How to Measure Customer Support Agent Performance

Every support metric can be gamed, and agents will optimise for whatever you measure them on. That is not cynicism about agents — it is how measurement works everywhere. The task is choosing metrics whose gamed version is still the behaviour you wanted.

Metrics worth tracking

First response time

How long a customer waits for the first human reply. Strong metric for chat and email, and one of the few that correlates directly with satisfaction. Hard to game destructively — a fast first response is genuinely good, provided you also measure whether it resolved anything.

Resolution rate

The share of contacts resolved without escalation or a repeat contact. This is the metric that keeps first response time honest, because an instant reply that solves nothing scores badly here.

Repeat contact rate

How often the same customer comes back about the same issue within a set window. Arguably the single most useful quality signal, because it catches the failure mode that every speed metric misses: closing quickly without actually fixing anything.

Quality score

A human review of sampled conversations against defined criteria — accuracy, tone, process adherence, documentation quality. Slower and more expensive than automated metrics, and the only one that reliably catches an agent who is technically compliant and unhelpful.

Documentation accuracy

Whether notes and dispositions are complete and correct. Underrated. Poor documentation costs the next agent time on every subsequent contact and quietly corrupts your reporting.

Adherence and availability

Whether agents are working the hours staffed. Not a quality metric, but a fairness one — you are paying for hours and are entitled to see that they were worked.

Metrics that cause damage

Average handling time as a target

Useful for capacity planning, actively harmful as a performance target. Agents pressured on handling time rush customers, skip documentation, and close contacts that were not resolved — which reappear as repeat contacts you then blame on something else. Track it, plan with it, do not incentivise against it.

Contacts handled per hour, in isolation

The same problem in a different shape. It rewards volume over resolution and punishes the agent who spent twenty minutes properly solving a hard problem.

Satisfaction scores with tiny sample sizes

Response rates on satisfaction surveys are low and skewed toward the very happy and the very angry. On a small team, three responses in a week tells you almost nothing, and treating it as a performance measure is unfair to the agent who happened to get the difficult customers.

Building a scorecard

A workable scorecard for a small team has four or five components, not fifteen.

Weights should reflect your business rather than this table. A sales-focused campaign might weight conversion; a technical queue might weight accuracy heavily. What matters is that quality carries real weight and no single speed metric dominates.

How to run quality reviews

1. Sample randomly. Reviewing only escalations or complaints gives a systematically distorted picture.
2. Review enough to be meaningful. Five to ten conversations per agent per week is a workable baseline for a small team.
3. Score against written criteria the agent has read. Surprise criteria are not coaching, they are gotchas.
4. Calibrate reviewers. Two reviewers scoring the same conversation should land close. If they do not, the criteria are too vague.
5. Deliver feedback with the specific conversation attached. "Your tone needs work" is useless; "in this call at 4:12 the customer asked twice and was not answered" is actionable.

Setting targets that are not fiction

Set the first month's targets from your own baseline, not from an industry benchmark you read somewhere. Measure for four weeks, look at the distribution, then set a target slightly above the current median. Targets pulled from someone else's business are either trivially easy or demoralisingly impossible, and you have no way of knowing which.

Review targets quarterly. A target that everyone hits every week has stopped doing anything. A target nobody has ever reached has stopped doing anything too, in a worse way.

What to do with a struggling agent

Before concluding that an agent is the problem, check the three things that usually are:

• Is the documentation adequate? Agents cannot answer questions nobody has answered for them.
• Are the escalation rules clear? Ambiguity produces either over-escalation or dangerous guessing.
• Is the target realistic? If nobody on the team hits it, it is not an agent problem.

When those are genuinely fine and performance is still poor, coach specifically, then replace if coaching does not work. But check the first three first — in our experience they explain most of it.

---