Best Practices
Tips for optimal AI performance
Best Practices for AI Success
Building a great AI agent is an iterative process. Follow these guidelines to ensure high-quality responses and a premium customer experience.
1. Data Quality & Structure
The quality of your agent's responses is directly tied to the quality of your data.
- Keep it Focused: Avoid uploading massive, generic documents. Smaller, focused documents (e.g., "Refund Policy.pdf") are easier for the AI to retrieve.
- Update Regularly: When your business policies change, update your data sources immediately. Enable Auto-Sync for websites.
- Use Q&A for Nuance: For specific, high-stakes questions, use the manual Q&A source. This overrides the generic documents and ensures 100% accurate wording.
2. Refining the Persona
In the Settings tab, provide clear "System Instructions".
- Be Specific: instead of "Help customers", use "You are a helpful support assistant for BotDesk. You describe technical features concisely and always offer to connect to a human if you can't solve a problem after two tries."
- Set Boundaries: Tell the AI what not to do. "Never discuss pricing for custom enterprise plans; instead, capture their email for sales."
3. The Hybrid Strategy (Human Handoff)
AI is powerful, but humans are essential for empathy and complex problem-solving.
- Enable Handoff: Configure a threshold or specific triggers (e.g., "I want to speak to a person") that escalate the chat to your human team.
- Monitor Daily: Check the Activity tab daily to see where the AI might be struggling. Use the "Thumbs Down" filter to find and fix poor responses.
4. Continuous Training Loop
- Step 1: Review logs in the Activity tab.
- Step 2: Identify missing knowledge or wrong answers.
- Step 3: Add new data or Q&A pairs.
- Step 4: Retrain and verify in the Playground.