Where the agents went deeper
The experiment, known internally as Project Exponential, took place this year. In some cases, AI examined specific issues in greater depth than the regular team and asked more precise questions, said Kat Burnet, head of audit at KPMG in the UK.
KPMG staff assigned tasks to the agents, provided the necessary context, checked their work periodically and challenged their findings before the opinion was prepared. The project involved two partners, a director, senior managers and junior staff. KPMG says the system provides full visibility and traceability, recording the agents’ interactions with documents down to individual characters and timestamps.
The company plans to expand the project to a maximum of 20 major audits. But the technology is not yet ready for use across thousands of audits. Few companies have data clean enough for agents to work with reliably, Burnet said.
Regulator flags auditor independence
A few days before the article was published, the UK audit regulator, the FRC, issued guidance on applying existing auditor-independence rules to AI. The focus is on when auditors can share AI-generated findings with clients without jeopardising their independence. Strict limits or a complete ban apply to some additional services for audit clients: a firm that gives recommendations may later assess those recommendations itself and have an interest in preserving its relationship with the client.
In the FRC’s view, AI agents intensify this conflict. Traditional auditors generally examine samples, while agents could potentially analyse far more transactions. They may also uncover not only audit-related issues, but inefficient processes, potential savings and patterns that a conventional audit might miss. The regulator is concerned that such findings could be used to promote additional services or attract clients to consulting. In some circumstances, sharing them with a client could pose a separate threat to auditor independence.
Mark Babington, the FRC’s executive director of regulatory standards, stressed that responsibility remains with the auditor in all cases. The auditor must explain how AI was used, check its results and take responsibility for the evidence. Auditors are required to report some findings to clients, including serious deficiencies in internal controls; the FRC does not consider this disclosure an additional, non-audit service.
Auditor independence is a particularly sensitive issue in the UK. After construction and outsourcing group Carillion collapsed in 2018, a parliamentary inquiry concluded that KPMG, which had audited the company for many years, failed for 19 years to challenge management’s increasingly unrealistic forecasts. Two KPMG units were fined a total of 21 million pounds as a result.
How AI challenges audit fees
KPMG’s experiment raises the question of how automation will affect the cost of audit services. If agents take over labour-intensive work currently done by large teams of junior auditors, the change could affect the industry’s entire pricing model. According to the Financial Times, KPMG sees clear opportunities to save time, but has not said whether that will lead to lower audit fees.
KPMG International has already agreed to a fee reduction with its own auditor, citing AI’s effect in lowering the cost of the work. Competitor PwC has also conducted several parallel audits.
Similar pressure surfaced at an internal Deloitte meeting in May. According to the Wall Street Journal, leadership showed a chart projecting that consulting’s share of hourly revenue would shrink to a thin strip by 2035 as the overall market grows. One Deloitte consultant took the message to mean that leadership was signalling the old model no longer worked and employees would effectively be replaced by robots.
Dataist’s take: Agents may broaden the scope of audits, but they have not yet removed the need for human review and accountability. And the ability to find more problems at the same time complicates the question of which findings an auditor can share with a client, and whether doing so could become a way to sell consulting.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X