Glossary
AI evaluation (evals), explained
AI evaluation, or evals, is the set of tests that measures whether an AI system responds correctly, safely and usefully, using test cases and criteria defined before going live.
A typical eval gathers real or simulated questions with the expected answer and checks accuracy, rule compliance, tone and when the agent should have handed off to a person.
Grading can be done by people, by automated rules or by another model acting as a judge, with samples reviewed by humans.
Running the same test set after every change to instructions, knowledge base or model shows whether the change made the agent better or worse.
Conversations aren't support.They're infrastructure.
Step into the new era of the relationship between companies and customers.