BI-Agent and BI-Bench: Towards Automating End-to-End Business Intelligence
Frontier LLMs scored under 50% on the paper’s end-to-end BI benchmark.
The authors built BI-Bench from real public BI projects, extracting dashboard questions and ground-truth answers. Their target workflow covers table discovery, transformations, joins, and final question answering without manual prep. They propose BI-Agent, a tool-augmented system that breaks those steps into structured-data subtasks and coordinates specialized methods. Reported gains reach up to 40 percentage points for vanilla LLMs, with post-training adding up to 30 points. HF Daily Papers' note
The authors built BI-Bench from real public BI projects, extracting dashboard questions and ground-truth answers. Their target workflow covers table discovery, transformations, joins, and final question answering without manual prep. They propose BI-Agent, a tool-augmented system that breaks those steps into structured-data subtasks and coordinates specialized methods. Reported gains reach up to 40 percentage points for vanilla LLMs, with post-training adding up to 30 points. HF Daily Papers' note
score 5