Support copilot
Live chat
Relay retrieves passages from the help centre, calls the order tool when it has an order number, and hands off when the evidence is thin.
Try
Answers quote the help centre and show the tool result. The demo never leaves this page.
Admin
Conversations
Every seeded conversation was produced by running the real agent, so the metrics below are computed from transcript data.
Conversation log
Seeded from the fixed demo clock · click a row to open the transcript
| Customer | First question | Confidence | Outcome | First reply |
|---|
Hand-offs
Tickets created with a full transcript
Confidence mix
Last agent message per conversation
Admin
Evaluation
Fifteen labelled questions, fixed clock. Hit@k measures retrieval; grounding checks that the cited article is the labelled one.
Labeled questions
15 questions · 13 retrieval labels · 2 out-of-scope
| Question | Expected | Top result | Conf. | Decision | Check |
|---|
Grounding means the answer quotes at least one article and its citation matches the label. Out-of-scope questions are scored on escalation only, so they are not part of the retrieval denominator.