Accounts payable in a small firm is five processes, loosely tied together by an inbox and somebody’s memory: the invoice arrives, someone works out what it is, someone decides it is legitimate, someone pays it, and someone ticks it off against the bank. Automating “AP” without saying which of the five you mean is how firms end up with a capture tool bolted to a process that still runs on chasing a partner for a nod in the corridor.
So take them separately.
| Stage | What it is | Automatable today? |
|---|---|---|
| Capture | Invoice arrives, becomes a structured record | Yes, well. This is the mature part |
| Coding | Nominal code, VAT treatment, department or job | Partly. Reliable on repeat suppliers, poor where judgement is involved |
| Approval | Someone with authority agrees to pay it | No, and it should not be. Route it, time it, log it, but a person decides |
| Payment | Money leaves the account | No. Automate the preparation of the run, never the release |
| Reconciliation | Ledger matched to bank | Yes. It is matching, which is what software is for |
Read the middle column first. Two of the five stages are data handling and two are decisions. Automation projects fail when they treat an approval as a data-handling step with a delay in front of it.
The biggest risk: supplier bank-detail fraud
The faster and smoother your payment process, the better it works for a fraudster, and most AP automation material leaves that out.
An invoice arrives that looks entirely ordinary, from a supplier you do use, for an amount in the range you normally pay, with a line saying the bank details have changed. In a slow manual process somebody might notice, because the invoice passes through enough hands that one pair of eyes is fresh. In a tidy automated one it is captured, coded against the supplier record, matched to an approval rule and queued into the next payment run without a person ever looking at the account number.
A better model won’t fix this. The control is separation of powers, written down and then enforced in the permissions: the person who can change a supplier’s bank details cannot be a person who approves invoices or releases the payment run. A change to bank details is treated as a new supplier, not an edit, and it is confirmed by telephone on a number you already held before the request arrived. Never the number on the letter, never a reply to the email, never a mobile given over chat.
Two smaller habits travel with it. Make the bank-detail field produce an audit entry naming who changed it and when, because a change nobody can account for is the thing you want to find in week one rather than at year end. Then decide in advance what happens when the supplier really has changed bank, because the real ones are rushed, apologetic and persuasive, and a control only holds if the compliant path is obvious.
Confirmation of Payee at the bank gives you a name check against the account you are about to pay, which is worth using and is a check rather than a guarantee. Treat it as one more signal, not as the reason you stopped phoning.
AP automation on Xero, QuickBooks or Sage
The shape is the same on all three: the ledger holds supplier records, bills and the approval state, and a capture layer feeds it. The differences are in what you already pay for.
Xero
Bills are a first-class object with an approval state, and Xero owns Hubdoc, its capture product. Which plans include which capture and payment features moves around, so read your own plan page before buying anything extra. On the UK pricing page today the entry-level Ignite plan is capped at 20 invoices and 10 bills a month and charges £0.20 per bill or payroll payment, while the plans above it bundle a small number of payments; list prices run £18, £39, £55 and £70 a month excluding VAT for Ignite, Grow, Comprehensive and Ultimate, with a 90% first-six-months offer running (verified 2026-10-22, xero.com/uk/pricing-plans). A bill cap matters more than it sounds for AP: it is the thing that decides your plan.
QuickBooks
QuickBooks has supplier bills, receipt capture and approval behaviour that varies by tier, with the richer workflow features sitting on the upper plans. Intuit’s UK pricing page renders its prices in the browser rather than in the page itself, and they did not load when I checked today, so I am not quoting a figure here. The shape of the answer is per-user-ish tiering, five plans, with payroll priced separately and an introductory discount almost always running. Get the current number from Intuit directly and get it in writing for year two, not year one.
Sage
Sage owns AutoEntry, a capable capture product that sells separately from the ledger, so a Sage firm can end up buying capture twice without noticing. Sage’s UK pricing pages refused automated access today, so no prices here either. Structurally, Sage is two products with one brand: Sage Accounting for small businesses and Sage 50 for established ones, and they are not the same automation proposition.
The honest comparison of the three platforms, including which of them automates best, is the companion page: Xero vs QuickBooks vs Sage.
Why you should never retype invoice data
Retyping breaks the point of the pipeline, and for VAT it breaks the rules. If your pipeline is “capture tool produces a CSV, someone keys it into the ledger”, you have built the expensive version of the old process. HMRC’s Making Tax Digital rules require that once data is in your digital records, any further transfer between programs happens by digital link, which HMRC defines as a transfer that needs no manual intervention, and it says in terms that copy and paste is not a digital link[1].
API, a genuine import, linked spreadsheet cells: all fine. A person with a keyboard in the middle: not fine, and also where your error rate lives.
Where to start, and where automation stops
Start with one supplier category; the steps above are the order I’d run it in, and the first one does most of the work. One supplier category, same format every month, low value per invoice, is enough to prove the capture accuracy on your own paperwork rather than on a vendor’s demo set. A month of that tells you more than any trial.
- Invoices that are not invoices. Supplier statements, pro formas, remittance requests and order confirmations all arrive in the same inbox and all look close enough to fool capture. Expect to teach the process the difference.
- Credit notes and part-deliveries. Matching a credit note to the right original across a month boundary is still largely manual, and in firms with returns it is where the time goes.
- The first invoice from a new supplier. No history, so no reliable coding suggestion, so a person. That is correct behaviour, not a failure.
- Judgement coding. Capital versus revenue, mixed-rate supplies, partial exemption, anything a reviewer would query. If the code depends on knowing why the money was spent, the software does not know why.
- Approval chains with a partner who will not log in. The commonest reason AP automation stalls in a practice has nothing to do with the software. Solve the human routing first or do not start.
Accounts receivable has the same anatomy and the opposite emotional problem, and the chasing side is covered in how to automate invoice chasing. The wider view of which practice processes deserve automation first, in what order, is in AI for accountancy firms, and the practice-wide version of the question lives at AI and automation for accountants.
This article is for informational purposes only and does not constitute tax, accounting or compliance advice. Consult a qualified professional for advice specific to your firm.
Sources
[1] HM Revenue & Customs, ‘VAT Notice 700/22: Making Tax Digital for VAT’, paragraph 3.2.1, updated 1 April 2022. Available at: https://www.gov.uk/government/publications/vat-notice-70022-making-tax-digital-for-vat/vat-notice-70022-making-tax-digital-for-vat