When an agent says or does something you didn’t expect, attach the lead to the assistant and ask why. It reads that lead’s real conversation and the context the agent actually had, explains what happened, and can propose a fix.
Attach a lead
There are three ways to bring a lead into the chat:
| How | Where |
|---|
| Open from the lead’s page | On a lead’s page, use the chat button to open the assistant scoped to that lead. |
| @mention it | In the chat, type @ and pick the lead from the menu (search by name, phone, or email). |
| Paste its link | Paste the lead’s URL (/leads/<id>) into the chat and the assistant attaches it. |
Typing a name as plain text doesn’t attach the lead — pick it from the @ menu so it becomes a chip.
Ask what happened
With a lead attached, ask things like:
- “Why did the agent reply that in its last message?”
- “Review this conversation — did the agent make a mistake?”
- “If something’s wrong, propose a fix so it doesn’t happen again.”
The assistant reads the lead’s full history (across every channel), the data the agent extracted, and the rules it followed, then explains the behavior — separating what it observed from what it’s inferring.
Turn the finding into a fix
If the explanation points to a configuration problem, ask the assistant to fix it. It proposes the change as the same before/after card you confirm — see Edit your agent by chat.
You can attach several leads at once (for example, @mention two) and ask the assistant to compare them — “what makes these behave differently?”.
Read the agent’s setup
You don’t need a lead to inspect what an agent is working with. Ask the assistant about your catalog/offers, your knowledge base, or your business context at any time, and it will read the agent’s configuration and answer. 
