Training Your Agent
Best practices for training your AI agent to deliver accurate, helpful responses.
Training Philosophy
Training an AI agent is about teaching it what your customers need to know and how you want it to communicate. The better your training data, the better your agent performs.
Key Insight: Start with your 20 most common questions. Perfect those before expanding.
Adding Training Data
Method 1: FAQ Pairs
Add question-answer pairs directly:
Q: What is your return policy?
A: We offer a 30-day return policy on all products. Items must be unused and in original packaging. To start a return, go to Orders → Select Order → Request Return, or just share your order number with me.
Method 2: Document Upload
Upload existing documentation - we'll extract and index the content:
- Product manuals (PDF)
- Help center exports
- Policy documents
- Training guides
Method 3: Website Crawl
Enter URLs and we'll automatically extract content:
https://yoursite.com/help
https://yoursite.com/faq
https://yoursite.com/shipping-policy
Writing Good Training Data
Do This
- • Be specific and detailed in answers
- • Include variations of how customers might ask
- • Add context about when answers apply
- • Use natural, conversational language
- • Include next steps and calls to action
Avoid This
- • Vague or incomplete answers
- • Jargon customers won't understand
- • Outdated information
- • Conflicting answers for similar questions
- • Overly long responses (keep it concise)
Example: Good vs. Bad Training
❌ Bad Example:
Q: Shipping?
A: Check website.
✅ Good Example:
Q: How long does shipping take? / What are shipping times? / When will my order arrive?
A: Standard shipping takes 3-5 business days within India. Express shipping (next-day delivery) is available for metro cities at ₹99 extra. Once shipped, you'll receive a tracking link via email. Would you like me to check the status of a specific order?
Testing & Iteration
Test with Real Questions
Use the test console to ask questions exactly as customers would.
Review Conversations
Check the Conversations dashboard weekly for questions the agent couldn't answer well.
Add Missing Knowledge
Create new FAQ entries for gaps you discover.
Refine Existing Answers
Improve answers that are technically correct but could be clearer.
Training Checklist
- Added top 20 frequently asked questions
- Uploaded product/service documentation
- Added company policies (returns, shipping, etc.)
- Tested with 20+ different questions
- Configured escalation triggers
- Set up weekly review schedule