CXBench: A Benchmark for Process-Compliant LLM Agents in Regulated Customer-Experience Workflows

Key points

• Tool-using language agents increasingly perform regulated customer-service workflows, where a correct final state is insufficient: agents must also provide disclosures, obtain consent, and follow required procedures.1
• Existing tool-agent benchmarks largely score final business states, accepting trajectories that reach correct outcomes through non-compliant processes.1
• We define a false-pass as an outcome-pass run that fails process checks.1
• Process-inclusive evaluation changes the observed model ordering and exposes differences hidden by outcome-only scoring. cxbench makes these failures inspectable through trace-linked criteria.1

Sources (1)

• [1]CXBench: A Benchmark for Process-Compliant LLM Agents in Regulated Customer-Experience Workflows
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 07:46 AM

  We introduce cxbench, a benchmark that evaluates outcome and process compliance separately across six regulated domains.

  Tool-using language agents increasingly perform regulated customer-service workflows, where a correct final state is insufficient: agents must also provide disclosures, obtain consent, and follow required procedures.

Extractive summary: sentences quoted from the sources.