Human Handoff & Self-Healing

A conversation reaches your team four ways: the visitor asks for a human ("can I talk to someone"), frustration signals accumulate, the visitor leaves negative feedback, or — with auto-escalation enabled — Orac's confidence on a query falls below your chosen threshold. In every case the visitor gets an acknowledgment, the conversation is flagged in your inbox with its reason, and your notification preferences decide who hears about it and how fast.

From the moment a conversation needs a human, Orac stops answering in it. While it waits for your reply, visitors who write again get a short acknowledgment — never silence, never the bot talking over your incoming answer. Once you reply, the visitor sees it labeled as a human response, live in their widget.

Escalated conversations sit in the Conversations inbox with unread badges, SLA timers, and team assignment. Reply directly from the thread — with @mentions, canned responses, and round-robin or manual assignment if you run a team. Resolution flips the conversation back to the bot when you're done.

Every human reply carries a Save as knowledge action. One click pairs your answer with the visitor's question and promotes it to a verified knowledge entry — embedded, prioritized in retrieval, and answered by Orac directly the next time anyone asks. The escalations you handle this week are the tickets that never reach you next week.

The same mechanism powers the Intelligence tab's gap healing: detected knowledge gaps come with drafted answers you approve or edit. Between the two paths — heal gaps proactively, save replies reactively — the knowledge base converges on what your visitors actually ask, not what you guessed they'd ask on day one.