Short answer: The seven metrics you can act on in customer service are first response time, resolution time, first contact resolution rate, customer satisfaction (CSAT), self-service rate, tickets per hundred orders and the share of recurring questions. Each points to a concrete action: staffing, knowledge, process or product. Measure medians, not means, and always read the numbers together, never one at a time.
Updated
: 3 October 2026
Most ticketing systems show thirty-odd measures, and most customer service managers look at three of them without knowing what to do when one turns red. A metric is only worth following if a bad value leads to a decision. Here are seven that do, with a definition, how to measure it, the most common mistake and what to do about a bad value, and finally how to measure AI support.
1. First response time
First response time is the time from the customer getting in touch to someone giving a substantive reply. It is the first thing the customer judges, and the measure that says most about staffing.
How to measure it. From the ticket being created to the first reply from a person or an AI that actually answers the question, per channel and as a median, ideally with the 90th percentile for those who wait longest. Zendesk distinguishes in its metric definitions between calendar time and business hours. Choose business hours if you do not staff around the clock, but tell the customer which hours apply.
Common mistake. Counting the auto-reply as the first reply, and using the mean. The mean is dragged out by a few tickets that sat over the weekend and hides that most customers get a fast reply, or the reverse.
If the value is bad. Look first at when tickets arrive per hour and when you are staffed. Then at how large a share are recurring questions (metric 7): they can be answered by self-service or a reviewed draft before anyone opens them. What customers expect per channel is covered in the article on response times.
2. Resolution time
Resolution time is the time from the ticket being created to it being resolved for the customer. It is the measure the customer actually cares about, and the one most often hidden behind a nice response time.
How to measure it. From the start of the ticket to it being closed as resolved, as a median and ideally also as the share resolved within a target, for example within eight business hours. Count again if the customer reopens the ticket.
Common mistake. Closing tickets to improve the number. A ticket closed after a "hope that sorted it" and reopened two days later does not have a resolution time of one hour.
If the value is bad. Look for where the tickets stand still. Usually it is waiting for someone else: warehouse, finance, a supplier. Then the fix is a clearer internal flow or a connection that gives the agent the data directly, not more agents.
3. First contact resolution rate
First contact resolution rate, often shortened to FCR, is the share of tickets resolved without the customer having to get in touch again. According to Zendesk's CX Trends 2026, 85 percent of CX leaders say customers leave brands that do not resolve the issue at the first contact.
How to measure it. The share of tickets where the customer does not come back on the same question within a set period, for example seven days, regardless of channel. SQM Group reports in its 2024 benchmark an average around 70 percent across all industries, with wide variation depending on how complex the issues are.
Common mistake. Measuring in one channel only. The customer who emailed and then called counts as resolved in the email channel. And letting the agent mark "resolved at first contact" themselves.
If the value is bad. Read the tickets that came back. Almost always the cause is one of three: the answer was incomplete, the answer was wrong, or the agent lacked the mandate to resolve it. The first two are knowledge questions and are fixed in the knowledge base. The third is a mandate question.
4. Customer satisfaction (CSAT)
CSAT, customer satisfaction score, is the share of customers who answer positively when asked how satisfied they were with an individual ticket. It is the only one of the seven measures where the customer says how it went; how it relates to NPS and CES is covered in the article on CSAT, NPS and CES.
How to measure it. A short question after the ticket is closed, often with two or five options, and an optional comment. Zendesk sends its survey per ticket, not per customer, a day after the ticket is solved. Divide positive ratings by valid responses and multiply by a hundred, decide in advance which ratings count as positive, and report the response rate next to it.
Common mistake. Reading CSAT as a grade for the agent. The customer often rates the decision, such as a rejected return, not the treatment.
If the value is bad. Read the comments on the negative replies and sort them into three piles: wrong answer, slow answer, bad decision. The action is different for each pile. A CSAT that drops after you introduce automatic replies is a clear signal to stop and review.
5. Self-service rate
Self-service rate, sometimes called deflection, is how large a share of customer questions are resolved without becoming a ticket: in the help centre, in the chat or in the form before it is submitted. It is the measure that decides how the queue grows as the company grows.
How to measure it. Simplest as a ratio: the number of help centre sessions divided by the number of tickets created in the same period. Zendesk calls it the self-service score, where 4:1 means four customers try to solve the question themselves for every customer who creates a ticket.
Common mistake. Counting page views as resolved questions. According to Gartner, only 14 percent of customer service issues are fully resolved in self-service. Many visits are followed by a ticket, and then self-service has lengthened the customer's wait instead of shortening it.
If the value is bad. Pull the help centre searches that returned nothing and the articles read just before a ticket was created. That is the list of what is missing or not working. Also make sure the team points to self-service: in a Gartner survey from 2025, 60 percent of agents did not.
6. Tickets per hundred orders
Tickets per hundred orders, or per hundred customers for service companies, is the number of incoming tickets in relation to how much you sell. It is the measure that makes customer service comparable over time as volume changes, and the one that says whether the problem sits in customer service or somewhere else.
How to measure it. The number of tickets created in a period divided by the number of orders (or active customers) in the same period, times a hundred. Split by cause: delivery, returns, product, payment.
Common mistake. Celebrating that tickets are falling when it is really sales that have fallen, or worrying when tickets rise during a campaign that doubled orders.
If the value is bad. A high number is almost never a customer service problem. It is an unclear product page, a delivery time that is wrong, an order confirmation without a tracking link. Take the number per cause to whoever owns the cause.
7. Recurring questions
The share of recurring questions is how large a part of the tickets concern something you have already answered many times. It is the measure that shows how much of the team's time goes to writing the same reply again, and therefore how much the knowledge base and automation can give.
How to measure it. Tag every ticket with the question it concerns, in plain words and not just the system's category, and count how many tickets the twenty most common questions account for. Repeat every month, because the list changes with season and range. A knowledge analysis of the ticket history does the same on a larger set; how that works is described on the knowledge analysis page.
Common mistake. Trusting the ticketing system's categories. "Delivery" says nothing about whether it is tracking, delay, wrong address or damaged parcel, and the actions are completely different. How to build a category list you can steer by is covered in the article on categorising tickets.
If the value is bad. Every question on the list without an approved article is a knowledge gap, a question that recurs without a settled answer. Write the article, publish it where the question is asked and let it be the basis for drafts in the inbox. Then follow whether the question disappears from the list. That is the closest you get to a direct causal link in customer service, and the foundation of a knowledge base that works.
How do you measure AI support, repeat contact and service targets?
With four complementary measures where each ticket is counted once. An AI draft that an agent reviews and sends is assisted handling, a correct automatic answer without an agent is something else, and an executed action, such as a changed order, is a third.
| Measure | Calculation |
|---|---|
| Repeat contact | Tickets with a new contact about the same question within the chosen period / measurable tickets × 100 |
| SLA compliance | Tickets within the defined service target / tickets covered × 100 |
| AI-assisted share | Human-handled tickets with AI support / chosen volume × 100 |
| Automatic resolution share | Confirmed automatically resolved tickets / chosen volume in the flow × 100 |
SLA, service level agreement, is the service target you have promised, for example a reply within one business day. The formulas are suggestions; adapt them to your system and write down the scope. If the denominator is zero, show the number as not measurable, not as zero percent. An automatic resolution share based only on nobody asking for a person overstates the help, so review accuracy and follow repeat contact.
Measured time savings can be converted into capacity in the ROI calculator.
What to do
- Produce the seven numbers for the last 30 days, per channel and question type where possible. Use the median, and write down the period and sample.
- Write one line per metric: value, target, and what you do if the target is missed. If you have no action, the number is not a metric but a curiosity.
- Pick two to improve this quarter, not seven. First response time and recurring questions are a good pair to start with, because the second drives the first.
- Set up a simple weekly report showing the seven numbers side by side, plus the queue as waiting tickets by age and category. One number at a time misleads you.
- Read ten randomly chosen tickets a week. The numbers say what, the tickets say why.
- Take tickets per hundred orders, split by cause, to the next management meeting.
Common questions
Which metric is most important?
First contact resolution rate, if you may only choose one, because it captures both speed and quality and is what customers say decides whether they stay. But none of the numbers works alone. A high resolution rate with a long response time, or a short response time with a low CSAT, are both warning signs that only show when the numbers are read together.
Does an AI draft count as an automatically resolved ticket?
No. If an agent reviews and sends the reply, the ticket is AI-assisted. Automatic resolution only applies to tickets where the customer got a correct answer without an agent and does not come back about the same question. Keep the measures apart, otherwise the same ticket is counted twice and the business case for AI looks too optimistic.
How often should we follow up the metrics?
Response time and queue every day, ideally on a screen the team sees. Resolution time, resolution rate, CSAT and self-service rate every week. Tickets per hundred orders and recurring questions every month, because they need a larger sample to avoid being random. Do not change targets more often than once a quarter; otherwise you never see the effect of anything.
What is a good CSAT?
It depends on industry, channel and how the survey is designed, so compare first with your own history. Look at the trend and the response rate rather than an absolute number. A CSAT that rises while the response rate falls may mean only the satisfied bother to answer.
Do we need an expensive tool to measure this?
No. All seven numbers can be produced from an ordinary ticketing system and a spreadsheet, and recurring questions mostly requires someone tagging the tickets in plain words for a month. Tools make it faster and continuous, but start by settling the definitions and the actions. A tool that counts the wrong definition quickly helps nobody.
Sources
- About native Support time duration metrics — Zendesk help centre, 2026
- Contextual Intelligence Becomes the New Standard for Exceptional Customer Experience in 2026 (CX Trends 2026) — Zendesk, 2025
- Call Center FCR Benchmark 2024 Results by Industry — SQM Group, 2024
- About the CSAT (Customer Satisfaction) user experience for email and messaging — Zendesk help centre, 2026
- Reporting tools for measuring self-service — Zendesk help centre, 2026
- Gartner Survey Finds Only 14% of Customer Service Issues Are Fully Resolved in Self-Service — Gartner, 2024
- Gartner Survey Finds 60% of Customer Service Agents Fail to Promote Self-Service — Gartner, 2025
Originally published on the Supportifier blog.
Top comments (0)