AI Models Outperform Licensed Accountants on Structured Tasks, Study Finds
A new study by Mercor shows that current AI models now surpass licensed CPAs in speed and accuracy on structured accounting work, a reversal from 18 months ago. However, on the more complex APEX Benchmark, no model completes all tasks, and human oversight remains essential for closing the books.
Artificial intelligence models have overtaken licensed certified public accountants (CPAs) in both speed and accuracy when performing structured accounting tasks, according to a study by Mercor. The findings mark a significant shift from just 18 months ago, when AI lagged far behind human professionals in this domain.
The study evaluated AI performance on routine accounting activities that follow clear rules and formats, such as data entry, reconciliation, and basic calculations. In these areas, the models demonstrated superior efficiency and precision compared to their human counterparts. This suggests that AI is becoming increasingly capable of handling the repetitive, rule-based aspects of accounting work.
However, the picture changes dramatically when tasks become more complex. On the APEX Benchmark, a more demanding assessment designed to test higher-order accounting skills, no AI model was able to fully complete all the tasks. This indicates that while AI excels at structured, predictable work, it still struggles with the nuanced judgment and problem-solving required for comprehensive financial management.
«Without human oversight, AI still can't close the books on its own,» the study notes. Closing the books involves a series of steps to finalize financial records at the end of an accounting period, including adjustments, accruals, and verification. This process often requires professional judgment, interpretation of ambiguous information, and an understanding of broader business context—areas where AI remains unreliable.
The implications for the accounting profession are twofold. On one hand, AI can serve as a powerful tool to automate routine tasks, potentially freeing accountants to focus on more strategic and analytical work. On the other hand, the technology is not yet ready to replace human expertise entirely. The study underscores the continued need for qualified professionals to supervise AI outputs and handle complex decision-making.
Mercor's research adds to the growing body of evidence that AI is transforming knowledge work, but also highlights its limitations. As models improve, the boundary between what machines can do independently and what requires human intervention continues to shift. For now, the accounting industry faces a future where collaboration between humans and AI is likely to be the norm, rather than full automation.
The study did not specify which AI models were tested or the exact metrics used, but its conclusions align with broader trends in AI development. While AI has made rapid strides in recent years, achieving full autonomy in professional domains like accounting remains a distant goal. The need for human oversight persists, especially in tasks that demand ethical judgment, contextual understanding, and accountability.
11
