Beyond Accuracy: A Multi-Dimensional Framework for Evaluating Enterprise Agentic AI Systems
Sushant Mehta · 2025
Read Beyond Accuracy: A Multi-Dimensional Framework for Evaluating Enterprise Agentic AI Systems by Sushant Mehta (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-14136.txt.
Description
Sushant Mehta argues that judging enterprise AI agents by accuracy alone is dangerously incomplete. Reviewing twelve major benchmarks and testing leading agents, he finds cost ignored even though similar precision can vary fifty-fold in price. Reliability is also under-measured: agents scoring sixty percent on one run fell to twenty-five percent when eight-run consistency was demanded.
His alternative is CLEAR, a framework scoring cost, latency, efficacy, assurance, and reliability together. On three hundred enterprise tasks, accuracy-first agents cost several times more than cost-aware rivals with comparable performance. Fifteen expert judges found CLEAR predicted production success far better than accuracy alone, with a much stronger correlation.
Catalog record
| CREATOR | Sushant Mehta |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-14136