Testing Agents

This comprehensive guide will walk you through testing your agents in the playground to ensure they work correctly and provide helpful responses to your users.

​

Before You Start

You’ll need:

• A Marketrix account with at least one connection
• At least one agent created and configured
• Knowledge uploaded to your connection (recommended)
• A clear understanding of what you want to test

​

Step 1: Access the Playground

1. Log into your Marketrix account at app.marketrix.ai
2. Click “Playground” in the left sidebar
3. You’ll see the playground interface with connection and agent selection

​

Step 2: Select Your Connection

1. Click the “Connection” dropdown
2. Choose the connection you want to test with
3. Verify the connection is correct
4. The connection should show as “Active”

Tips:

• Choose the connection where your knowledge is stored
• Make sure the connection is properly configured
• Verify the connection has the knowledge you want to test

​

Step 3: Select Your Agent

1. Click the “Agent” dropdown
2. Choose the agent you want to test
3. Verify the agent is linked to the selected connection
4. The agent should show as “Active”

Tips:

• Choose the agent you want to test
• Make sure the agent is properly configured
• Verify the agent has access to the knowledge you want to test

​

Step 4: Start Testing

Once you’ve selected your connection and agent, you can start testing:

​

Basic Test

Start with a simple greeting:

• Type: “Hello, can you help me?”
• Press Enter or click Send
• Review the agent’s response

What to look for:

• Agent responds appropriately
• Response is friendly and helpful
• Agent introduces itself or offers help
• Response matches your brand tone

​

Knowledge Test

Test if your agent uses your knowledge:

• Type: “What does your company do?”
• Press Enter or click Send
• Review the agent’s response

What to look for:

• Agent references your knowledge base
• Response includes information from your documents
• Information is accurate and current
• Response is relevant to your business

​

Step 5: Systematic Testing

​

Test Basic Functionality

Test fundamental capabilities:

1. Greeting and Introduction:
   • “Hello, can you help me?”
   • “What can you help me with?”
   • “Who are you?”
2. Company Information:
   • “What does your company do?”
   • “What are your business hours?”
   • “How can I contact you?”
3. General Questions:
   • “What products do you offer?”
   • “What services do you provide?”
   • “How do I get started?”

What to verify:

• Agent responds to all questions
• Responses are helpful and accurate
• Agent uses your knowledge base
• Responses match your brand tone

​

Test Knowledge Integration

Test if your agent uses your knowledge effectively:

1. Product Questions:
   • “What are the features of [your product]?”
   • “How much does [your service] cost?”
   • “What’s included with [your offering]?”
2. Process Questions:
   • “How do I [specific process]?”
   • “What’s the process for [specific task]?”
   • “Can you walk me through [specific procedure]?”
3. Policy Questions:
   • “What’s your return policy?”
   • “What are your terms of service?”
   • “How do you handle [specific situation]?”

What to verify:

• Agent finds relevant information in your knowledge
• Responses are accurate and complete
• Agent provides specific details from your content
• Information is up-to-date and relevant

​

Test Interaction Modes

Test Tell Mode (Answering Questions):

• “What are your business hours?”
• “How do I reset my password?”
• “What features does this have?”
• “Can you explain how this works?”

Test Show Mode (Demonstrating Tasks):

• “Show me how to create an account”
• “Show me how to use this feature”
• “Show me how to navigate your website”
• “Show me how to set up the product”

Test Do Mode (Performing Actions):

• “Help me fill out this form”
• “Guide me through the checkout process”
• “Automate this task for me”
• “Complete this process for me”

What to verify:

• Each mode works as expected
• Responses are appropriate for the mode
• Agent provides the right type of help
• User experience is smooth and helpful

​

Test Edge Cases

Test challenging scenarios:

1. Unclear Questions:
   • “I don’t understand what you’re saying”
   • “Can you help me with something not related to your business?”
   • “What if I’m not satisfied with your service?”
2. Complex Questions:
   • “I’m having trouble with [specific issue]”
   • “Can you help me with a complex problem?”
   • “What should I do if [specific situation]?”
3. Follow-up Questions:
   • Ask a follow-up question to test context
   • See if the agent remembers the conversation
   • Test if the agent can build on previous responses

What to verify:

• Agent handles unclear questions appropriately
• Agent stays on topic when appropriate
• Agent escalates complex issues when needed
• Agent maintains context throughout conversations

​

Step 6: Document Your Testing

​

Keep Testing Notes

Record what works well:

• Note successful test cases
• Record helpful responses
• Document good agent behavior
• Track positive user experiences

Record issues and problems:

• Note questions the agent can’t answer
• Record inaccurate or unhelpful responses
• Document areas for improvement
• Track problems that need fixing

​

Create a Testing Checklist

Basic functionality checklist:

• Agent responds to greetings
• Agent provides company information
• Agent answers general questions
• Agent uses knowledge base effectively

Knowledge integration checklist:

• Agent references uploaded documents
• Agent provides accurate information
• Agent answers product/service questions
• Agent explains processes and procedures

Interaction modes checklist:

• Tell mode works correctly
• Show mode provides step-by-step guidance
• Do mode performs actions appropriately
• All modes provide helpful responses

Edge cases checklist:

• Agent handles unclear questions
• Agent stays on topic appropriately
• Agent escalates when needed
• Agent maintains conversation context

​

Step 7: Iterate and Improve

​

Based on Testing Results

If your agent works well:

• Document successful test cases
• Note what makes responses effective
• Consider deploying to production
• Plan for ongoing monitoring

If your agent needs improvement:

• Identify specific issues
• Update agent instructions
• Add missing knowledge
• Test again to verify improvements

​

Common Improvements

Improve agent instructions:

• Make instructions more specific
• Add more business context
• Clarify how to handle different situations
• Include more examples

Add more knowledge:

• Upload additional documents
• Add more videos
• Include missing information
• Update outdated content

Refine agent behavior:

• Adjust tone and style
• Improve response quality
• Better handle edge cases
• Enhance user experience

​

Testing Best Practices

​

Test Systematically

Start with basics:

• Test fundamental functionality first
• Progress to more complex scenarios
• Test different interaction modes
• Try edge cases and challenging questions

Test thoroughly:

• Don’t rush through testing
• Test multiple scenarios
• Try different question types
• Test with various user personas

​

Test Realistically

Use real user questions:

• Test with questions real users might ask
• Use common use cases
• Try different user personas
• Simulate real conversations

Test edge cases:

• Try unclear or ambiguous questions
• Test with questions outside your scope
• Try complex or multi-part questions
• Test error handling and edge cases

​

Document Everything

Keep detailed notes:

• Record all test results
• Note what works and what doesn’t
• Document improvement areas
• Track progress over time

Create test cases:

• Develop standard test questions
• Create test scenarios
• Build a testing library
• Use for ongoing testing

​

Common Testing Issues

​

Agent Not Responding

Problem: Agent doesn’t respond to questions Solutions:

• Check that connection and agent are selected
• Verify the agent is active and configured
• Check that knowledge is properly assigned
• Test with simple questions first

​

Agent Gives Generic Responses

Problem: Agent gives vague or unhelpful answers Solutions:

• Check that knowledge is properly assigned
• Verify knowledge content is relevant
• Improve agent instructions
• Test with more specific questions

​

Agent Doesn’t Use Knowledge

Problem: Agent doesn’t reference your uploaded content Solutions:

• Check that knowledge is linked to the connection
• Verify the agent is linked to the same connection
• Ensure knowledge content is relevant to questions
• Test with questions that should use your knowledge

​

Agent Tone Doesn’t Match Brand

Problem: Agent sounds too formal/casual for your business Solutions:

• Review and update agent instructions
• Adjust the tone and style in instructions
• Test with different question types
• Refine the personality and voice

​

Next Steps

After testing your agents thoroughly, you’re ready to:

1. Use the chat interface - Master the playground interface
2. Refine your agents - Improve your agents based on testing
3. Create integrations - Deploy your tested agents to your website
4. Monitor performance - Track how well your agents work in production

Thorough testing in the playground is essential for creating successful AI agents. Take time to test systematically and thoroughly, and you’ll have agents that provide excellent user experiences.

​

Getting Help

If you need assistance with testing:

• Check the troubleshooting guide for common issues
• Review testing best practices for effective testing
• Test systematically to identify issues
• Contact support for additional help

You’re now ready to test your agents effectively and ensure they provide the best possible experience for your users! 🚀