KPMG’s 100-agent audit found better questions—and one basic regulatory failure

KPMG deployed more than 100 AI agents across every phase of an audit of an unnamed FTSE 250 company, the Financial Times reports. The four-month project ran in parallel with the conventional engagement. An audit partner signed off the experiment, but the firm did not rely on it when issuing the company’s formal opinion.

The agents gathered and analysed records, documented their work, proposed next steps and produced draft findings. KPMG says they sometimes went deeper and asked better questions than the human team. That is the firm’s assessment, and the client remains unidentified, so outsiders cannot compare the two audits or inspect which issues were found. One failure is more concrete: a human had to intervene when an agent could not match a set of regulations to the company’s documents.

Humans did not merely approve the last page. Staff defined objectives, supplied business context and prior audits, checked work periodically and challenged conclusions. Two partners, a director, senior managers and junior staff also built and operated the system. The trial therefore changes the shape of an audit team before it eliminates the team; the expensive judgement moves toward framing, exception handling and responsibility for the opinion.

KPMG plans to extend the method to as many as 20 larger audits, but says the technology cannot yet support thousands and many clients lack sufficiently clean data. The firm sees clear time savings while declining to say whether fees will fall. That is the next test. If agents expand coverage without reducing price, the surplus accrues to the auditor; if fees fall, the model changes the labour economics of a profession whose junior staff traditionally perform the exhaustive checking. Either way, the signed liability remains human.