Plan requirement
| Subscription | Suite Team or higher, Support Team or higher |
| Access | Admin |
A conversation counted as resolved and a person still handled it. The counting rule explains it: only conversations that were not escalated are evaluated for automated resolutions, so a resolution on a ticket an agent worked is usually a second conversation rather than a miscount. Where the labels themselves contradict each other, a failed handover is the third explanation.
Symptoms
- The reporting shows an automated resolution on a ticket an agent worked.
- Resolution counts look higher than the conversations you recognise as automated.
How the counting works
| Per conversation, not per ticket | One user's visits across channels, browsers or devices count as separate interactions. |
| Escalated conversations are excluded | They are not evaluated for automated resolutions at all. |
| An LLM verifies the rest | It checks the text of the conversation to confirm the request was resolved without human intervention. Conversations that pass become a Verified resolution; those that fail become a Contained resolution, which does not draw on your allowance. |
This is also the billing unit: automated resolutions are what your AI agent usage is measured and charged on, and plans include a resolution allowance with overage beyond it. That makes an unexplained count worth understanding rather than shrugging at.
Were there two conversations?
The customer asked something, the AI agent answered it, and the exchange ended. Later they came back about something else and reached a person. Two separate conversations on one ticket, one resolved and one escalated. Reporting is right; the ticket view makes it look like one story.
Did the customer come back on the same subject?
The conversation ended without a handover, so it was evaluated and counted, and the customer came back afterwards on the same subject.
This one matters. The resolution is technically correct and practically wrong: the answer did not hold, and you paid for it.
The mechanism is a clock. A conversation the customer stops replying to is checked and counted once it counts as ended, and that verdict cannot be reversed. When it ends depends on the channel: email after 72 hours, messaging after 2 hours by default, voice the moment the caller hangs up. On messaging that clock is short, which is often the whole explanation. So an answer that only failed on the fourth day is already in the figures as a success.
Or did the handover fail?
A third case, and the one that produces two labels that contradict each other. In messaging, the AI agent can reach the escalation step and try to hand over while no human agent is available to take it. The handover then does not complete: the AI agent raises a ticket for follow-up and finishes the conversation itself.
You see Custom resolution: Escalated to agent, because the escalation step was reached, alongside Conversation status: Automated resolution, because nobody joined the live conversation. Both are accurate; they describe different parts of the same flow. If this is your pattern, the thing to fix is availability at that moment, not the reporting.
How to tell them apart
Read the transcript and the subsequent conversation. Same subject within a short period means the second case; a different question means the first.
Read the ticket as well as the transcript, because the two diverge here. Once that clock has run and the conversation is counted, a later customer reply lands on the ticket but is not added back into the AI agent's conversation log. So a transcript that appears to end in silence can sit on a ticket where the customer answered on day four, and only the ticket shows it.
Why this is worth measuring
Repeat contact after a resolution is the number that keeps the resolution rate honest, and the dashboard does not report it, so you have to assemble it. A high resolution rate with high repeat contact means customers are being sent away rather than helped.
Track them together and neither can mislead you.
What to do about a pattern
Where one subject shows this repeatedly, the article behind it answers the general question and not the one customers have. That is a content fix, and it is usually a small one.
Comments
0 comments
Article is closed for comments.