Support metrics worth counting (and the ones that mislead)
A dashboard with twenty metrics is a dashboard nobody looks at. A small team needs the four or five numbers that change what happens tomorrow.
Support reporting has a way of growing until it measures everything and explains nothing. The test for keeping a metric is simple: if it moved by a third, would anyone do something differently this week? If not, stop counting it.
Here are the ones that pass that test for a team under ten people.
The seven worth counting
1. First response time (90th percentile)
How long a customer waits for a reply that moves their question forward. Use the 90th percentile, not the average — the average hides the cases that lose customers. The full argument, and per-channel expectations, is in first response time.
2. Resolution rate
The share of tickets closed as actually resolved, not as "customer stopped replying". A high resolution rate with a slow response time is a competent team that is understaffed. A fast response time with a low resolution rate is a team sending acknowledgements.
3. Reopen rate
How often a resolved ticket comes back. This is the honesty check on resolution rate. A rising reopen rate almost always means answers are being sent that were technically correct and practically insufficient.
4. First contact resolution
Questions resolved without a handover or a "let me check with a colleague". This is the metric a knowledge base moves most directly — it measures whether the person answering had what they needed.
5. Contact rate per 100 orders
Questions received per hundred orders. It is the only metric here that improves by changing something outside support: clearer product pages, a tracking number sent automatically, an honest delivery estimate. A falling contact rate means the business got easier to buy from.
6. Deflection
The share of questions resolved without a person — answered automatically, from a source. Count only questions that ended there: if the customer asked again afterwards, it was not deflected, it was delayed.
7. Draft acceptance
Of the AI drafts offered to managers, how many were sent without edits. This is the practical measure of whether your knowledge base is good, and it is the number to watch in the first month of using AI. See AI answers with sources.
| Metric | Healthy direction | Usually broken by |
|---|---|---|
| First response (p90) | down | channels nobody has open |
| Resolution rate | up | no owner, no status |
| Reopen rate | down | answers that skip the condition |
| First contact resolution | up | knowledge in one person's head |
| Contact rate / 100 orders | down | unclear pages, silent delivery |
| Deflection | up, slowly | thin or stale knowledge base |
| Draft acceptance | up | documents written as legal text |
The ones that mislead
- Total tickets. Goes up when sales grow and down when customers give up. On its own it says nothing.
- Average handling time. Push it down and you get faster, worse answers. It is a capacity number, not a quality one, and it should never be a target for an individual.
- Messages sent per agent. Rewards chattiness. The best answer is often one message.
- CSAT as a single number. Useful as a trend, meaningless as a score, because only the delighted and the furious respond. Read the comments; ignore the decimal place.
- Automation percentage. A bot that "handles" 70% of questions by replying "a manager will contact you" handles nothing.
Pick one metric per problem you are actually trying to fix, and leave the rest as context. A team fixing evening waits needs first response by hour of day, not a scorecard.
How to read them together
Metrics only mean something in pairs:
- Fast responses, low resolution — you are acknowledging, not answering.
- High resolution, high reopens — answers are incomplete, usually missing a condition.
- High deflection, falling CSAT — automation has crossed into questions that needed a person.
- High contact rate, good everything else — support is excellent at absorbing a problem that belongs to the product or the delivery process.
Start with two: first response time and resolution rate. Add the others only when someone asks a question the first two cannot answer.
Frequently asked
Which metrics should a small support team start with?
First response time at the 90th percentile and resolution rate. They cover speed and outcome, and most other numbers are explanations of those two.
Is average handling time a good target?
No. Pushing handling time down produces faster, worse answers. Use it to plan capacity, never as an individual target.
What counts as a deflected question?
One answered without a person and not asked again. If the customer came back with the same question, it was delayed rather than deflected.