Support metrics explained
How each support metric is calculated, so you know what you're actually reading.
On this page
First response time
From the customer's first message to the first genuine agent or AI reply.
An automated acknowledgement does not count. If it did, every team could report a two-second first response time by sending "we got your message" and the number would mean nothing.
Resolution time
From arrival to the ticket being marked resolved. A reopened ticket continues its original clock rather than starting a new one — otherwise closing prematurely would improve your numbers, which is exactly backwards.
SLA states
Each ticket resolves to one of:
| State | Meaning |
|---|---|
| Met | Replied before the target |
| Breached | Replied after the target, or not at all |
| Pending | Target has not yet passed |
Targets are calculated in business minutes. A ticket arriving Friday at 16:30 with a four-hour target is due Monday at 11:30, not Friday at 20:30 — so out-of-hours arrivals do not manufacture breaches.
See SLA policies.
Automated resolution rate
Tickets closed by the AI with no human message, divided by all resolved tickets.
Read alongside reopen rate. Any automation number can be improved by lowering your confidence threshold; only the pair tells you whether that was a good idea.
Reopen rate
Resolved tickets that a customer replied to, reopening them, divided by all resolved tickets.
This is the most honest quality metric available, because it is the customer's judgement rather than yours. It is very hard to game — you cannot improve it by changing a setting, only by actually solving problems.
Handoff rate
Conversations the AI escalated, divided by all conversations it handled. A falling handoff rate over weeks means the knowledge base is improving.
A high handoff rate is usually a knowledge problem, not a threshold problem. See Train the AI.
Median, not mean
Response and resolution times are reported as medians. One ticket that sat over a holiday weekend drags a mean into uselessness while the median keeps describing the typical experience.
If you want to understand the tail, look at breach rate rather than switching to averages.
Related