Testing Lab

Test your agent before it goes live. Run simulated conversations, compare prompt versions, and measure performance.

Test scenarios

Create test scenarios with expected outcomes and run them against any prompt version.

{
  "scenario": "New lead calls to book a cleaning",
  "inputs": ["Hi, I need to book a dental cleaning"],
  "expected": {
    "collects_name": true,
    "checks_calendar": true,
    "books_appointment": true
  }
}

A/B testing

Save prompt versions and run Version A against Version B on the same scenarios. The lab reports which one performs better on each metric.

Metrics

• Completion rate — Did the agent finish the goal?
• Accuracy — Did it collect the right information?
• Politeness score — Was it friendly and professional?
• Transfer rate — How often did it give up?