KPMG Deploys 100 AI Agents To Audit a Company, Says They Asked Better Questions Than Human Accountants

The experimental Project Exponential audit found that autonomous systems could investigate financial issues deeply, while also highlighting the continuing need for human judgment and oversight

Accounting
The Big Four accounting firm used the agents to examine the financial statements of an unnamed FTSE 250 company, and said AI sometimes investigated financial issues more deeply than the human team Apex Virtual Education/Unsplash

KPMG has put more than 100 artificial intelligence agents through an experimental audit of a UK-listed company, with the technology reportedly asking sharper questions than its human accountants in some areas.

The Big Four accounting firm used the agents to examine the financial statements of an unnamed FTSE 250 company, running the exercise alongside a conventional audit by its accountants. KPMG's UK head of audit, Cath Burnet, said the AI sometimes investigated financial issues more deeply than the human team.

'In several cases the AI was able to go so much deeper,' Burnet told the Financial Times, adding that the agents had asked 'better questions' in certain situations.

The experiment, known internally as Project Exponential, offers a glimpse of how agentic AI could change a profession built around checking records, identifying financial risks and challenging the information companies provide.

How KPMG Put 100 AI Agents to Work

Unlike AI tools designed to assist with individual accounting tasks, KPMG's agents were deployed across the different stages of the audit.

They collected financial information, analysed and documented data, prepared draft conclusions and observations, and suggested further steps. The agents could also work autonomously alongside one another, allowing different parts of the audit process to interact.

A team developed the system over four months. KPMG staff supplied information about the company's operations and previous audits, set objectives for the agents and reviewed their work throughout the exercise.

The firm did not disclose the exact size of the team involved in building and running the system. It said two partners and one director worked alongside senior managers and junior employees.

However, the trial was not used to formally approve the company's accounts. KPMG relied on its conventional audit when issuing the official opinion, meaning the AI's findings were tested alongside human work rather than used as the final basis for the decision.

One Mistake Showed Why Human Auditors Still Matter

The experiment also exposed a limitation that could become significant if AI agents are given greater responsibility for financial checks.

Burnet said a human auditor had to intervene when one agent failed to match regulatory requirements with the company's documents. The incident highlighted the importance of professional knowledge and human judgement when interpreting complex financial and regulatory information.

KPMG maintained that human involvement remained critical. Staff periodically checked the agents' work, challenged their conclusions and remained responsible for the final audit opinion.

The firm also said its system provided detailed records of the agents' interactions with documents, including individual characters and timestamps, allowing their work to be traced.

KPMG Plans to Expand Its AI Audit Experiment

KPMG intends to extend Project Exponential to as many as 20 larger audits, with the longer-term ambition of offering AI-based audit services to clients.

There are still practical barriers to wider adoption. Burnet said the technology was not yet ready to support thousands of audits, while many companies lacked the clean, organised data needed for the agents to work effectively.

The potential cost savings also remain uncertain. KPMG acknowledged that AI could reduce the time required for audit work but declined to say whether those efficiencies would translate into lower fees for clients.

Regulatory scrutiny is another consideration.

On 16 November, the UK's Financial Reporting Council published guidance on maintaining auditor independence when using technology, including AI. It warned that sharing technology-generated insights with audit clients could create risks to auditors' integrity, objectivity and independence.

KPMG's experiment suggests that AI agents may be capable of doing more than automating routine checks. But until their findings can be independently assessed and consistently relied upon, the responsibility for questioning the numbers and signing off the accounts remains with human auditors.