Aipedia: Auditors
Auditors can use a large language model to score customer contracts by revenue risk, but confidence in the tool must be linked to the substantive testing that follows. If the model misreads a contract, the item may never be tested.
Part of: AI in Auditors
In short
Contract review for revenue testing is one of two worked examples in the FRC's March 2026 generative and agentic AI guidance. An LLM reads customer contracts and classifies them, so the team can select higher-risk contracts for detailed testing. The risks are deficient output (missed or misread clauses), misuse (treating the score as evidence) and methodology that does not meet the standards. The firm and engagement partner keep full responsibility. If a misstatement sits in a contract the model scored as low risk, the claim will test how the selection was designed and reviewed.
Revenue often turns on contract terms: variable consideration, rights of return, bill-and-hold arrangements, side agreements, multiple performance obligations. Reading every contract by hand is slow, so teams typically sample. An LLM can read the whole contract population, extract key terms and score each contract against risk factors the team defines, for example non-standard payment terms or acceptance clauses.
The FRC uses this as a worked example in its Generative and Agentic AI Guidance (30 March 2026). The tool selects contracts for testing; it does not do the testing. The guidance stresses that the confidence placed in the tool should be linked to the substantive work that follows.
Illustrative scenario: a team uses an LLM to classify a software client's customer contracts. Several contracts include acceptance clauses in a schedule attached as a scanned image, which the tool did not read. The model scores those contracts as standard, so none is selected. Revenue on them was recognised before acceptance, and the error comes to light when a customer disputes payment. The client restates, and its lender queries why the audit did not pick it up.
The FRC's three risks show up clearly here. Deficient output: the model omitted terms it never saw. Misuse: the team treated a low score as if it were evidence about recognition. Methodology: the selection approach was not designed to cover the risk of unread attachments.
A quieter risk is drift. If the tool or underlying model is updated between interim and final audit, the same contract can be scored differently, and the team may not notice unless the version is recorded each time it is run.
A claim that the audit failed to detect a revenue misstatement would usually be handled under the firm's PI policy as any audit negligence claim would, subject to the policy terms. The insurer and defence team will want to see why the selection design was reasonable, how the tool was certified, and what review sat over the model's scores. Liability may be limited by a section 534 agreement if one was properly made, and by scope of duty principles. See audit firm PI cover explained and AI audit errors, liability caps and PI.
See also FRC guidance explained and board minute summaries, the other FRC worked example.
Apex arranges PI for firms carrying out statutory audits. A broker can review how your cover responds where AI-selected samples are challenged. See the AI in audit hub.
It can support the selection, as in the FRC 2026 worked example, but the auditor designs the selection approach and remains responsible for it. The output should be reviewed and some low-scored items checked, so the team knows whether the model is reliable for that client.
Generally it is a basis for selecting items rather than evidence about whether revenue is correctly recognised. The FRC links confidence in the tool to the substantive testing that follows. Treating the score as evidence is the kind of misuse the FRC warns against.
Treat them as exceptions to be reviewed manually or selected for testing. A tool that silently scores unreadable documents as low risk creates a gap in coverage that the file should not hide.
Yes, in the sense that the firm must be satisfied it is appropriate before use. The FRC 2025 guidance says firms using third-party tools should still obtain evidence on how the tool meets auditing and ethics requirements, lawful use of training data, versioning and updates, and assurance reports can help.
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 callbackApex 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.