Search papers, labs, and topics across Lattice.
This paper introduces FinProBench, a benchmark designed to evaluate financial AI agents using Role-Grounded Rubric Construction (RGRC), which derives evaluation criteria from actual deliverables produced by financial professionals. The study reveals that while prompt-based rubrics can perform comparably for conventional roles, RGRC significantly outperforms them for specialized roles, highlighting the importance of professional grounding in evaluation standards. By leveraging 1,723 curated deliverables across various financial occupations, the authors demonstrate that RGRC not only captures tacit standards but also reduces rubric construction effort by 6.7 times, enhancing the efficiency of evaluation processes.
Role-Grounded Rubric Construction reveals that professional deliverables are essential for accurately evaluating specialized financial AI agents, outperforming traditional prompt-based methods.
Evaluating financial AI agents requires criteria aligned with real professional work. Existing rubric methods typically derive criteria from task prompts or model outputs, overlooking tacit standards visible only in practitioner deliverables. We introduce FinProBench, a benchmark for professional financial tasks, and Role-Grounded Rubric Construction (RGRC), a reusable pipeline that derives rubrics from deliverables produced by practitioners in the same role. RGRC comprises four stages: Deliverable Collection, Competency Extraction, Rubric Synthesis, and Validation. Its rubrics capture tacit standards, distinguish quality levels, and transfer across tasks within a role. Before analysis, we classified 57 occupations by deliverable genre into 30 prior-rich conventional roles and 27 prior-sparse role-specialized roles. Across all roles, Prompt-only nearly matches RGRC for conventional roles (89.2% vs. 90.7%), but RGRC substantially outperforms it for role-specialized roles (99.1% vs. 78.0%). This split indicates that prompt engineering can approximate rubrics when conventions are well represented in model priors, while professional grounding is essential for standards beyond those priors. FinProBench is built from 1,723 curated deliverables spanning 57 occupations, 8 financial sub-industries, and 161 deliverable types, and releases an initial evaluation set of 20 complete tasks covering 20 roles in 7 sub-industries. With heterogeneous LLM judges and role-level rubrics, human deliverables rank first on average (73.7 vs. 70.3, 70.2, and 69.6 out of 100), while all four systems show overlapping 95% confidence intervals and complementary strengths. Reusing rubrics at the role level reduces estimated per-task construction effort by 6.7 times relative to authoring each rubric from scratch.
