Reference · Glossary
Support AI playbook
Last updated
A **support AI playbook** defines when an assistant may draft replies, which knowledge it may cite, and when to escalate to a human — including tone and refund authority limits.
#When to use
Help desks with documented policies and clear escalation paths.
#When not to
Unsupervised refunds, legal advice, or medical claims. Do not let the bot invent policy.
#Quality checklist
- Allowed knowledge sources (RAG / macros)
- Escalation triggers
- Ban list (promises, legal language)
- Audit samples weekly
#Learn next
- Lesson: `ai-for-customer-support`
- Related: RAG