Four support metrics worth watching (and three that mislead you)
Most support dashboards optimise for the number that's easiest to compute. Here are the four we'd actually run a team on, and the three that quietly reward the wrong behaviour.
Support metrics have a bad habit: the easiest number to compute becomes the number everyone is judged on, and within a quarter the team has reorganised itself around gaming it. These are the four we'd keep, and the three we'd treat with suspicion.
Keep: median first response time
Median, not mean. Averages in support are dominated by a handful of conversations that sat over a weekend, so a genuinely fast team looks mediocre and a slow team can hide behind a good week. The median tells you what a typical customer actually experienced, which is the only thing the number is for.
Keep: resolution rate
Of the conversations that arrived in a period, what fraction were actually closed? It's boring and it's honest. A queue that grows faster than it closes is a staffing problem that no amount of response-time improvement will fix, and this is the metric that surfaces it first.
Keep: deflection, defined honestly
If you run an AI agent, you want to know what share of conversations it resolved without a human. The honest definition is strict: the customer asked, the AI answered, nobody escalated, and the customer didn't immediately come back with the same question. Anything looser and you're counting conversations the AI merely touched.
Keep: satisfaction, split by who handled it
One blended CSAT number tells you almost nothing once part of your support is automated. Split it — AI-resolved versus human-resolved — and it starts answering real questions. If AI satisfaction is well below human, your knowledge base has gaps. If they're close, you can safely widen what the agent handles.
Suspicious: ticket volume
Volume going down is not obviously good. It can mean your product improved, or it can mean customers gave up on contacting you. Volume is a denominator for other metrics, not a goal on its own.
Suspicious: average handle time
Reward short conversations and you'll get short conversations — closed early, reopened tomorrow. If you track it at all, track it alongside reopen rate, or you're paying people to end conversations before they're finished.
Suspicious: first-contact resolution
Attractive, and almost impossible to measure honestly. Most implementations count a conversation as first-contact resolved if nobody replied again, which also describes every customer who quietly gave up. Unless you can distinguish “solved” from “abandoned”, you're measuring silence.
The test for any support metric: if the team optimised for this number and nothing else, would customers be better off?
Median first response and resolution rate pass that test. Handle time doesn't. That's the whole framework.