Brief IA
›

KPMG Tests Over 100 AI Agents in Parallel Audit

AI Tools·Tom Levy·

KPMG Tests Over 100 AI Agents in Parallel Audit

KPMG Tests Over 100 AI Agents in Parallel Audit
Key Takeaways
1KPMG tested over 100 AI agents with a FTSE 250 company.
2One agent required human correction to align regulations and documents.
3The firm is considering up to 20 larger audits.
Why it matters : The experience covers all phases of the audit, but the official opinion still relies on conventional work.

KPMG has deployed over 100 AI agents this year for an experimental audit conducted alongside that of a British FTSE 250 company. The official opinion still stems from the conventional procedure. The firm is considering up to 20 larger audits as the British regulator clarifies the independence obligations related to these tools.

The FRC Regulates Findings Shared with Clients

The FRC, the UK authority responsible for accounting regulation, recently published guidelines applying existing independence requirements to AI. These guidelines specify under what conditions the results of an audit using AI can be communicated to the client without compromising that independence. Certain services outside of the audit are prohibited or heavily regulated when they concern an audited client. A consulting firm may indeed need to evaluate its own recommendations while being incentivized to accommodate its client.

According to the FRC, the agents exacerbate this potential conflict: where professionals sample, these tools could examine more transactions. They could thus identify inefficiencies, potential savings, or issues beyond the scope of the audit that might escape a traditional audit. The FRC fears that communicating these findings could be perceived as promoting additional services or even serve to sell consulting. In certain circumstances, the regulator sees this as a specific threat to independence.

Mark Babington, Executive Director of Regulatory Standards at the FRC, reminds that the responsibility remains with the auditor, regardless of the tools used. The auditor must be able to explain their use of AI, discuss their conclusions, and justify the evidence. Major deficiencies in internal controls must still be reported to the client. The FRC excludes these mandatory communications from the category of services other than the audit.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Auditors Had to Correct an Agent

Within KPMG, a member of the Big Four, the experience is internally referred to as Project Exponential. The agents collected, analyzed, and documented data autonomously, then drafted conclusions. Cath Burnet, head of audit at KPMG UK, believes that some of their work went further than that of the traditional team and that their questions were sometimes better.

However, one agent failed to correctly match regulatory texts with the company's documents, necessitating human intervention. KPMG considers this presence essential. Two partners, a director, senior managers, and junior staff participated in the work. The teams provided the agents with their objectives and necessary context, then periodically checked their work and discussed their conclusions before finalizing the opinion.

The firm also claims to have complete visibility into the system's operation. It states that it logs interactions with the documents at the character level, along with their timestamps.

KPMG Does Not Promise to Lower Its Fees

KPMG identifies potential time savings without stating whether these will lead to a reduction in fees. KPMG International has already negotiated a reduction with its own auditor, arguing that AI lowers the cost of work, according to the Financial Times. PwC, for its part, has conducted several parallel audits.

At Deloitte, an internal meeting held in May unsettled consultants. According to the Wall Street Journal, management presented a graph showing that hourly billed consulting revenues would occupy only a small share of a growing global market by 2035. One consultant interpreted this presentation as an acknowledgment of an obsolete model being replaced by robots.

Data Still Limits Broader Deployment

Cath Burnet warns: the technology is not yet accessible at the scale required to conduct thousands of audits. In her view, there are few companies today with sufficiently high-quality data to ensure the reliability of the agents' operation.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.