KPMG UK says AI agents sometimes asked better questions than auditors

KPMG UK has disclosed an experiment in which more than 100 AI agents examined the accounts of a major UK-listed company and, in KPMG's account, sometimes investigated financial issues more deeply and asked better questions than the firm's human audit team. The article, from European Business Magazine and citing the Financial Times, gives the project's codename as Project Exponential and says the company was an unnamed FTSE 250 firm.
The agents worked alongside a conventional audit team and took on tasks across the audit process, not just isolated administrative jobs. They collected information, analysed and documented data, developed draft conclusions and observations, and suggested next steps. They could also work alongside one another, dividing responsibilities and coordinating parts of the investigation.
KPMG UK head of audit Cath Burnet said the technology had, in several instances, gone considerably deeper into issues than the regular audit team. It identified areas needing further investigation and asked questions that human auditors had not necessarily raised. This is reported speech, not a direct quote.
The technology was developed over approximately four months. KPMG was careful to separate the trial from a formally approved statutory audit: an audit partner signed off on the experimental exercise, but its findings were not used as the basis for formally approving the company's accounts.
Human oversight stayed essential. During the trial an auditor had to step in when one AI agent failed to correctly match regulatory requirements with the company's documentation.
KPMG plans to expand the initiative to as many as 20 larger audits. The firm acknowledges the technology is not yet mature enough to be deployed across thousands of engagements. Data quality is another significant obstacle, since some businesses lack the clean, structured information agents need to work well.
The article frames the wider stakes for the Big Four. Traditional audits rely on large teams of junior accountants doing detailed, repetitive work under managers and partners; agents could automate much of the data collection, analysis and documentation and leave senior staff to focus on complex judgements and exceptions. KPMG has acknowledged the potential for significant time savings but has not confirmed whether AI-driven efficiencies will translate into lower audit fees. The UK's Financial Reporting Council has warned that AI use must not undermine auditor independence, integrity or professional scepticism. The article also lists concerns about data security, over-reliance on automated conclusions and AI generating inaccurate or fabricated information, and notes that responsibility for an audit opinion stays with the human auditor.
Its own verdict is cautious: the experiment does not establish that AI can replace accountants or independently deliver a reliable statutory audit.
Key facts
- KPMG UK used more than 100 AI agents alongside a conventional audit team on the accounts of an unnamed FTSE 250 company, in a project codenamed Project Exponential (per the Financial Times).
- Head of audit Cath Burnet said the technology in several instances went considerably deeper than the regular audit team and asked questions humans had not necessarily raised.
- It was not a formally approved statutory audit: a partner signed off, but the findings were not used to approve the company's accounts, and an auditor had to correct one agent that mismatched regulatory requirements with the company's documents.
- The technology took approximately four months to develop; KPMG plans to extend it to as many as 20 larger audits but says it is not yet mature enough for thousands of engagements.
- KPMG has not confirmed whether AI-driven efficiencies will lead to lower audit fees.
Why it matters
The agents did more than assist with isolated chores. They collected information, analysed and documented data, drafted conclusions and suggested next steps across the audit process, and coordinated with one another. If the claim that they sometimes went deeper than people holds up, it points to a different operating model for audits, where agents handle data collection, analysis and documentation and senior professionals concentrate on complex judgements and exceptions. That bears on the economics of firms such as KPMG, Deloitte, PwC and EY.
Who it affects
Audit firms, above all the Big Four, and the junior accountants who traditionally do detailed, repetitive audit work. It also touches the audited companies, who do not yet know whether any savings will reach them, and regulators such as the UK Financial Reporting Council, which has set expectations on independence, integrity and professional scepticism.
How to use it
This is not a product you can use; it is an internal KPMG experiment. The practical lesson from the account is the setup: agents ran alongside a conventional team, an audit partner signed off, and a human stepped in when an agent erred. KPMG plans to extend the approach to as many as 20 larger audits, with no timescale given. Clean, structured company data is a precondition for agents to work effectively.
How solid is it
Thin on evidence. The article relies on the Financial Times for the project details and on KPMG's own account, with Burnet's view given in reported speech. No figures are given for time or cost savings, no examples of the questions the agents asked, and the audited company is not named. The source does not say the agents outperformed the humans overall, only that they sometimes or in several instances went deeper or raised questions humans had not necessarily raised. It also does not name the AI models or vendors behind the agents.
Risks and caveats
The exercise was not a formally approved statutory audit, and its findings were not the basis for approving the accounts. One agent failed to match regulatory requirements with the company's documentation and needed an auditor to intervene. The technology is not mature enough for thousands of engagements, and data quality is a significant obstacle. The article also cites concerns about data security, over-reliance on automated conclusions and fabricated output, and the FRC warning. Responsibility for the audit opinion remains with the human auditor. It is unconfirmed whether efficiencies will lower audit fees.
“The KPMG experiment does not establish that AI can replace accountants or independently deliver a reliable statutory audit.”
— European Business Magazine article on the KPMG UK trial