Walk into most support operations and you will find a dashboard with fifteen metrics on it. Ask which three drive decisions and the room goes quiet. The problem is rarely a shortage of data — it is that the easiest things to measure are not the most useful, and some of the most popular metrics actively damage the thing they claim to improve.
Here is a shorter list that holds up.
The metrics worth managing
Customer Satisfaction (CSAT)
A single question after a resolved contact: how satisfied were you? Reported as the percentage of positive responses.
CSAT is the closest thing support has to a ground truth, because it asks the only party whose opinion settles the matter.
Make it useful by segmenting it. Aggregate CSAT is nearly useless — it moves slowly and hides everything. Broken down by channel, by language, by agent and by contact reason, it tells you precisely where quality is leaking.
Read the comments, not only the score. The free-text field is where the actionable information is. A dissatisfied customer explaining why is worth more than fifty silent fives.
Watch the response rate. Below about 15%, you are mostly measuring people with strong feelings. Treat the number as directional.
A healthy benchmark across most industries is 85–90% positive. Below 80% signals something structural.
First Contact Resolution (FCR)
The percentage of contacts resolved without the customer needing to come back.
FCR is the metric most strongly correlated with satisfaction, and it is also an efficiency measure — every repeat contact costs you twice. It is rare for a metric to align quality and cost this cleanly, which is why it belongs at the centre of the dashboard.
Measure it properly. The useful definition is: no further contact from the same customer about the same issue within seven days. Measuring "agent marked it resolved" measures optimism.
Low FCR usually is not an agent problem. It typically means agents lack the authority or information to finish the job — they cannot issue the refund, cannot see the order system, must hand off to a team that responds in two days. Fix the permissions and the tooling before coaching the agents.
Typical range is 70–75%. Above 80% is strong.
Customer Effort Score (CES)
How much work did the customer have to do to get their problem solved?
CES predicts loyalty better than satisfaction does. Customers rarely leave because an interaction was pleasant or unpleasant; they leave because it was exhausting. Repeating information to three different agents, re-explaining after a transfer, chasing for an update — these are what drive churn.
CES is particularly good at exposing process problems that CSAT misses, because a polite agent can produce a satisfied customer at the end of a genuinely painful journey.
The metrics worth watching
These should inform decisions, not drive targets.
Response time. First response and resolution time, measured at the 90th percentile rather than the average. Averages hide the tail, and the tail is what generates complaints. If your average first response is two hours but 10% of customers wait two days, those customers are your problem.
Contact volume per customer. Rising volume per customer usually means a product or communication problem, not a support problem. This is the metric that turns support into an early-warning system for the rest of the business.
Backlog age. Not the size of the queue — the age of the oldest item. A stable backlog of 200 tickets is fine; a backlog containing something from eleven days ago is not.
Agent attrition. Support quality depends on accumulated product knowledge. High churn means that knowledge never accumulates. Attrition above 40% annually will cap your quality regardless of what else you do.
The metrics doing harm
Average Handle Time as a target
AHT — the average duration of a contact — is the most damaging metric in customer service when used as a target.
Measured, it is informative: a sudden rise may indicate a confusing product change. Targeted, it reliably produces rushed conversations, premature closures and customers who have to come back. You get a better AHT and a worse operation, and the cost reappears as a worse FCR.
Watch it. Never incentivise it.
Tickets closed per agent
Rewards closing, not solving. Agents optimise for the easy queue and close ambiguous tickets prematurely. Volume per agent is useful for capacity planning and useless as a performance target.
Aggregate CSAT as the single headline number
Not harmful in itself, but harmful when it is the only thing reported. It moves too slowly to act on and conceals the segments where quality is failing.
Setting targets that work
Pair every efficiency metric with a quality metric. Response time alone produces fast, unhelpful answers. Response time and FCR produces fast answers that work.
Set targets per segment. One response time target across channels and languages means you will either over-serve some or systematically miss others.
Use percentiles, not averages, for anything time-based. "90% of first responses within one hour" is a commitment. "Average first response of one hour" is compatible with a tenth of your customers waiting a day.
Review targets quarterly. A target that has not been missed in a year is not a target.
A dashboard that fits on one screen
If you reduced everything to what genuinely drives decisions:
| Metric | Segmented by | Why |
|---|---|---|
| CSAT | Channel, language, contact reason | Where quality is leaking |
| FCR | Contact reason | Where agents lack tools or authority |
| First response, 90th percentile | Channel, language | Whether you keep your promises |
| Backlog age | Channel | Whether anything is being abandoned |
| Contact volume per customer | Contact reason | Early warning for the product |
| Agent attrition | Team | Whether quality is sustainable |
Six metrics. Everything else is diagnostic — useful when investigating something specific, noise on a daily dashboard.
The question behind the numbers
The most useful review meeting we have seen does not start with the dashboard. It starts by reading five real conversations aloud — two rated well, two rated badly, one chosen at random — and asking what they reveal. The metrics then explain how widespread what you heard actually is.
Numbers tell you the size of a problem. Conversations tell you what it is.
We report these metrics by channel and by language as standard for the teams we run, with the raw data available to clients rather than a summary. If your current reporting does not let you see quality per language, let us know what you are working with.