BankerToolBench: Evaluating AI Agents in End-to-End Investment Banking Workflows
BankerToolBench (BTB) is an open-source benchmark testing AI agents on complex, end-to-end investment banking workflows that take junior bankers up to 21 hours. The benchmark requires agents to navigate data rooms, use industry tools, and produce multi-file deliverables like Excel models and pitch decks. Findings show frontier models fail nearly half the grading criteria due to severe cross-artifact consistency issues.