Aipedia: Accountants

AI anomaly detection in client ledgers

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

AI anomaly detection can help accountants spot unusual transactions, but it does not show that unflagged entries are fine. Over-reliance on what the tool did not flag can lead to failure-to-spot-fraud claims, which turn on the engagement scope.

Part of: AI in Accountants

In short

Machine-learning tools scan client ledgers and bank feeds for duplicate payments, unusual suppliers, odd timing and other patterns that may indicate error or fraud. They are useful, but they create automation bias: staff trust the flags and assume the rest is clean. The ethics Code requires accountants to understand the technology, recognise data limitations (300.6 A2) and explain the limits of the work (R113.3). When fraud later emerges, the client may say the firm should have found it. Whether PI responds, and how strongly the firm can defend, often depends on what the engagement letter said the firm would do.

How accountants use anomaly detection

Bookkeeping and accounts preparation software increasingly includes machine-learning features that flag entries outside normal patterns: a new supplier paid a round sum, duplicate invoices, payments made at weekends, or a sudden change in a payee's bank details. Some firms use separate analytics tools across a whole ledger as part of a year-end review or a special investigation.

These tools are good at pointing to entries worth a look. They are less good at showing what they could not see, such as fraud that mimics normal patterns or transactions recorded outside the ledger.

A model trained on a client's own history also learns that client's habits. If a fraud has been running for some time before the tool is switched on, the fraudulent payments may form part of what the tool treats as normal, and it will not flag them at all.

The duties that apply

113.1 A3: competence includes understanding the technology, including how the tool decides what is unusual. This is part of the Code ICAEW enforces from 1 July 2025, as announced in April 2025.
300.6 A2: data may not be sufficient for effective use of the technology, for example where a client's ledger is incomplete.
R320.11: decide whether output is appropriate for the purpose before relying on it.
R113.3: explain the limitations of the work, including what the tool does not test.
ICAEW's Ethics and AI roundtable report warns against automation bias and "token oversight" and calls for meaningful human review.

What can go wrong

Illustrative scenario: a practice runs an anomaly tool over a client's purchase ledger each quarter and reports "no significant exceptions". An employee has been paying a fictitious supplier in amounts and at intervals that match genuine suppliers, so nothing is flagged. When the fraud is discovered, the client points to the quarterly reports and argues the firm took on a fraud detection role.

The firm's defence depends on what it agreed to do. If the engagement letter said the service was accounts preparation and that the firm would not search for fraud, the client's case is weaker. If the firm marketed the tool as fraud protection, it is stronger. Reports matter too: a quarterly note saying "no significant exceptions" without explaining what was tested can read to a client like an assurance the firm never meant to give. Wording such as "no items met the tool's criteria for review" is more accurate.

How PI usually responds

A claim alleging the firm negligently failed to spot fraud usually falls within a civil liability PI wording, subject to the terms and excess. The key questions are scope and causation: what the firm agreed to do, and whether doing it properly would have stopped the loss. Engagement terms are central; see engagement letters and liability caps. Examples of how claims develop are in accountant negligence claims. Audit work raises separate issues covered in AI in audit.

Practical controls

Describe the tool in client reports as an aid, and state what it does not test.
Keep fraud detection out of scope unless you have agreed and priced it.
Sample unflagged transactions, not just flagged ones.
Record who reviewed each flag and what was concluded.
Use the same approach in AI-assisted due diligence and AI AML checks.

How Apex helps

Apex arranges PI for accountants and can check that your policy's description of services matches what you offer clients. See the accountancy AI overview and AI for year-end accounts.

Sources

Frequently asked

Does using an anomaly detection tool make us responsible for finding fraud?

Not automatically. Responsibility depends on what the engagement letter says and how the service was described to the client. Presenting the tool as fraud protection can widen the duty the firm is taken to have accepted.

What is automation bias?

It is the tendency to trust automated output and to treat the absence of a flag as evidence that nothing is wrong. ICAEW's 2024 roundtable report warns against it and calls for meaningful human review rather than token oversight.

Will PI cover a missed-fraud claim?

Usually a civil liability PI policy responds to an allegation that the firm negligently failed to detect fraud, subject to the terms and excess. The strength of the defence often turns on engagement scope and whether proper work would have stopped the loss.

Should we tell clients we use anomaly detection?

Yes, with a clear statement of what it does and does not do. The PCRT guidance and CCAB draft favour telling clients about AI use, and explaining limitations is a Code requirement under R113.3.

Check your PI covers how you use AI

Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.

Start a PI quote Request a callback

Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.