Test your agent

Validate and optimize your AI agent in the playground and with a real WhatsApp conversation before putting it into production.

Before connecting your agent to a real conversation you have two ways to test it: the Playground, the chat that lives to the right of the editor, and a test conversation that arrives on your own WhatsApp.

Why use the Playground?

🚀 Quick validation without needing to send a flow via WhatsApp

• Test complete conversations in seconds
• You don’t need to send real WhatsApp messages
• Experiment without limits until you’re satisfied

🔍 Find problems before your customers do

• Identify confusing or incorrect responses
• Detect gaps in your knowledge base
• Test difficult scenarios and objections

Important: The Playground only tests the agent’s conversational logic. CRM integrations do not run here. You’ll need to test those functions with a real WhatsApp conversation.

🎯 5-step testing methodology

Step 1: Test the basic flow (Happy Path)

Simulate an ideal customer who follows the intended flow:

Scenario: a contact who follows the intended path from start to finish

YOU: "Hi, I'd like some information"
AGENT: [Greeting and first question]

YOU: "My name is John Smith"
AGENT: [Confirms and asks the next question]

YOU: "[e-mail]"
AGENT: [Next question]

[Continue until the agent closes or hands off the conversation]

✅ Verify that the agent:

• Asks questions in a logical order, one at a time
• Confirms the information it receives
• Uses the right tools at the right moment
• Closes or hands off the conversation according to the rules you defined
• Keeps an appropriate tone from start to finish

Step 2: Test real customer scenarios

Simulate typical behaviors of your customers:

🤔 Undecided customer

How to act:

• Ask vague questions: “Can you help me?”
• Request comparisons: “What’s better, A or B?”
• Show doubts: “I’m not sure if this is for me”

What to evaluate?

• Agent clarifies specific needs
• Provides useful comparisons
• Handles uncertainty without pushing

⚡ Rushed customer

How to act:

• Use short phrases: “Price”, “Hours”, “Availability”
• Interrupt explanations: “Just tell me the cost”
• Indicate urgency: “I need this now”

What to evaluate?

• Concise and direct responses
• Essential information first
• Doesn’t over-explain when not asked

💰 Price-sensitive customer

How to act:

• Ask about prices immediately
• Compare with competition: “At X place it’s cheaper”
• Look for discounts: “Do you have promotions?”

What to evaluate?

• Appropriately handles price objections
• Highlights value, not just price
• Offers alternatives when relevant

🔒 Cautious customer

How to act:

• Avoids giving personal information
• Asks many questions before deciding
• Expresses distrust: “Is this safe?”

What to evaluate?

• Respects user privacy
• Provides security information
• Generates trust progressively

Step 3: Test extreme cases and objections

Difficult scenarios reveal weaknesses:

🚫 Boundary tests:

❌ "This is too expensive, do you have anything free?"
❌ "I don't understand anything you're saying"
❌ "I want to talk to a human NOW"
❌ "Why do you need my personal information?"

✅ Evaluate if the agent:

• Handles objections without being pushy
• Explains clearly when it doesn’t understand
• Respects user decisions

Step 4: Validate your knowledge base

Ask specific questions about your business:

📋 Basic information

• What are your hours of operation?
• Where are you located?
• What payment methods do you accept?
• Do you deliver/ship?
• Do you have a warranty?

🔧 Technical details

• Ask about specific products/services
• Request technical specifications
• Ask about compatibilities or requirements
• Inquire about internal processes
• Request comparisons between options

💼 Complex cases

• Situations requiring multiple sources of information
• Questions combining multiple topics
• Scenarios your competition handles differently
• Special cases or exceptions
• Questions about recent updates

Step 5: Iterative optimization

🔄 Continuous improvement process:

1. Identify problems

   Note incorrect, confusing, or incomplete responses during testing.

2. Update your knowledge base

   Add missing information or clarify ambiguous content.

3. Reset and test again

   Use “Reset chat” and repeat problematic scenarios.

4. Validate the improvement

   Confirm that identified problems have been resolved.

Reset the conversation after each change: The “Reset chat” button clears the entire session. Use it every time you modify the agent configuration to test changes.

Signs you need more optimization

🔴 Critical problems:

• Incorrect answers about basic information
• Can’t guide toward the main objective
• Provides contradictory information
• Gets confused with simple questions

🟡 Areas for improvement:

• Responses too long or short
• Difficulty with jargon or specific terms
• Doesn’t handle objections well
• Missing information about special cases

🟢 Ready for production:

• Responds consistently well in all scenarios
• Handles extreme cases professionally
• Provides clear value in each interaction
• Users can easily achieve their objectives

⚡ Tips for efficient testing

Simulate real personalities: Think of 3-4 typical customer types you have and act like each one during testing.

Document problems: Keep a list of problematic responses so you can correct them systematically.

Send yourself a test conversation on WhatsApp

The Playground doesn’t run integrations. To verify everything works end to end, send the agent to your own number.

1. Create the test conversation

   After saving the agent, choose Go to create conversation. Treble builds an outbound conversation with your agent already connected.

   
2. Name it and pick an HSM template

   Outbound conversations need a Meta-approved template to start the contact. You can use one of yours or the one Treble creates by default.

   
3. Add your number

   Enter the WhatsApp number where you want to receive the conversation and send it.

4. Test it from your phone

   Reply the way a real customer would and validate the complete flow.

5. Validate the integrations

   If you use HubSpot or Salesforce, confirm the data was saved correctly and that the tickets or transfers ran.

The test conversation is saved in the Conversations section, so you can send it again as many times as you need while you adjust the agent.

Next steps

Is your agent working well in the Playground? It’s time to:

1. Use your agent in a conversation - Integrate your agent into WhatsApp flows
2. Optimize your knowledge base - Improve response quality
3. Review the AI Agent detail - Available tools and how to adjust the instructions