents do for UK SMEs?
Thred is a team of AI agents that takes the recurring finance work off your team, on the accounting software you already use. The Connector collects bills and receipts from email, Outlook, Google Drive and WhatsApp. The Bookkeeper keeps the record up to date: it reads and codes each document, codes bank lines, matches payments and splits Stripe and PayPal payouts. The Payer manages your payables and checks every bill for duplicates and changed bank details. The Collector manages your receivables.
The Accountant handles filings and prepares the close. The Analyst monitors performance and variance, and The CFO is your finance lead when you have none. On Xero the agents read your organisation and create and update records, with each write shown for approval until you allow it. See Thred Layer.
Frequently asked questions
Can AI do my bookkeeping?
Much of it, yes. Current AI agents read documents, code bank transactions, match payments to invoices and prepare VAT and month-end work in your accounting software. You still need someone to set the rules, approve unusual items and take responsibility for filings.
Is AI bookkeeping software accurate?
It depends on the tool and how it is measured. Ask for a measured share of work completed without a human change, the period, the sample size and how corrections were counted. Thred publishes its method: 99.5% of 36,963 work units in production were not changed by a person, measured in September 2026.
Does AI bookkeeping work with Xero?
Yes. Several tools, including Thred, work with Xero through its API: they read your organisation and create or update bills, invoices and bank matches. Check whether the tool writes to Xero or only suggests, and how approvals work.
Is AI bookkeeping compliant with Making Tax Digital?
MTD for VAT requires digital records and filing through compatible software. AI that books into Xero or another MTD-compatible package keeps your records digital; the VAT return is still filed through that software, and you remain responsible for it.
Will AI replace my bookkeeper or accountant?
It replaces much of the data entry and checking, not the judgement. A bookkeeper or accountant who used to key in bills now reviews exceptions, advises and signs off. Many firms use AI to take on more clients with the same team.
What does AI bookkeeping cost?
Prices vary widely: some tools charge per user, others per transaction, per entity or by usage. Compare the cost with the hours of entry, matching and chasing it removes, and include the time you will spend reviewing.
Sources
- GOV.UK, Making Tax Digital for VAT
- GOV.UK, Making Tax Digital for Income Tax: who needs to use it
- GOV.UK, Roadmap for e-invoicing implementation (published 28 October 2026)
- Thred Layer, AI agents on your existing accounting software
Read next
- What your next finance hire should actually do
- The bookkeeping work OCR never did
- Bookkeeper, accountant, agent: who does what in a small company's finance in 2026
- How we measure bookkeeping without a human: the 99.5% method
- Bank reconciliation: how to do it step by step, and how to automate it
- What does Xero JAX do? Features, pricing and limits, checked 16 November 2026
Where Thred comes in
Finance agents running on top of your existing ledger. You set the rules, approve the work and handle the exceptions. No migration.
