Skip to main content
Deep Analytics → Agent KPIs. Sales agents only — oversight roles are not measured here.
Three different questions are being answered in one table, and mixing them up is the usual cause of “these numbers look wrong”:
  • Owned — chats the agent owns now
  • Messages sent — messages the agent sent
  • Replied — customer questions the agent answered
They will not tie to each other, and are not supposed to.

​
The columns

​
What does not count as a reply

This matters more than any other rule here. Only a message sent by a CRM user counts. The AI bot, your page’s away message, and anything typed in Facebook’s own inbox all arrive with no user behind them. They:
  • never stop the response clock, and
  • never earn credit for the agent.
So a conversation the bot handled beginning to end shows as unanswered by a human, because it was. That is intentional: it stops automation flattering your response times.

​
Bursts

When a customer sends five messages in a row, the clock starts at the first and the whole burst counts as one question — not five, and not five separate clocks.

​
Unanswered

Counted against the conversation’s owner. The table distinguishes three cases:
  • Unanswered — nobody ever replied
  • Amber — the bot or the page answered on its own, no human did
  • Gray — still inside its response threshold; not late yet

​
Conversions

Credited to the lead’s assigned agent at the moment the lead enters a win stage — including custom win-stage names — whether moved from the leads board or the inbox.

​
Handovers

Deals that took two agents:
  • ↙ received — closed after the chat was handed to this agent. Counted in their Converted.
  • ↗ handed off — closed by somebody else after leaving them. Not counted in their Converted.
Each deal is still counted once in the team total, against whoever owned it at the win. So per-agent columns will not sum to the team figure, by design.

​
Satisfaction

The average of scores customers gave, projected onto 1–5 whatever scale the survey used, credited to whoever owned the conversation. Always read it next to the response count — “4.8 from 3 of 200 surveys” is not a 4.8. The table shows answered-out-of-sent for exactly this reason.

​
Getting ratings at all

Satisfaction is empty unless something asks for it. Add the Ask for rating block to an automation — see Triggers and blocks.