What they did
KPMG in the UK ran an internal experiment it calls Project Exponential, in which more than 100 AI agents worked on the audit of an unnamed FTSE 250 company next to the regular team. The system took about four months to build. The agents gathered, analysed and documented data on their own and drafted audit conclusions. Staff set their goals, gave them context, checked their work at intervals and pushed back on conclusions before the audit opinion was signed. Two partners and a director worked on it with senior managers and junior staff. The system logs every agent interaction with documents down to the character and timestamp. KPMG has not named the agent product, so it is tagged as custom here.
How it went
Cath Burnet, KPMG's UK head of audit, said that in several cases the agents went much deeper than the human team and sometimes asked better questions than the firm's auditors. In one area a person had to step in because an agent could not correctly match a set of regulations against the company's documents. KPMG plans to extend the trial to up to 20 larger audits and sees clear time savings, but says it cannot yet run this at the scale of thousands of audits.
Worth knowing
Burnet says few companies have data clean enough for agents to work with reliably. The UK regulator, the FRC, has just issued guidance that the auditor stays accountable and must be able to explain and challenge the AI's results, and that sharing extra agent findings with a client can threaten independence. The account comes through press reporting citing the FT, not a KPMG write-up. An audit partner signed off the exercise, but its findings were not used to approve the company's accounts.