Recent empirical studies comparing human CPAs and advanced artificial intelligence models across core accounting workflows have revealed striking performance disparities. As firms evaluate how to integrate automation into daily operations, evaluations from technology developers provide fresh perspective on where machine intelligence excels and where human expertise remains essential.
Evaluating Month-End Close Performance
In an evaluation conducted by artificial intelligence development firm Mercor, twelve certified public accountants with an average of five years of experience were pitted against advanced language models on simulated month-end close scenarios. The tasks required participants to examine working files, apply specific calculation rules, and generate accurate summary tables. Human participants operated within virtual desktop environments designed to replicate standard workplace setups, with each professional given a three-hour window per task and guaranteed compensation regardless of completion speed.
Conversely, the automated agents operated via an application programming interface gateway and completed the identical workflows in a fraction of the time. While human accountants typically required hours to complete the complex reconciliations and frequently scored below fifty percent accuracy without assistance, the automated systems achieved high accuracy rates consistently. Interestingly, trials involving human-machine collaboration showed that participants using assistant tools often required significantly more time per task, occasionally overriding correct automated suggestions during final reviews.
Transaction Categorization and Ledger Accuracy
A separate benchmark organized by ledger solutions provider Digits focused on transaction categorization, examining how well artificial intelligence models and human bookkeepers handle high-volume entries. The dataset comprised two thousand financial transactions collected from active businesses, reflecting typical spending distributions and varying charts of accounts. A panel of outsourced accountants, ranging from senior students to seasoned professionals, established the human baseline across structured team assignments.
- High-volume daily transaction sorting relies heavily on contextual judgment and specific business conventions.
- Top-tier automated models achieved categorization accuracy that closely matched or exceeded the human benchmark panels.
- Minimizing manual intervention in foundational bookkeeping allows financial teams to redirect focus toward strategic advisory roles.
The findings across both evaluations highlight a rapidly evolving operational landscape for financial departments. While automated platforms demonstrate remarkable speed and precision in structured tasks such as month-end reconciliations and categorization, industry participants emphasize that human oversight continues to play a vital supervisory role in validating complex financial records and managing broader client relationships.