Test and deploy your agent

Previewing the agent before customers see it

Previewing your Customer Agent before it goes live is a crucial step to ensure it delivers accurate, appropriate, and polished responses. This controlled testing environment lets you experience the agent exactly as your customers would, helping you catch errors, verify Procedures, and confirm personalization settings before deployment. By previewing, you reduce the risk of incorrect or embarrassing answers reaching your audience.

what is preview and why use it

Preview is a feature that allows you to test the Customer Agent from the customer’s perspective before pushing any changes to production. It replicates real customer situations by letting you set conditions such as Audience, topic, Living Context state, and surface (for example, marketing site, product, or helpdesk). This means you can simulate how the agent will behave for different customer profiles and contexts.

Using Preview helps you:

• Verify that updates to your knowledge base or Procedures appear correctly
• Confirm that personalization and context settings work as intended
• Test new components or Moment configurations before publishing
• Reproduce and diagnose confusing or unexpected answers seen in production

Because Preview sessions do not affect live metrics or customer experience, they provide a safe space to experiment and validate changes.

how to start and use preview

To begin a Preview session:

1. Navigate to Test > Preview in your Unless platform.
2. Set the session conditions by selecting the Audience, topic, Living Context, and surface relevant to the scenario you want to test.
3. Click Start Preview to open the agent interface as a customer would see it.

Once the session starts, you can type questions or simulate customer interactions in the conversation panel. The agent will respond just as it would in production.

key features during preview

• Conversation panel: This is where you interact with the agent, identical to the customer view.
• Inspection panel: A side panel that reveals what the agent “sees” internally, including which Sources it used, available Procedures, and current Living Context values. This is invaluable for diagnosing why the agent answered a certain way.
• Session log: Every Preview session is saved automatically. You can share the session URL with colleagues for review or return to it later.

testing different scenarios

You can switch the Audience, topic, or surface mid-session to test different customer journeys or agent behaviors. For example, you might preview how the agent responds to a new product question on the marketing site versus a support query inside the product.

when to use preview and when not to

Preview is ideal for:

• Checking changes after updating a Source or Procedure
• Validating new Moment configurations before going live
• Preparing for pilot reviews or stakeholder demos
• Reproducing and troubleshooting confusing production answers

However, Preview is not designed for large-scale testing or scoring agent performance. For those purposes, use Simulations to test at scale or Quality reports to evaluate answer accuracy over many interactions.

why might preview answers differ from production

If the agent answers differently in Preview than in production, it is usually because the session conditions differ. Double-check that the Audience, topic, and Living Context match the live environment. Also, verify that the live Sources have not changed since you set up the Preview session.

conclusion

Previewing your Customer Agent before deployment is an essential best practice to ensure your AI delivers the right answers in the right tone and context. It provides a safe, realistic environment to test, diagnose, and refine your agent’s behavior, helping you avoid mistakes and improve customer experience. Use Preview regularly as part of your update and release workflow to maintain high-quality AI interactions.