Simulations in AI Agent Evaluation

Updated 

Simulations enable you to test and evaluate AI Agent performance using predefined scenarios and scenario sets. During a simulation, selected scenarios are executed against the AI Agent and assessed using configured evaluation metrics. This helps validate agent behavior, identify gaps, and measure readiness before deployment.

Evaluation Workflow

Simulations are part of the AI Agent evaluation workflow:

  1. Configure the AI Agent

  2. Generate Scenarios

  3. Create Scenario Sets

  4. Run Simulations

  5. Review Evaluation Results

    Simulation List View

    The Simulations page displays all created simulations with the following details:

    • Simulation Name
    • Scenario or Scenario Set
    • Status
    • Created Date

    Access Simulations

    1. Create or open an AI Agent.

    2. Navigate to Evaluate from the left pane.

    3. Select Simulations.

    4. Click + Simulation.

    Create a Simulation

    Configure Details

    1. Provide the following information:

      • Simulation Name*: Enter a unique name to identify the simulation.

      • Description: (Optional) Add details to describe the simulation purpose.

      • Scenario Set*: Select the scenario set to use for evaluation.

      • Scenarios: Choose all or specific scenarios from the selected set.

    2. Click Select, then Next.

    Review Evaluation Metrics

    • Review metrics configured for the AI Agent.
    • Remove any metrics not relevant to the simulation.

    Example: Focus on resolution metrics for issue handling or communication metrics for quality evaluation.

    Click Start Simulation.

    Simulation Status

    During execution, simulations display the following statuses:

    • Running
    • Completed
    • Failed

    These indicators provide visibility into progress and results.

    Launch Simulations from Scenarios

    You can also initiate simulations directly from:

    • Scenario records
    • Scenario Set records

    The simulation form is pre-filled based on the selected asset.

    Evaluation Metrics

    Simulations use configured metrics to evaluate performance, including:

    • Agent Success Rate
    • Containment
    • Conversation Opening
    • Issue Understanding & Probing
    • Courtesy & Professionalism
    • Empathy
    • Active Listening
    • Communication Clarity
    • Ownership
    • Issue Resolution
    • Conversation Closure