Enterprise · · 5 min read
What Happens to Audits When AI Goes Deeper than Humans?
You'll learn how KPMG's AI agents performed a full company audit, what this means for audit accuracy and efficiency, and the challenges of integrating AI into core business assurance.

Imagine over 100 diligent workers. They pore over every financial detail of a large company. They cross-reference documents, draft conclusions, and suggest next steps. All this, without constant supervision.
This isn't a scene from a futuristic office. It's what KPMG, the global professional services giant, recently achieved in a groundbreaking pilot.
KPMG deployed a fleet of over 100 AI agents. They conducted an entire audit of an unnamed FTSE 250 British company.
The news of this pilot, codenamed 'Project Exponential', broke on November 18, 2026. It ran parallel to a traditional human audit.
The results were striking. KPMG's UK head of audit, Cath Burnet, noted the AI went "so much deeper" into financial statements than human auditors. It even asked "better questions" in some situations.
Until now, AI in auditing has mostly been confined to specific, isolated tasks, like crunching numbers or detecting anomalies.
No large UK audit had ever used the technology end-to-end, across every single phase.
But KPMG's pilot suggests a fundamental shift. AI agents collaborated and operated autonomously through complex processes.
This isn't just about faster number-crunching. It's about a new way of seeing a business's financial health.
That raises three questions.
Question 1
What Did KPMG's AI Agents Achieve in Their FTSE 250 Pilot?
KPMG's 'Project Exponential' marked a significant departure. It moved away from previous AI applications in auditing.
KPMG deployed over 100 AI agents. They worked across an entire audit of a listed British FTSE 250 company. This was unlike using AI for a single, specialised task.
These agents weren't just assisting. They were actively involved in every audit phase. They collected and analysed data. They meticulously documented findings. Agents drafted conclusions and suggested next steps for the audit process.
For context, according to research notes from Google Search, earlier uses of AI in auditing were typically limited to specific, often repetitive tasks, like data extraction or simple reconciliation.
The human audit team worked alongside the AI agents, providing a crucial benchmark. What surprised many was the AI's depth of analysis.
Cath Burnet, KPMG's UK head of audit, highlighted the AI's capabilities. She said it went "so much deeper" into financial statements than human auditors. It even "asked better questions" in some situations.
How did this 'deeper' analysis actually work? You see, the AI agents functioned like a highly coordinated team of junior accountants, but with superhuman speed and memory.
The agents processed vast information. They cross-referenced it against regulations and company policies. A comprehensive record of their actions was maintained throughout.
The system provided "full visibility and full traceability." It recorded interactions with documents down to individual characters and timestamps. This ensured a clear audit trail.
And that brings us to the second question.
Question 2
How Do Autonomous AI Agents Perform Complex Audits?
The key to the AI agents' effectiveness was their ability to work with one another "without step-by-step direction."
This doesn't mean they were entirely unsupervised. Instead, KPMG staff set the overall goals for the agents. They briefed them on the specific business being audited. They also provided context from previous audits.
Think of it like a fleet of delivery drivers (the AI agents) working for a logistics company (KPMG). The company owner (human auditors) sets the delivery target. For example, ensuring all packages reach their destination on time with correct documentation.
The drivers are given the overall route plan and the inventory (the company's financial data and audit objectives). Each driver (AI agent) then autonomously figures out their best delivery route. They communicate with other drivers if coordination is needed.
What if a driver encounters an unexpected roadblock? Or a package is damaged? (An AI agent might struggle to reconcile a regulation with client paperwork.) They flag it for the central dispatcher (human intervention).
In KPMG's pilot, human intervention was indeed required once. An AI agent couldn't align relevant regulations with the client's paperwork. This highlights that while autonomous, human oversight remained "critical," as stated by Cath Burnet.
So, how much autonomy did these agents truly have? Staff reviewed the agents' output. They challenged their findings. This acted as quality control and strategic guidance, rather than micro-management.
This brings us to the third question.
Question 3
What Does "Project Exponential" Mean for the Future of Business Audits?
KPMG's successful pilot is not just a one-off experiment. The firm intends to apply this end-to-end AI approach to as many as 20 bigger audits. The ultimate goal is selling AI-based audits to clients in the future.
This points to a strategic shift. AI could become a standard tool in the auditor's arsenal, offering "clear time-saving opportunities."
However, the road ahead isn't without its bumps. Burnet cautioned that the technology cannot yet run thousands of audits. Why? Few companies possess data clean enough for the agents to process effectively.
This means that for many businesses, preparing their data for AI-driven audits will be a significant undertaking.
Furthermore, regulatory bodies are watching closely. The Financial Reporting Council (FRC) has already issued guidance. It warns that AI-derived insights could potentially threaten auditor independence.
There's also the question of trust. A Centre for Public Interest Audit study found 23% of audit committee chairs had "limited" confidence in AI.
KPMG didn't confirm if audit fees would fall. However, the promise of deeper insights and efficiency could reshape an audit's value proposition. According to its global assurance leader, Andy Hammond, PwC has also run its own parallel audit projects to test AI.
This suggests the industry is collectively exploring how AI can enhance, or fundamentally change, what an audit truly is.
The Bigger Picture
KPMG's 'Project Exponential' offers a glimpse into a future. Here, financial scrutiny could reach remarkable new levels of depth and breadth.
Imagine a financial review that not only catches errors but proactively identifies subtle risks that even the most seasoned human eye might miss.
This shift isn't just about automation; it's about augmentation. It's about empowering auditors to ask better questions. It helps them achieve a more profound understanding of a company's financial health. This transforms an essential, but often slow, process.
The real challenge now is not just building smarter AI, but building smarter businesses ready to work with it.
The next generation of audits is here.
Quick Answers
What did KPMG's AI audit pilot achieve?
KPMG deployed over 100 AI agents to perform a full audit of a FTSE 250 company, running parallel to a human audit. The AI was noted to go 'so much deeper' than the human team in analysis and even asked 'better questions' in some situations.
How do AI agents conduct an audit autonomously?
The AI agents worked with one another without step-by-step direction, collecting data, analysing it, documenting findings, drafting conclusions, and suggesting next steps. Human staff set overall goals, briefed the agents, and reviewed their output, providing critical oversight.
What are the implications of KPMG's AI audit for businesses?
KPMG's pilot suggests future audits could offer deeper insights and time-saving opportunities. However, it also highlights challenges like the need for cleaner company data, regulatory concerns regarding auditor independence, and the ongoing need to build trust in AI-driven assurance processes.
Sources
- Resultsense18 Nov · 14:55 IST
- https://streamlinefeed.co.ke/18 Nov · 12:23 IST
Facts are credited to the outlets above. The analysis, calculations and Signal Check are by the Finnect AI Insights desk.
