CRMArena
- Tests
- CRM workflows for service agents, analysts, and business operations.
- Best for
- CRM and customer-ops agents
- Classification
- agent benchmark · emerging · mixed
AI evaluation domain · 5 source-backed records
Customer Support AI Benchmarks
Customer-support evals should measure resolved user goals, correct tool and CRM actions, grounded policy answers, and high-quality handoffs—not just conversational fluency.
Decision guidance
What should a customer-support agent benchmark measure?
Start with end-to-end resolution and final system state. Add checks for policy compliance, tool side effects, knowledge grounding, escalation timing, and whether a human receives enough context to finish the job.
5
Benchmarks in this domain
Sources first · alphabetical collection order
Runnability describes access to a usable repository, dataset, or harness—not whether setup is easy.
2026-07-15
tau-bench
- Tests
- Customer-service agents in retail and airline domains using APIs and policy guidelines.
- Best for
- Support-agent reliability
- Classification
- agent benchmark · current · open
tau-knowledge
- Tests
- Knowledge-intensive support extension to the tau-bench family.
- Best for
- Support agents that retrieve policy knowledge
- Classification
- agent benchmark · emerging · mixed
tau-voice
- Tests
- Full-duplex voice customer-service tasks scored against final database state.
- Best for
- Voice support agents
- Classification
- agent benchmark · current · open
tau2-bench / tau3-bench
- Tests
- Customer-service simulation framework with text, voice, policies, tools, multiple domains, and tau3 task-fix updates.
- Best for
- CX eval harness design
- Classification
- agent benchmark · current · open