What it does
Roughly two-thirds of a support queue is the same small set of questions asked again and again. This agent reads an incoming ticket, searches your Zendesk Help Center and your own resolved-ticket history for an answer that actually covers it, and drafts a reply that cites the article it came from. When the knowledge base does not answer the question, it does not improvise — it writes an internal note explaining what it looked for and what was missing, tags the gap, and hands the ticket to a human. The deflection decision is the product. The agent replies only when a specific article addresses the specific question asked; partial matches, policy questions, billing disputes, anything touching a refund, an outage, an account action, or an angry customer go straight to a person. A wrong confident answer costs more than a slow one, so the default on any doubt is escalation. Every customer-facing reply is held for human approval before it sends. Escalation, by contrast, is frictionless and ungated — the safe path should never be the slower one. Ticket text is treated as untrusted input throughout: a customer who pastes instructions into a ticket is a customer, not an operator. A useful side effect: the internal notes on escalated tickets become a ranked list of what your knowledge base is missing.
Example prompts
- Look at ticket 48219 and either draft a reply from the Help Center or escalate it with a note.
- Ticket 51004 asks how to reset SSO. Find the article, draft the answer, and cite it.
- Work through the unassigned tickets from this morning and tell me which ones are safely deflectable.
Before you connect
The credentials this agent will ask you for — the full setup is on the Setup tab.
Needs 1 credential (1 required) to connect. See setup
