Testing your AI Agent before going live

After building your AI Agent, it's important to test it before going live. Testing helps prevent errors, incorrect answers, or technical issues. Below are several steps you can follow to properly test your AI Agent.

1. Create test scenarios

Design scenarios that represent the conversations your customers will have with the AI Agent. Examples:

  • A customer asks a frequently asked question like, "What are the opening hours?"
  • A customer requests help with an order, such as, "Where is my package?"

Also, consider exceptional situations, such as:

  • A customer uses unclear or vague terms.
  • A customer enters incorrect information, such as a wrong order number.
  • A customer leaves a conversation halfway through.

2. Test in the interactive tester

Use the interactive tester in Watermelon to see your AI Agent in action:

  • Test different scenarios and check whether the AI Agent provides logical answers. Add or adjust instructions to improve results.
  • Ensure that the tone of voice matches your brand. Not satisfied with your AI Agent's personality? Adjust it in Domain Knowledge.

 

Tip: Use the sources in the Interactive Tester. The AI Agent might be pulling outdated information from your website.

3. Test with a test group

Invite a small test group to use the AI Agent in a controlled environment. This can include colleagues, friends, or a group of customers. Have them test the AI Agent in different scenarios and ask for clear feedback on the experience and response quality.

 

Tip: We recommend testing your AI Agent on an internal or test website. This way, your test group cannot access the chatbot settings, ensuring the best test results.

4. Process feedback

Analyze the feedback, optimize your AI Agent by adding or adjusting instructions, and resolve any issues.

5. Check the integrations

Test whether integrations work properly:

  • Are the channels correctly linked? Do messages come through properly on all connected channels, such as the Website Widget, Facebook, and WhatsApp?
  • Are emails, notifications, or Actions correctly executed?

6. Live testing and monitoring

Go live with the AI Agent and monitor its performance:

  • Analyze incoming messages with this article to track how well the AI Agent is performing.
  • Adjust Domain Knowledge/Instructions when incorrect answers appear.

 

With this test plan, you ensure your AI Agent gets off to a great start. Do you have any questions or run into any issues? Contact our support team – we’re happy to help!