AI accounting means using artificial intelligence to do accounting work: categorising bank transactions, reading receipts and invoices, matching payments, flagging anomalies, forecasting cash and drafting analysis. It is already built into mainstream bookkeeping software and is changing how accountants and auditors spend their time. What it does not change is who is responsible. Directors must still be satisfied that the accounts give a true and fair view. Taxpayers must still take reasonable care over their returns, and a careless error can cost a penalty of up to 30% of the tax understated. An "AI accountant" is a tool, not a person who carries that responsibility.
What AI does in accounting today
Most of the practical value sits in the high-volume, rules-heavy parts of the job:
- Transaction coding. Bank feeds are categorised automatically from patterns in past postings, with suggested matches against invoices and bills;
- Document capture. Receipts and supplier invoices are read from photos, emails or PDFs, and the supplier, date, amount and VAT are extracted;
- Reconciliation. Payments are matched to invoices across bank, card and payment-provider accounts, including part-payments and batch receipts;
- Anomaly detection. Duplicate invoices, unusual amounts, new bank details on a supplier record and postings outside normal patterns are flagged for review;
- Forecasting. Cash flow projections are built from historic patterns, open invoices and recurring costs;
- Drafting. Generative AI tools produce first drafts of management commentary, variance explanations, client emails and summaries of long documents.
For a small business, the visible effect is that bookkeeping becomes continuous rather than a quarterly catch-up. For a finance team, time moves from processing to reviewing. That is only an improvement if the review is actually done.
What AI cannot do
The limits are as important as the capabilities:
- Judgement. Whether a cost is capital or revenue, whether a payment is a director's loan or remuneration, or whether revenue should be deferred are judgements about facts the software does not see;
- Context. A model trained on last year's postings will repeat last year's mistakes with great consistency;
- Accuracy guarantees. Generative AI can produce fluent, confident text that is wrong, including invented figures and references. Every output that matters needs checking against a source;
- Accountability. No tool signs the accounts, the tax return or the audit report.
Responsibility stays with people
UK law puts the obligations on identifiable people, and none of them is delegated to software.
Accounting records. Section 386 of the Companies Act 2006 requires every company to keep adequate accounting records. AI-coded records are still the company's records, and if the coding is wrong, the records are wrong.
The accounts. Under section 393 of the Act, directors must not approve accounts unless they are satisfied that they give a true and fair view of the assets, liabilities, financial position and profit or loss. "The software produced it" is not a basis for that satisfaction.
Tax returns. HMRC's Compliance Handbook says "every person must take reasonable care", that people meeting unfamiliar transactions should find out the correct treatment or seek appropriate advice, and that remaining uncertainty should be disclosed with the return. Under Schedule 24 to the Finance Act 2007, a careless inaccuracy can attract a penalty of up to 30% of the potential lost revenue. That rises to 70% where the inaccuracy is deliberate and 100% where it is deliberate and concealed, with higher percentages for some offshore matters. A VAT return built on AI-suggested coding that nobody reviewed is a likely candidate for "careless".
AI in audit and assurance
Audit firms are adopting AI for risk assessment, journal testing, document review and data analysis across whole populations rather than samples. The Financial Reporting Council has issued two pieces of guidance. Its AI in Audit guidance was published on 26 June 2025. It covers traditional machine learning and generative AI and sets out principles for proportionate documentation of AI tools. On 30 March 2026 it followed with guidance on generative and agentic AI, focused on the risks those tools pose to audit quality and how to mitigate them. The 2025 guidance states that it does not introduce new regulatory requirements, and the FRC describes the 2026 guidance as codifying good practice. The FRC is explicit that accountability for AI tools and for the quality of audit outputs "remains unchanged".
For companies being audited, this means auditors may ask for complete data extracts rather than samples, and may raise questions from analysis that covers every transaction. Clean, well-structured records make that process faster. Our guide to the impact of AI and cyber risks on audit opinions looks at the other side: how AI-related risks inside a business can affect the audit.
Risks to manage before you adopt
- Data protection. Feeding payroll, customer or supplier data into an AI tool is processing personal data. The UK GDPR applies, so check where the data goes, whether it is used to train models and what the contract says;
- Confidentiality. Public generative AI tools are not the place for unpublished results, deal documents or client information;
- Controls. Automated approvals and bank rules need the same segregation of duties and review as manual processes, or fraud simply moves inside the automation;
- Audit trail. Keep evidence of who reviewed and approved AI-generated postings, so the records still show how each figure was arrived at;
- Over-reliance. Staff who never code a transaction lose the feel for when something looks wrong. Build review into the role.
Will AI replace accountants?
It is replacing tasks rather than accountability. Data entry, matching and first-draft reporting are already largely automated. Judgement, tax planning, sign-off, dealing with HMRC and explaining the numbers to a board are not. The practical change for businesses is that a well-configured system plus a qualified reviewer now produces better records, sooner, than either alone. For more on where the tools fit, see our guide to tax technology.
If you want to automate your bookkeeping without losing control of the numbers, our cloud accounting team can set up the software, bank rules and review process. Our process automation specialists can look at the wider finance workflow and the controls it needs.