AI beats licensed accountants on speed and accuracy, but still can’t close the books without supervision
What happened
A new Mercor study shows AI models now outperform licensed Certified Public Accountants on structured accounting tasks in both speed and accuracy. This marks a major shift from just eighteen months ago, when AI lagged well behind human professionals. However, when tested on the more complex APEX Benchmark, no AI fully completed all required tasks. Without human oversight, AI cannot yet close the books independently.
Why it matters
This study pressures accounting workflows to rethink the balance between automation and human review. Faster, more accurate AI can reduce routine workload, lower errors, and speed month-end processes. At the same time, the inability of AI to fully handle complex closing tasks alone exposes limits that still require skilled accountants in the loop. For finance teams and their leadership, this means tightening supervision rather than removing it. Mistakes from unsupervised AI could raise audit costs and regulatory risks.
What to watch next
Look for developments that expand AI’s grasp beyond structured tasks into less predictable accounting challenges. Vendors pushing AI adoption will need to show tighter guardrails, transparency, and audit trails. Keep an eye on enterprise trials and whether these tools shift hiring needs toward hybrid roles combining finance expertise and AI monitoring skills. Regulatory attention on AI’s role in financial reporting may also increase as usage expands.
AI Quick Briefs Editorial Desk