ImplementationAdoption storyFor: Finance and accounting · Industry operations

KPMG ran 100+ AI agents through a FTSE 250 audit; up to 20 more next

KPMG says AI agents went deeper than its auditors in a parallel audit of a UK-listed company. It won't say whether fees fall, and few clients have clean data.

At a glance

Company
KPMG UK · Professional services · Size not stated
Team and use case
Finance and accounting · External audit of financial statements
Tools
Not disclosed
Stage
Pilot, since 2026
Results
  • AI agents used across the audit: More than 100
  • Audits covered: 1 FTSE 250 company; up to 20 planned
  • Time saved: Not disclosed
  • Reported failures needing a person: 1 (matching rules to documents)
Cost
Not disclosed. KPMG will not say whether time saved would lower audit fees.
People
Two partners, a director, senior managers and junior staff set goals, reviewed the agents' work and challenged conclusions. The effect on junior auditors' jobs and any training are not disclosed.
Evidence
Company claim, no method

The company states a result without saying how it was measured.

What this leaves out: KPMG's head of audit describes the firm's own experiment, relayed by the FT and The Decoder. No count of findings, hours, error rate or cost is given, the client is unnamed, and "deeper" and "better questions" are not defined.

KPMG's UK audit practice ran more than 100 AI agents through every phase of the annual audit of a FTSE 250 company this year, alongside the conventional audit by its own staff, The Decoder reported on 18 November 2026, citing the Financial Times. An agent here is software that carries out a multi-step task on its own: these collected, analysed and documented data and drafted audit conclusions. The official audit opinion was based on the regular audit, not the experiment, which KPMG calls Project Exponential.

In several cases "AI was able to go so much deeper" than the regular team, said Cath Burnet, KPMG's UK head of audit, and in some situations it asked "better questions" than the firm's auditors. That is KPMG describing its own trial. The report gives no count of findings, hours, errors or cost, and does not name the client or the software.

What went wrong, and who checked

  • One failure is reported: a person had to step in because an agent could not correctly match a set of regulations against the company's documents.
  • KPMG says human involvement remained "critical". Staff set the agents' goals, gave them context, reviewed work periodically and challenged conclusions before the opinion was final. Two partners and a director worked with senior managers and junior staff.
  • KPMG says the system gives "full visibility and full traceability", recording what each agent did with a document down to the character and the timestamp.

What it means for the company being audited

  • Your data is the limit. Few companies have data clean enough for the agents to work on reliably, Burnet said. KPMG plans to extend the project to as many as 20 larger audits, and says the technology is not yet available at the scale thousands of audits would need.
  • More gets looked at. Auditors test samples; agents can potentially analyse far more transactions, The Decoder notes in describing the UK regulator's concerns. Wren's read: a finance team should expect more queries, not fewer.
  • Fees are open. KPMG sees "clear time-saving opportunities", according to the FT as relayed by The Decoder, but will not say whether fees would fall. The same report says KPMG International has negotiated lower fees from its own auditor on the argument that AI makes the work cheaper.
  • Extra findings come with a rule. Guidance published days earlier by the Financial Reporting Council (FRC), the UK audit regulator, says sharing what agents turn up beyond the audit, such as inefficient processes or possible savings, can under certain circumstances pose a "distinct threat" to auditor independence if it works as a pitch for consulting. Major control weaknesses must still be reported to you.

"Whatever technology a firm uses, the auditor remains accountable," said Mark Babington, the FRC's executive director of regulatory standards. PwC has run several parallel audits of its own, The Decoder adds.

The people side, and the pushback

The report does not say what happens to junior auditors, who do most of the labour-intensive testing, or how staff were trained. On Hacker News, where the story drew more than 450 points, the strongest objection was that models can ask sharp questions and still miss an obvious one, and that pressure to show a productivity gain pushes reviewers to skim the output. Another commenter argued that testing everything instead of a sample will flag small errors a sample would never have reached, costing the audited company time.

What to do if you run finance or the controller's office

  • Before the next audit is scoped, ask your audit partner in writing whether AI agents will be used on your engagement, on which areas, and who reviews their work.
  • Ask what data, in what format, the firm would need for fuller testing, and give your controller that list now: KPMG's own account is that client data is the constraint.
  • Raise the fee in the same conversation. KPMG International got a lower fee from its auditor with this argument; yours has no reason to offer first.
  • Tell your audit committee how you will handle findings that arrive with an offer of advisory work.
Was this useful for your work?

Take this to work

Forward this

KPMG ran more than 100 AI agents through a FTSE 250 company's audit in parallel with its normal team this year and plans up to 20 more, The Decoder reports, citing the Financial Times; KPMG says the agents went deeper in places but will not say whether fees fall. Our auditor should be asked before the next audit is scoped whether agents will be used on it, what data they need from us, and what it does to the fee.

Questions to ask

  1. Will AI agents be used on our audit next year, on which areas, and does the opinion rely on their work or on a parallel human test?
  2. What data, in what format and how clean, would you need from us, and who pays to prepare it?
  3. If the work takes fewer hours, how is that reflected in our fee?
  4. Who at the firm reviews an agent's conclusion before it reaches us, and can we see the record of what it examined?
  5. How will you handle findings outside the audit so that they do not become a pitch for advisory work?

Questions this article answers

Did AI agents sign off a KPMG audit?

No. The Decoder, citing the Financial Times, reports that the agents ran in parallel with the conventional audit of a FTSE 250 company and that the official audit opinion was based on the regular audit.

What is KPMG's Project Exponential?

It is KPMG's internal name for an experiment in which more than 100 AI agents worked through every phase of a UK-listed company's annual audit alongside the human team. KPMG plans to extend it to as many as 20 larger audits.

Will AI make audit fees cheaper?

KPMG has not said. The Decoder reports that the firm sees "clear time-saving opportunities" but will not say whether fees would fall, and that KPMG International has negotiated lower fees from its own auditor on the argument that AI makes the work cheaper.

What does the UK regulator say about AI in audits?

The Financial Reporting Council's Mark Babington said "the auditor remains accountable" whatever technology is used. The Decoder reports that FRC guidance warns that sharing findings beyond the audit could threaten auditor independence.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail