TauIndianBankBench: Procedure-Aware, Action-Oriented Evaluation of Tool-Using Conversational Agents

TauIndianBankBench introduces a 1,000-task action-oriented benchmark for evaluating tool-using conversational agents in Indian retail banking, checking that assistants follow required procedural steps and make no unrequested changes, with 6.5-12.0% of sessions judged successful when only final records are checked.