Artificial intelligence is becoming a key foundation of finance operations. According to Deloitte‘s Q2 2026 CFO Survey, 73 per cent of UK CFOs are now optimistic about AI’s ability to improve business performance, while nearly all expect digital technology investment to continue increasing over the coming years.

Richard Jones, VP of product at ExpenseIn, argues that black box AI cannot police expense fraud and finance teams need explainable pattern recognition.

When it comes to expense fraud, however, businesses need systems that go beyond the excitement and can clearly explain why a claim has been flagged, so that decisions stand up to audits and internal governance.

Nearly every employee spending management problem comes down to the same thing: the gap between money leaving the business and finance seeing the full picture. Most of the AI tools built to catch fraudulent claims are marketed as black boxes: flag first, explain later. Rather than relying on
these opaque models, organisations should prioritise explainable pattern recognition that mirrors
how finance teams already investigate claims.

Fraud is not background noise

The ACFE’s Occupational Fraud 2026: A Report to the Nations analysed 2,402 real-world fraud cases across 143 countries, with total losses exceeding $3.4 billion. Asset misappropriation, the category under which expense fraud sits, remained the most common form of occupational fraud, appearing in 90 per cent of cases.

These figures are a reminder that occupational fraud is not a problem finance teams can afford to treat as background noise. Certain kinds of spend leave patterns that experienced finance professionals are trained to recognise; however, if you cannot show why a claim has been flagged, it becomes difficult to validate during audits.

The best use for AI is not replacing finance teams in identifying these behaviours, but providing explanations and guidance for them to act upon. For example, issues such as missing receipts or duplicate claims are straightforward for finance teams to spot, but AI can provide context behind a flagged claim to understand its origin and source. Approvals are therefore based on evidence rather than assumptions, and the audit trail holds up.

Volume is where it breaks down

For larger organisations, finance teams process numerous submissions, making it challenging to identify suspicious activity. Plenty of teams run expense processes well, but volume is where it starts to break down, with businesses spending hours a month chasing claims, checking logs and  reconciling payments manually.

AI can focus attention where it is needed most by identifying the transactions, across thousands, that require closer review. For example, it can point out claims that fall outside normal patterns or repeated claims just below approval thresholds. Teams can then spend more time investigating genuinely suspicious activity rather than manually checking every submission.

Most people think of AI fraud detection as reactive, catching a bad claim after it has been submitted, but the bigger opportunity is spotting the warning signs. By analysing historical data and wider spending behaviours, AI can highlight changes across employees, departments or expense categories that may show potential misuse in the future. Spotting this is key to preventing fraud before losses escalate. Organisations can then strengthen controls and make more informed decisions about where additional review may be required.

AI is not going to replace the judgement finance teams bring to a suspicious claim, and it should not try to. What it can do is make that judgement faster and better evidenced. As spend volumes grow, that combination of human oversight and explainable AI is what will actually keep fraud losses down, rather than adding another black box for finance teams to second-guess.

The post Expense Fraud Detection Needs Explainable AI, Not Black Boxes appeared first on The Fintech Times.