Help Center

The Voice AI Prompt Optimizer is a testing and improvement tool that helps you evaluate and enhance your Voice AI Agent's performance using realistic call scenarios. This feature allows you to generate test scenarios, conduct voice-based tests, analyze results, and compare different prompt versions before implementing changes to your live agent. It enables teams to verify agent behavior, enhance call outcomes, and minimize risks before deploying the Voice AI for actual customer interactions.

Understanding the Prompt Optimizer

This tool performs structured, scenario-based call tests on your Voice AI Agent. It assesses whether the agent's responses align with the expected behavior for each scenario and can produce improved prompt variations based on the test outcomes. The Prompt Optimizer moves beyond simple one-off tests by systematically measuring how effectively the agent manages specific caller inputs, navigates workflows, handles objections, utilizes tools, and achieves desired results.

Key Advantages

Using the Prompt Optimizer provides a methodical approach to testing and refining your Voice AI Agent's behavior, whether before its initial launch or after making updates.

Automated Scenario Creation: The tool generates relevant test scenarios automatically, based on your agent's existing prompt, language settings, instructions, and configured actions.
Real Voice Call Testing: It conducts actual test calls, allowing you to assess the agent's behavior in genuine conversations.
Action Verification: It checks if configured actions—such as knowledge base searches, appointment scheduling, sending SMS or emails, and other tools—are triggered correctly during the call.
AI Assessment and Explanation: Each call is evaluated against expected outcomes, and the system provides reasoning for why a result was considered a pass or fail.
Transcript and Recording Access: You can review the complete call transcript and listen to the audio recording of the test.
Prompt Variation Testing: The tool can generate optimized versions of your prompt based on failed test calls and then re-test these new variations.
Safe Testing Environment: Tests are run on a cloned copy of your agent. The live agent remains unchanged until you approve the results and choose to apply the new prompt.
Multi-language Support: Scenario generation, test calls, and evaluations are supported in the agent's selected language.

When to Use the Prompt Optimizer

This tool is most beneficial when you need systematic testing across specific call situations to ensure your agent can reliably handle important workflows, not just perform well in an isolated manual test. Consider using it:

Before directing live customer calls to a new or updated Voice AI Agent.
After modifying the agent's prompt, goals, tools, actions, or knowledge sources.
Following unsuccessful tests from other Voice AI testing methods like Web Call or Phone Call tests.
When you need to validate specific behaviors, such as booking appointments, call transfers, sending SMS/emails, updating CRM records, or accessing the knowledge base.
For testing multilingual capabilities or uncommon call scenarios.
When comparing the performance of different prompt variations against the same set of scenarios.
Whenever you need measurable data on the agent's accuracy, tool usage, or adherence to expected behavior.

Important Considerations

While designed to make agent testing more efficient and reliable, results from the Prompt Optimizer should be reviewed carefully.

Test calls are real voice calls, not text simulations.
Any actions configured in your agent (appointments, SMS, emails, etc.) may be triggered during testing.
Your account is allocated 20 free testing minutes per location per day.
Individual test calls will automatically end after 5 minutes.
The automatic optimization process can generate and test up to 5 prompt variations in a single session.
You can run a maximum of 10 test calls simultaneously.
AI evaluations are not perfectly predictable; treat accuracy scores as indicative rather than absolute.
Evaluation criteria can be strict, so it's important to review the call transcript and recording in addition to the pass/fail score.
The live agent is only updated if you manually choose to apply an optimized prompt.

How to Use the Prompt Optimizer

Proper setup is key to getting valuable results. The process follows a step-by-step workflow: configure the test, run calls, review outcomes, and refine prompts as needed.

Navigate to your CRM's AI Agents section, select Voice AI, and then open the Agent List.
Open the specific Voice AI Agent you wish to test.
From within the agent builder interface, click on the Prompt Optimizer option.
The tool will open, using a cloned copy of your agent for testing. This ensures your live agent remains unaffected.
Click on the option to Generate Scenarios. The tool will automatically create contextual test scenarios based on your agent's configuration to simulate realistic caller interactions and expected behaviors. These scenarios may include details like a caller persona, sample caller input, expected agent response, and expected tool usage, and may be assigned a priority level.
Select the specific scenarios you want to include in your test run.