If you run the finance function for an SME, you have almost certainly faced this moment: an invoice lands in the inbox, it looks completely normal, the amount is plausible, and the supplier name matches someone you already pay. You approve it. Three weeks later the real supplier calls chasing payment on an invoice you already settled — except you settled it to a bank account that was never theirs. This is not a rare horror story. UK businesses lose tens of millions of pounds a year to invoice and mandate fraud, and small and mid-sized companies are disproportionately targeted because their accounts payable controls are often manual, trust-based, and stretched thin. This article explains, in plain terms, how AI-driven pattern recognition is now being used to catch fraudulent invoices before payment leaves the bank — using a real worked example of a £12,400 fake supplier invoice.
Why Accounts Payable Is a Prime Target for Fraud
Accounts payable is attractive to fraudsters because it is a process built on repetition and trust. Once a supplier is set up in your system, invoices from that supplier tend to get waved through with minimal scrutiny — the name is familiar, the amounts are within a normal range, and the person approving payment is often juggling dozens of invoices a week. Fraudsters exploit this by either impersonating an existing supplier (changing bank details via a convincing email) or inventing a supplier that looks almost identical to a real one, sometimes with a single letter changed in the company name or a near-identical invoice template. For an SME without a dedicated fraud team, these invoices can sail through because nobody has the time to check every bank account number against a verified source.
How AI Pattern Recognition Actually Works
You do not need to understand machine learning to use this technology, but it helps to know what it is actually doing. Rather than checking an invoice against a single rule — such as ‘is the amount under £15,000’ — AI fraud detection tools build a profile of normal behaviour for every supplier you deal with. That profile includes things like typical invoice amounts, how often invoices arrive, which email addresses and domains are used, the format and sequence of invoice numbers, and which bank account has historically been paid. When a new invoice comes in, the system compares it against that learned pattern in seconds, not against a fixed checklist, and flags anything that deviates. Crucially, it can spot subtle anomalies a busy human would miss — a bank sort code that has never been used before, an invoice number sequence that does not follow the supplier’s usual numbering, or an email domain that is spelled almost, but not quite, correctly.
The key shift is this: traditional controls check whether an invoice looks correct in isolation. AI pattern recognition checks whether an invoice looks correct compared with everything that supplier has ever sent you before. That comparison is where fraud gets caught.
Worked Example: The £12,400 Fake Invoice
Consider a mid-sized SME that regularly pays a genuine supplier, ABC Supplies UK Ltd, for office and warehouse materials. A fraudulent invoice arrives from ‘ABC Supplies Ltd’ — almost identical, but missing ‘UK’ from the name — for £12,400, with new bank details attached and a note saying the supplier has ‘recently changed banks.’ On the surface it looks entirely plausible: the goods described match previous orders, the amount is within the normal range the business pays this supplier, and the layout of the invoice is a close copy of a genuine one. A manual reviewer working through a stack of forty invoices that week could easily miss the missing ‘UK’ and the unfamiliar sort code. The table below shows what the AI system compared when the invoice arrived.
| Detail checked | Genuine supplier on file | Invoice received |
|---|---|---|
| Company name | ABC Supplies UK Ltd | ABC Supplies Ltd |
| Email domain | @abcsuppliesuk.co.uk | @abc-suppliesuk.co.uk |
| Bank sort code | 20-11-22 (used for 3 years) | 40-05-17 (never used before) |
| Invoice numbering | Sequential, e.g. INV-4471, INV-4472 | Out of sequence: INV-9931 |
| VAT registration number | GB 123 4567 89 | Not quoted on invoice |
None of these details alone would necessarily stop payment — a supplier can genuinely change banks, and VAT numbers are sometimes left off invoices by mistake. The value of the AI system is that it weighs all of these signals together and produces a single risk score, rather than relying on one person to notice five small inconsistencies spread across a document. The calculation below shows how the system built up its score for this invoice.
| New bank account never used before (high weight) | +35 points |
| Email domain altered by one character | +25 points |
| Invoice number outside normal sequence | +15 points |
| Company name missing ‘UK’ suffix | +12 points |
| VAT number absent from invoice | +8 points |
| Total fraud risk score (threshold for hold is 50) | 95 points |
A score of 95 out of 100 placed the invoice well above the business’s pre-set threshold of 50, which automatically paused it for manual review instead of allowing it into the weekly payment run. The finance manager received an alert, phoned ABC Supplies UK Ltd on the number held on file (not the number on the suspicious invoice), and confirmed within minutes that no such invoice had been issued and no bank details had changed. The payment was never made.
It is worth seeing how this would have looked in the accounting records if the invoice had been processed in the normal way before the fraud was caught, and how it is corrected once flagged. Many finance teams post purchase invoices to the ledger as soon as they are received, ahead of payment, so that the trade creditor balance is accurate — this is exactly the point at which the AI check intervenes, before the linked payment run executes.
| Account | Dr | Cr |
|---|---|---|
| Purchases / Materials Expense | £10,333 | |
| VAT Input (recoverable) | £2,067 | |
| Trade Creditors (Accounts Payable) — ABC Supplies Ltd | £12,400 |
Provisional posting made automatically when the invoice was received into the purchase ledger, before the AI fraud check and payment run.
| Account | Dr | Cr |
|---|---|---|
| Trade Creditors (Accounts Payable) — ABC Supplies Ltd | £12,400 | |
| Purchases / Materials Expense | £10,333 | |
| VAT Input (recoverable) | £2,067 |
Reversal posted once the invoice was confirmed fraudulent, removing it from the ledger entirely before any payment was released.
AI fraud detection reduces risk, but it does not eliminate the need for human judgement and verification. A flagged invoice should always be confirmed by phoning the supplier on a number taken from existing records — never a number printed on the suspicious invoice itself. Treat a high risk score as a reason to pause and verify, not as proof of fraud on its own, since genuine suppliers do occasionally change banks or update their branding.
What This Means for Your Business
For an SME owner or finance manager, the real benefit here is not the technology itself but what it frees you up to do. Instead of every invoice requiring the same level of scrutiny, AI pattern recognition lets your team focus attention on the small minority of invoices that genuinely look unusual, while low-risk, repeat invoices from established suppliers move through faster. This also strengthens your position with auditors and lenders, since a documented fraud control process — including an audit trail of flagged invoices, risk scores, and verification calls — demonstrates a level of financial control that many SMEs currently lack. It is also considerably cheaper than the alternative: a £12,400 loss is roughly equivalent to the annual running cost of a part-time finance hire for many small businesses, and that is before accounting for the time spent chasing recovery, which is rarely successful once funds have left a UK account for an unfamiliar one.
- Set a clear risk score threshold with your finance team or software provider (commonly around 50 out of 100) above which an invoice is automatically held from the payment run until verified.
- Make bank detail changes a mandatory trigger for manual verification by phone, using a number already on file — never the number printed on the invoice or email requesting the change.
- Review your supplier master file at least quarterly to remove dormant suppliers, since fraudsters often target accounts that are rarely checked but still technically ‘live’ in the system.
- Keep a simple log of every flagged and verified invoice, including the outcome, so you can show auditors, lenders, or your board that fraud controls are active and working, not just theoretical.
Strengthen your accounts payable controls
If invoice fraud risk is keeping you up at night, we can help you build practical, right-sized controls that protect your cash without slowing down your finance team.
