Training Your AI
With your knowledge base imported, it is time to fine-tune how the AI responds to customers.
Setting the AI Persona
Navigate to Settings > AI Configuration and define:
Name — Give your AI a name customers will see (e.g., "Bee" or "Support Assistant").
Tone — Choose from presets or write custom instructions:
- "Be friendly but concise. Use simple language. Avoid jargon."
- "Match a professional tone suitable for enterprise clients."
Response Length — Set preferred response length. Shorter responses resolve faster; longer ones feel more thorough.
Configuring Behavior Rules
Set guardrails for the AI:
- Topics to avoid — Competitors, pricing negotiations, legal advice
- Mandatory escalation triggers — Angry customers, refund requests over a threshold, security issues
- Required information — Always ask for order number before processing returns
Confidence Threshold Tuning
The confidence threshold determines when the AI responds vs. escalates:
| Threshold | Behavior | Best For |
|---|---|---|
| 0.85+ | Conservative — escalates often | Sensitive industries (finance, healthcare) |
| 0.70 | Balanced — recommended starting point | Most teams |
| 0.55 | Aggressive — answers more questions | High-volume, low-risk support |
Start at 0.70 and adjust based on your first week of data.
Testing With Real Scenarios
Before going live, test with actual customer questions:
- Open the AI Playground in your dashboard
- Paste 10-15 real customer messages from recent tickets
- Review each AI response for accuracy and tone
- Adjust knowledge base articles or behavior rules where needed
Feedback Loop Setup
Enable the feedback system so customers can rate AI responses:
- Thumbs up/down on each response
- Optional text feedback
- Automatic flagging of negative-rated conversations for review
This data feeds directly into improving the AI over time.
Pre-Launch Checklist
- AI persona name and tone configured
- Behavior rules and guardrails set
- Confidence threshold chosen
- 10+ real scenarios tested successfully
- Feedback system enabled
Next up: Going live with your AI support.