AI Support Bot Playbook
1) Define what the bot can and cannot do
- Bot can: answer FAQs, guide self-serve steps, find docs.
- Bot cannot: override security, handle refunds without approval.
2) Add escalation rules
- Escalate when user is angry, stuck, or requests a human.
- Escalate when data is missing or uncertain.
- Escalate security and billing issues by default.
3) Keep the user experience simple
- Short answers with steps.
- Link to the right help page or action.
- Ask clarifying questions when needed.
What to Put in the Knowledge Base
Best sources
- Help center articles
- Product docs and release notes
- Top support ticket resolutions
What to avoid
- Private customer data
- Secrets and API keys
- Outdated policy pages
Rule: if a human needs approval, the bot should escalate.
Examples
Example: Password Reset
- User: “I can’t log in”
- Bot: checks if it’s password vs email issue, shares reset steps
- If failed twice → escalate to support
Example: Refund Request
- Bot: explains refund policy and collects order id
- Bot: escalates to billing team with collected details
Example: Product How-To
- User: “How do I export a report?”
- Bot: step-by-step instructions + link to docs
- Bot: asks if they want a short video or screenshot guide
What to Measure
| Metric | Why it matters |
|---|---|
| Deflection rate | How many tickets are avoided |
| Escalation rate | Whether the bot is safe and honest |
| CSAT after chat | Customer satisfaction and trust |
| Resolution time | Speed improvements |
Want an AI Support Chatbot Built Safely?
We build AI chatbots with guardrails, escalation flows, and knowledge base design that improves support without hurting customer experience.