FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Bookkeepers
AI auto-categorisation posts transactions to nominal codes from learned patterns, and when it is wrong it is usually wrong repeatedly. A single bad rule can miscode months of entries or several clients before anyone notices.
Part of: AI in Bookkeepers
In short
Cloud software learns how a bookkeeper codes each payee and then codes new transactions the same way, often without review. The danger is systematic error: a wrong rule, copied across months or applied to similar clients, misstates profit, VAT and tax. No bookkeeping body has issued AI rules, but ICB s.36 and s.1 require careful work and truthful records, and in an AAT Comment interview AAT's president said accountability does not transfer to the vendor. PI usually responds to resulting client losses, but aggregation and the policy limit matter when one mistake touches many files.
Coding tools combine fixed bank rules with machine learning. They look at payee, description, amount and the bookkeeper's past choices, then suggest or apply a nominal code and VAT rate. Some may offer coding "predictions" across all clients of a practice. Once trusted, they can process a quarter's transactions in minutes.
The weak points are payees that sell several kinds of goods, transactions that look routine but are capital items, personal spending through a business account, and VAT rates that differ by item rather than by supplier.
These are in the ICB Professional Conduct Regulations. For AAT members, the PCRT AI guidance applies to tax work; scenario 11 deals with tax software that can run without human review, and paragraph 3.5 says members "remain ultimately accountable".
In an AAT Comment interview of 18 August 2026, AAT's president said: "Accountability does not transfer to the vendor when the processing does." The article is opinion, not AAT policy, but it reflects the usual legal position. The client contracts with the bookkeeper. Software terms commonly limit the vendor's liability. So the bookkeeper is the one the client will claim against. See AI software errors: bookkeeper or vendor?.
Illustrative scenario: a practice sets a learned rule that codes all payments to a builders' merchant as repairs with standard-rate VAT. For one client, those payments include materials for a property being built for sale, which should not be treated as repairs. The rule runs for two years. Profit is understated, the client's accountant files on the wrong figures, and the correction brings back tax, interest and professional fees. The same rule was applied to two other trade clients.
Nobody spotted it because each individual entry looked plausible, the VAT returns balanced and the software showed the coding as confirmed. Systematic errors like this are typically found by someone outside the process: the year-end accountant, a lender reviewing accounts, or HMRC. By then the work has been relied on several times, which is why reviewing rules, not just transactions, matters.
Miscoding that causes loss is usually treated as a professional negligence claim within the policy, subject to terms. The harder questions are about size. If one rule caused errors for three clients, the policy's aggregation clause decides whether that is one claim with one limit and one excess, or three. ICB requires at least £50,000 of cover appropriate to the size of the business; AAT sets minimums linked to fee income. A practice that relies on cross-client rules should check that its limit reflects that concentration of risk. See aggregation of PI claims.
Apex arranges PI for bookkeepers; a broker reviews your limit, aggregation wording and how the policy treats AI-assisted coding. Related: AI bank reconciliation and the AI in bookkeeping hub.
In practice, usually the bookkeeper. The client engaged you, vendor terms commonly limit liability, and you chose to accept the coding. Any recovery from the vendor depends on its contract, and your insurer may seek it in your name.
It depends on the aggregation clause in your PI policy. Some wordings group claims arising from the same act, error or omission; others are narrower. The answer affects how many excesses you pay and whether one limit is shared across clients.
It is sensible. AAT members doing tax work are advised by the PCRT AI guidance to include a statement on AI use in engagement letters (1.3). For ICB members there is no specific rule, but setting out how coding is done and reviewed helps if responsibility is later disputed.
No body sets a timetable. A quarterly review, aligned with VAT or MTD submissions, is a practical minimum, with an immediate review when a client changes its trade or adds a new income stream.
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.
