How ServiceOps AI works
ServiceOps AI is not a chatbot bolted on the side. It is one AI core with two faces:
- The email agent reads incoming requests and answers them — autonomously if you allow it.
- The copilot (Assist) helps agents inside a ticket: drafting replies, summarizing, suggesting next steps, and carrying out tasks you delegate to it.
The principle: grounded answers
Section titled “The principle: grounded answers”A request arrives ↓The AI reads it — and the ticket's real situation ↓It retrieves what YOUR organization knows(knowledge base · AI data sources) ↓It answers from what it found — citing it ↓Confident enough? → it repliesNot confident? → it hands off to your teamThe AI answers from your content, not from its general training. If your knowledge base says nothing about the question, the honest outcome is an escalation to a human — not an invented answer. You control the confidence threshold that separates the two.
The guardrails, in one view
Section titled “The guardrails, in one view”| Guardrail | What it controls |
|---|---|
| Sender policy | Who the agent may answer at all |
| Response mode | Answer autonomously, or draft for human review |
| Confidence threshold | Below it, the AI hands off instead of replying |
| Reply caps | Per conversation and per sender per hour — loops cannot run away |
| Visibility | Answers to outsiders use public articles only |
| AI disclosure | Optionally tells recipients the reply came from an AI assistant |
Always attributable
Section titled “Always attributable”Every action the AI takes is recorded and labeled: an AI reply appears in the ticket as the agent’s named persona, status changes it makes show in the history, and the usage ledger records each call. You can always answer “who did this?” — even when the answer is “the AI, at 3 am, correctly”.
What it never does
Section titled “What it never does”- It never answers from thin air: no grounding, no autonomous reply.
- It never uses internal articles with people outside your team.
- It never exceeds the sender policy — an address the policy excludes gets a human, not a bot.
- It never buries its identity in ambiguity when you enable the disclosure.
Set it up
Section titled “Set it up”Configuration lives in Organization → Settings → AI and takes ten minutes: the email agent, the copilot, data sources, and providers & BYOK.