AI agents ask better questions than KPMG accountants in audit trial
KPMG's latest experiment reveals that AI agents are outperforming human accountants in audit trials, asking deeper questions and identifying more issues. This surprising finding has ignited a familiar Hacker News debate: is this a testament to AI's rapid advancement or a scathing indictment of human professional services? The implications for the future of auditing, and white-collar jobs more broadly, are now front and center.
The Lowdown
KPMG UK's "Project Exponential" has demonstrated that AI agents can sometimes conduct financial investigations more deeply and ask better questions than their human counterparts during audits of a major FTSE 250 company. This experiment, involving over 100 AI agents, signals a significant shift in the application of artificial intelligence from mere support tools to autonomous roles within the financial services industry, challenging traditional auditing models.
- The AI agents actively collected, analyzed, and documented data, developing draft conclusions and even coordinating tasks among themselves, showcasing a level of autonomy beyond simple assistance.
- While an audit partner signed off on the experimental exercise, the findings were not used for a formal statutory audit, and human oversight proved essential when an AI agent failed to correctly match regulatory requirements.
- KPMG plans to expand the initiative to more audits, but acknowledges current limitations such as technological maturity and the need for clean, structured data.
- The potential implications for the "Big Four" accounting firms are substantial, suggesting faster audits and lower operating costs, though whether these efficiencies will translate to reduced client fees remains an open question.
- The transition also raises regulatory concerns about auditor independence, data security, and the possibility of AI-generated inaccuracies, with the ultimate responsibility for audit opinions remaining with human auditors.
This experiment by KPMG highlights that the discussion is no longer if AI will integrate into the audit process, but rather the extent to which machines will take over these functions and redefine the role of human accountants.
The Gossip
The Big Four's Big Blind Spot
Many commenters weren't surprised by AI outperforming KPMG, suggesting it says more about the current state of human auditing—often described as exploitative, inexperienced, or compromised by misaligned incentives (e.g., trading audits for consulting contracts). Some even quipped about KPMG's German nickname, "Kinder Prüfen Meine Gesellschaft" (children auditing my company), to underscore the perceived lack of quality in human-led audits.
AI's Auditing Acuity: Beyond Human Habit
Discussion focused on AI's inherent advantages in tasks like auditing. Commenters posited that AI agents lack human tendencies towards laziness, can process exhaustive data rather than samples, and excel in structured, rule-bound tasks. However, others countered that AI still has "blind spots," can get stuck on "obvious" problems, and might exhibit a different kind of "laziness" due to training constraints, emphasizing that human judgment and oversight remain crucial.
Automating the Algorithm of Employment
Many participants debated the broader implications of AI for jobs, particularly in "rules-based" professions. Some asserted that auditing, being essentially "counting beans and reading rules," is an obvious target for AI automation. This sparked a comparison to other professions like coding, where a significant portion of work also involves following rules. Skeptics, however, warned against "childlike predictions" of rapid GDP growth, citing regulatory hurdles and the human element still required for nuanced judgment and non-verifiable tasks.