Mix different emotional states and expertise levels to test how your agent adapts its communication style.
How toEvaluate Workflows via API
Simulate multi-turn conversations
Test your AI's conversational abilities with realistic, scenario-based simulations
Why simulate conversations?
Testing AI conversations manually is time-consuming and often misses edge cases. It helps you:
- Test how your AI maintains context across multiple exchanges
- Evaluate responses to different user emotions and behaviors
- Verify proper use of business context and policies
- Identify potential conversation dead-ends
1. Create a realistic scenario and be specific about the situation you want to test
- Customer requesting refund for a defective laptop
- New user needs help configuring account security settings
- Customer confused about unexpected charges on their bill
2. Define the user persona
- Frustrated customer seeking refund
- New user needing security help
- Confused customer with billing issues
After defining the user persona, select the field where your agent's replies come from:
3. Advanced settings (optional)
- Maximum number of turns: Set a limit for conversation turns. If no value's set, the simulation ends when complete
- Reference tools: Attach any tools you want to test with the simulation. You can learn more about setting up tools here
- Reference context: Add context sources to enhance conversations. Learn more here
Example simulation
Here's a real-world example of a simulated conversation:
This tests a refund scenario where:
- Customer needs refund for defective product
- Agent verifies purchase
- Policy guides the process
- Must resolve in 5 turns