Customer support metrics that matter: NPS, CSAT, CES, FRT and AHT
What NPS, CSAT, CES, first response time and average handle time each measure, how to report them honestly, and how they fit in one monthly report.
By the Paperlink AI teamPublished 7 min read
Short answer
NPS, CSAT and CES come from surveys and say how customers feel; first response time (FRT) and average handle time (AHT) come from your help desk and say how the team works. Report each operational metric next to an experience metric, use medians for time, and check the margin of error before you compare survey scores.
A support team can measure dozens of numbers. Five of them answer the questions that managers and customers actually ask: would customers recommend you (NPS), were they happy with this contact (CSAT), was it easy (CES), how long did they wait (FRT), and how long did the work take (AHT). This guide explains what each one tells you, how to measure it without fooling yourself, and how the five fit together in one monthly report.
The five metrics at a glance
| Metric | Question it answers | Formula | When you measure it |
|---|---|---|---|
| NPS | Would customers recommend us? | % scoring 9 or 10, minus % scoring 0 to 6 | Every quarter, or after key moments |
| CSAT | Was the customer satisfied with this contact? | Answers of 4 or 5 on a 1 to 5 scale, divided by all answers | After a ticket closes |
| CES | Was it easy to get the issue solved? | Average answer on a 1 to 7 scale | After a ticket closes |
| FRT | How long did the customer wait for a first reply? | Time from the first message to the first reply from a person | Every ticket, reported as a median |
| AHT | How much agent time does a contact take? | (Talk + hold + wrap-up time) divided by contacts handled | Every contact, by channel |
Each metric has a free calculator with the formula shown: the NPS calculator, the CSAT and CES calculator, and the Erlang C staffing calculator, which works out AHT and turns it into a team size. The calculator guides explain the math in detail. This post is about choosing and reading the numbers.
Two kinds of metric
The five metrics fall into two groups. Experience metrics (NPS, CSAT, CES) come from surveys and say how customers feel. Operational metrics (FRT, AHT) come from your help desk and say how the team works.
Report them in pairs. An operational metric on its own is easy to improve in a way that hurts customers. A team can cut AHT by closing tickets before the problem is solved, and the CSAT for those tickets drops. A team can cut FRT with a fast first reply that says nothing, and CES gets worse because the customer has to write again. Put each operational metric next to an experience metric, and a change in one tells you to look at the other.
| If you push this down | Watch this | Because |
|---|---|---|
| AHT | CSAT | Rushed contacts leave problems unsolved |
| FRT | CES | A fast but empty first reply adds a step for the customer |
| Ticket volume | NPS | Fewer tickets is good only if customers did not give up |
First response time: measure the median, not the average
FRT is the metric customers feel most directly, and the one that is easiest to report in a misleading way. Three rules keep it honest:
- Count only replies from a person, or from a bot that actually answered. An automatic "we received your message" is not a response.
- Say which clock you use. Business hours and calendar hours give very different numbers for a message that arrives on Friday evening. Pick one, write it on the report, and keep it.
- Report the median and a high percentile. Averages hide the customers who waited longest.
An example with ten tickets, first response times in minutes:
2, 3, 3, 4, 5, 6, 8, 12, 45, 240
- The average is 328 divided by 10, which is 32.8 minutes. No customer waited about 33 minutes.
- The median is 5.5 minutes, the middle of the list. Half the customers waited less than that.
- The 90th percentile is 45 minutes: 9 of the 10 customers waited 45 minutes or less.
The one ticket that waited four hours moves the average more than the other nine together. The median tells you what a typical customer gets, and the 90th percentile tells you how bad it gets. Report both, per channel, because customers expect a chat reply much sooner than an email reply.
AHT: an input, not a target
AHT is the main input for staffing: contacts multiplied by handle time is the workload, and the Erlang C calculator turns that workload into the number of agents you need for a target wait. As a target for agents, AHT does harm. Agents who are measured on speed transfer hard contacts, skip notes and close tickets early.
AHT also moves for reasons that have nothing to do with how well the team works. When simple questions go to a help page or a chatbot, the contacts that are left are the harder ones, so AHT goes up while the total work goes down. A worked example with invented numbers:
| Simple contacts (3 min) | Complex contacts (10 min) | Total agent time | AHT | |
|---|---|---|---|---|
| Before | 400 | 600 | 7,200 min | 7.2 min |
| After 300 simple contacts are self-served | 100 | 600 | 6,300 min | 9.0 min |
The team spends 900 fewer minutes on contacts, and AHT still rises by almost two minutes. Read AHT together with the total agent time and the mix of contact reasons, or a good change looks like a bad one.
NPS, CSAT and CES: sample size and timing
Survey metrics are estimates from the customers who answered. Two habits make them useful:
- Look at the margin of error before you compare. With 100 answers, a CSAT change of a few points from one month to the next can be noise. The NPS and CSAT calculators show the margin for your numbers.
- Ask at the same moment every time. A survey sent one hour after a ticket closes and one sent a week later measure different memories. Change the moment and you change the score.
Use each survey for its own job. CSAT and CES describe one contact, so they point at a fix in your support process. NPS describes the whole relationship, so a low score can come from price, product or delivery, not only from support. Read the comments behind the low scores before you act on any of them.
Two numbers that connect support to the business
Churn and customer lifetime value. A support team that lowers churn adds more value than one that lowers cost. The customer lifetime value calculator works out churn and lifetime value from your own numbers. Run it twice, with your current churn and with one point less, and the difference is the number to bring to a budget meeting.
Uptime. Outages create contacts. If your product promises a level of availability, the uptime calculator converts the promise into minutes of allowed downtime per month. Plan the extra contacts for those minutes, and have the status update template ready before you need it.
A one-page monthly support report
| Section | Numbers | Question for the meeting |
|---|---|---|
| Customers | CSAT and CES with margin of error, NPS each quarter | What do the low scores have in common? |
| Speed | FRT median and 90th percentile, per channel | Who waited longest, and why? |
| Workload | Contacts by reason, total agent time, AHT | Which reason grew, and can a help page answer it? |
| Business | Monthly churn, downtime minutes | Did support contacts rise with churn or outages? |
What to add when a chatbot answers first
When a chatbot answers before your team does, three more numbers matter: how many questions it answered, which questions it could not answer, and which answers customers rated as not helpful. In Paperlink AI, the dashboard Overview shows an Answered rate (answered questions divided by all questions), the Unanswered report lists the questions your documents did not cover, and Needs review collects the answers that visitors marked as wrong, out of date or not what they asked. Each unanswered question is a help page to write, and each new page lowers the contacts in the workload row of your report.
Questions people ask
Which metric should a small team start with?
CSAT and FRT. They need no extra tools beyond a one-question survey and your help desk timestamps, and together they cover how customers feel and how long they wait. Add the others when you have enough answers for the margin of error to be useful.
What is a good score?
Scores depend on the industry, the channel, the survey moment and who answers, so benchmarks from other companies rarely compare like with like. Your own trend, measured the same way each month, is the benchmark that matters.
Is resolution time better than first response time?
They answer different questions. FRT says how long a customer waited to hear from you. Resolution time says how long the problem lasted. Track both if you can, and report both as medians.