Customer-support evals must measure whether the customer goal was met. They must also check tool and CRM (customer relationship management) actions, policy answers, and handoffs. Fluent chat alone is not success.
What should a customer-support agent benchmark measure?
Start with end-to-end resolution and the final system state. Then add checks for policy compliance, tool side effects, knowledge grounding, escalation timing, and the context a human agent receives.
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.
CRMArena
watchpartial
2026-10-26
Tests
Tests AI agents on nine CRM (customer relationship management) tasks in a realistic Salesforce environment. Tasks cover three personas: service agent, analyst and manager.
Extends tau-bench with a banking support domain where the agent must search about 700 knowledge documents. It must combine that knowledge with tool calls to make correct account changes.
Simulates customer-service talks in which both agent and user can act on a shared environment. It covers airline, retail, telecom and banking domains, in text and voice.