KPMG UK says 100 AI agents sometimes asked better audit questions than human team
KPMG UK said more than 100 AI agents sometimes outperformed human auditors in depth and questioning during a four-month trial on an unnamed FTSE 250 company. The exercise shows agents can coordinate across audit tasks, but human sign-off, regulatory matching and clean data remain necessary. It matters because it demonstrates real deployment in professional services, not just benchmark claims.
KPMG UK conducted Project Exponential using more than 100 AI agents across an audit workflow for an unnamed FTSE 250 company, the Financial Times reported. The agents collected information, analysed documents, drafted conclusions and suggested next steps while working alongside a conventional audit team. KPMG audit head Cath Burnet said the technology went deeper than the regular team in several instances, identifying issues for further investigation and raising questions human auditors had not necessarily asked.
The trial lasted about four months and was not a formally approved statutory audit. An audit partner signed off on the experimental exercise, but its findings were not used to approve the company’s accounts. Human oversight remained central: an auditor had to intervene when one AI agent failed to correctly match regulatory requirements with the company’s documentation. KPMG said data quality also remains a major obstacle because some companies lack clean, structured information.
The result moves AI agents from isolated assistance toward coordinated work across an audit process. Auditing traditionally depends on large teams of junior accountants performing repetitive testing, with managers and partners supervising and making final judgements. In the trial, agents gathered evidence, documented findings and flagged exceptions, placing some routine work with software while senior professionals retained oversight. The regulatory matching error shows human review is still required before statutory reliance.
KPMG plans to expand the initiative to as many as 20 larger audits, while acknowledging the technology is not mature enough for thousands of engagements. The stakes are concrete for audit firms, listed companies and investors: audit quality, cost and staffing are affected where agents are used for evidence collection and exception detection. For now, the documented evidence supports augmentation, not replacement, because human sign-off and data readiness remain necessary.
It affects audit firms, listed companies and investors because AI agents are now being tested on core audit judgment tasks, not just administrative support. The trial shows real operational use while confirming that human review and data quality still limit statutory reliance.
Sources (1)
- [1]European Business Magazine(https://europeanbusinessmagazine.com/ai-agents-better-questions-kpmg/)