Most support teams report one satisfaction number upward. It goes in the board pack, it gets compared to last quarter, and almost nobody asks the question that would make it useful: which part of the work is it describing?
Read across 32,000 companies and 1.2 billion tickets and the answer is uncomfortable. The same company, in the same period, scores 95.15% CSAT on conversational channels and 88.00% on ticketing. Those are both real and both honestly calculated. Blend them and you get a number that describes neither.
This is a guide to reading your own support metrics: which ones carry information, which ones mislead, and what to do when two reputable benchmark sets disagree with each other, which they do.
Never report a blended figure. Split every metric by channel, report the median and the 90th percentile rather than the average, and know which benchmark set you are comparing yourself against, because the two main ones measure different work and differ by more than ten points.
Why a blended number hides the thing you need to see
Chat and email are not two delivery routes for the same work. They attract different problems. Chat gets the question with a known answer, because the customer expects a reply in seconds and will not open a chat window to describe something complicated. Email and case work gets the problem that needs somebody to think.
So they score differently, and the gap is not small.
| Conversational support | Top tier | Median | Bottom tier |
|---|---|---|---|
| First response time | 10 seconds | 41 seconds | 4m 42s |
| Resolution time | 2m 7s | 10m 2s | 41m 39s |
| CSAT | 99.58% | 96.22% | 88.40% |
| First contact resolution | 97.83% | 83.32% | 63.54% |
Those CSAT tiers are the full conversational matrix across all company sizes. Narrowed to the 250 to 5,000 employee band, which is the comparable group for most readers here, the conversational median is 95.15% against 88.00% on ticketing. Both are from the same dataset and the difference is the segment, which is exactly the kind of thing worth checking before quoting any benchmark.
On ticketing, the same dataset shows the bottom tier taking up to 36 hours to resolve, where the top tier manages it in 32 minutes. The conversational spread between bottom and top is 41m 39s against 2m 7s, a twentyfold difference on work that is supposed to be comparable.
What to do about it. Pull every metric you report by channel, for one month, and put the two columns side by side. If you have never done this, the first time is uncomfortable and it is the most useful hour you will spend on reporting this year.
The two benchmark sets disagree, and both are right
This trips up everyone who tries to benchmark themselves, so it is worth stating plainly.
Freshworks, reading 32,000 companies and 1.2 billion tickets, puts median first contact resolution at 83.32% on conversational channels. SQM Group, reading 500+ North American contact centers, puts the industry average for first call resolution at 70%, and says only 5% of centers clear 80%.
Thirteen points apart. Neither is wrong.
A ticket closed on first touch and a customer confirming their problem went away are different events, and the second is a higher bar. So before you compare yourself to any benchmark, find out whether it counts tickets or asks customers. If your own number is survey-based, SQM is your comparison. If it comes out of your helpdesk, it is Freshworks.
Your benchmark depends on what your tickets are about
Even once you have split by channel and picked the right source, there is a third correction almost nobody makes. First contact resolution varies enormously by what the ticket is about, and the spread is wider than the spread between good and bad companies.
| Contact type | Median first contact resolution |
|---|---|
| General inquiries | 73% |
| Account maintenance | 72% |
| Orders | 71% |
| Billing | 69% |
| Claims | 61% |
| Technical support | 60% |
| Complaints | 48% |
Twenty-five points separate a general inquiry from a complaint. A technical support team at 62% is performing above its own benchmark. The same number on a general inquiry queue is eleven points below. Reported side by side with no context, the first looks worse and is better.
There is a related finding in the same dataset worth knowing before you build any escalation process: calls where the agent had to get help mid-conversation resolve 16% worse on first contact than calls they handled alone. Assisted is not free. It is better than a transfer and it is not the same as knowing the answer.
Enter what you have. Nothing is sent anywhere and nothing is stored.
Seven numbers that carry information
Four that mislead
Blended CSAT. Covered above. It is the single most reported and least informative number in support.
Average handle time on its own. It can fall because agents got better or because they started closing tickets early. Without reopen rate beside it, it is unreadable in either direction.
Deflection rate. A deflected contact is not a solved problem, it is a customer who stopped trying in your channel. Gartner found only 14% of issues are fully resolved in self-service, against a far higher share that start there. And only 27% of customers would try a chatbot again after one bad experience, so a deflection today can cost you the channel tomorrow.
Ticket volume as a measure of team performance. It mostly measures your product, your pricing page and your onboarding. Treating it as a support metric puts the incentive in the wrong place.
If a metric cannot be made worse by somebody gaming it, it is probably not measuring anything that matters. The useful ones are all slightly uncomfortable.
What the national picture says about all of this
One thing worth holding in mind while you tune your reporting. The American Customer Satisfaction Index, built on roughly 200,000 interviews a year across 400+ companies, currently sits at 76.7. It sat at 76.7 in 2013.
Thirteen years, enormous investment in support tooling, and the national number has not moved. That is not an argument against measuring. It is an argument for measuring the things that would actually change if the work got better, which is a much shorter list than most dashboards contain.
Frequently asked questions
What is a good first contact resolution rate?
It depends entirely on how it is measured. On survey-based measurement in contact centers, the industry average is about 70% and only 5% of centers clear 80%. On helpdesk telemetry across conversational channels, the median is 83.32%. Compare yourself to the one that matches your own method, and never quote a figure without saying which it is.
Should I report the average or the median?
The median, with the 90th percentile next to it. Averages on support timings are dragged around by a small number of very long cases, so the average often describes no actual ticket. The median tells you the typical experience and the 90th percentile tells you the bad one, and you need both.
How do I measure escalation rate if engineering work is not tracked in the helpdesk?
Count tickets that were reassigned out of the support queue, or that reference an engineering ticket number, over total tickets. It will be approximate for the first month or two. Approximate and tracked beats exact and absent, and the trend matters more than the level.
How many metrics should a support team actually report?
Five or six, split by channel, with a stated definition for each. Longer dashboards are read less carefully, and a metric nobody reads is worse than no metric because it creates the impression of oversight.
Does an outsourced team change which numbers matter?
It adds two. Agent tenure on your account, because a vendor’s company-wide attrition figure tells you nothing about your team, and the percentage of tickets handed back to you unresolved. Everything else on the list applies identically whether the team is yours or not.
Where to start
Pick one month of data. Split first contact resolution and CSAT by channel. Add the 90th percentile to every timing you already report. Then count how often support escalated to an engineer, which you almost certainly are not tracking yet.
Four numbers, one afternoon. If the channel split surprises you, that is the finding, and it was sitting in your helpdesk the whole time.
Our benchmark report carries all of them with sample sizes and data years attached, and ends with a scorecard you can grade your own team against.
Figures: Freshworks Customer Service Benchmark Report 2025, 32,000+ companies and 1.2 billion tickets, 2024 data. SQM Group, 500+ North American contact centers, minimum 400 surveys each, 2024. Gartner self-service survey, 5,728 customers, 2024, and chatbot survey, 3,566 customers, fielded February to March 2026. ACSI National Index Q1 2026, approximately 200,000 annual interviews. All are listed with links in The State of Customer Support Operations 2026.
