guide
What Is a Good CSAT Score? Benchmarks by Industry (2026)
78 is average. Whether that is good depends entirely on what you sell.
The cross-industry average CSAT is 78 out of 100. Almost every page that tells you this stops there, which is the point at which the number becomes useless.
Seventy-eight is poor for financial services and strong for telecoms. Without a sector it is not a benchmark, it is trivia.
The formula, quickly
Satisfied responses divided by total responses, as a percentage. Satisfied usually means 4 or 5 on a five-point scale.
CSAT = (satisfied responses ÷ total responses) × 100
Everything interesting is in what counts as satisfied, who gets asked, and when — which is where the rest of this page goes.
Benchmarks by industry
| Sector | 2026 benchmark |
|---|---|
| Financial services and banking | 83 |
| Software and SaaS | 78 |
| Health insurance | 78 |
| Cross-industry average | 78 |
| Airlines | 72 |
| Cable and telecoms | 62 |
A twenty-one point spread between the top and bottom of that table. A telecoms operation scoring 75 is outperforming its sector by thirteen points; a bank scoring 75 is underperforming by eight.
So the first question is not what your score is. It is what your sector's is.
The general bands
| Score | Reading |
|---|---|
| 90%+ | Exceptional. Check your sampling before celebrating |
| 75-85% | Good, in most sectors |
| 65-75% | Acceptable in low-scoring sectors, weak elsewhere |
| Below 65% | A problem, or a measurement artefact |
That first row is not cynicism. A score above 90 is more often a sampling problem than an achievement, for reasons in the next section.
Three ways a score gets inflated without service improving
Each of these moves CSAT by several points and none of them involves helping a customer more.
1. Who gets asked
Surveying only resolved tickets removes the customers whose problem you failed to solve. It is the single most common inflation and it is usually accidental — the survey fires on ticket closure, and unresolved tickets do not close.
Check whether abandoned and escalated conversations are in your denominator. If they are not, your score describes your successes.
2. When you ask
Immediately after resolution scores higher than a week later, because relief is doing some of the work. Neither timing is wrong; comparing across a change in timing is.
3. What counts as satisfied
On a five-point scale, counting only 5 gives a very different number from counting 4 and 5. On a seven-point scale, counting 5, 6 and 7 gives a different number again.
None of these is dishonest and all of them are choices. A ten-point swing is available without anyone lying, which is why comparing your CSAT to a published benchmark is only meaningful if the methodology matches — and published benchmarks rarely state theirs.
Response rate is the number underneath
CSAT surveys typically get single-digit to low-double-digit response rates, and the people who respond are not a random sample. Very satisfied and very dissatisfied customers respond; the indifferent middle does not.
The practical consequence: at low response rates your CSAT is a measure of your extremes. Track the response rate next to the score, and treat a rise in score with a falling response rate as noise rather than progress.
Where CSAT is genuinely useful
- As a trend against itself, on stable methodology. Your score last quarter is a better comparison than any industry average.
- Split by handler. AI-handled against human-handled, tracked separately — this is the split that catches automation quietly degrading service.
- Split by ticket type. An aggregate hides that billing conversations score twenty points below order status.
- Alongside a volume metric. CSAT rising while contacts fall is good; CSAT rising while contacts fall because people gave up trying to reach you is not.
The split that matters most now
If any share of your contacts is handled by automation, the single most useful thing you can do with CSAT is stop reporting it blended.
A blended score is propped up by human-handled conversations. Deflection rising while AI-handled CSAT falls is the clearest early signal that you are suppressing contacts rather than resolving them — and the blended number will look stable throughout, which is exactly why teams miss it.
CSAT against the other two
| Metric | Asks | Best for |
|---|---|---|
| CSAT | Were you satisfied with this interaction? | Transactional quality, per contact |
| NPS | Would you recommend us? | Relationship health, not support |
| CES | How easy was it to get this resolved? | Effort, and the best predictor of loyalty |
NPS gets used as a support metric constantly and it is not one — it measures how someone feels about your company, which support influences and does not control. CES is the more useful companion to CSAT, because effort predicts churn better than satisfaction does.
Benchmarks read from published 2026 industry research, cited below. We have not run our own survey; the contribution here is placing the numbers against sector and naming the measurement choices that move them.
Frequently Asked
What is a good CSAT score?
Generally 75-85%, with above 90% considered exceptional. But the cross-industry average is 78 and the spread between sectors is twenty-one points, so the sector benchmark matters more than the general band.
What is the average CSAT score?
78 out of 100 across industries in 2026. Financial services lead at 83; cable and telecoms trail at 62.
What is a good CSAT score for SaaS?
The software and SaaS benchmark is 78, in line with the cross-industry average. Above 82 is genuinely strong for the sector.
Is 80% a good CSAT score?
In telecoms, airlines or cable, yes — well above sector benchmarks of 62 to 72. In financial services at 83, it is slightly below par.
How is CSAT calculated?
Satisfied responses divided by total responses, times 100. Satisfied usually means 4 or 5 on a five-point scale, though counting only 5s produces a very different number.
Why is my CSAT score so high?
Check who is being surveyed. Firing the survey on ticket closure excludes unresolved tickets, which removes your failures from the denominator. It is the most common inflation and it is usually accidental.
What response rate should a CSAT survey get?
Single digits to low double digits is typical. At those rates the score measures your extremes, because satisfied and dissatisfied customers respond and the indifferent middle does not.
Should I track CSAT separately for AI-handled conversations?
Yes, and it is the most useful thing you can do with the metric. A blended score is propped up by human-handled conversations, which hides automation quietly degrading service.
Is CSAT better than NPS for support?
Yes. NPS measures how someone feels about your company, which support influences but does not control. CES — how easy was it to get resolved — is the better companion, because effort predicts churn better than satisfaction.
Can I compare my CSAT to a published benchmark?
Only loosely. Scale length, what counts as satisfied, survey timing and who gets asked all move the number by several points, and published benchmarks rarely state their methodology.
Tools Mentioned
Full reviews, pricing tiers and where each one breaks.
Zendesk QA
ZendeskFormerly Klaus. Auto-scores every conversation and now grades AI agents alongside humans.
MaestroQA
Custom scorecards and screen capture, for teams that want QA defined their own way.
Observe.AI
Conversation intelligence and real-time assist, strongest on the analytics side.
EvaluAgent
One of the few QA vendors publishing per-seat pricing, plus per-conversation QA for AI agents.
You Can Also Look Into
AI Customer Service Metrics Explained - What to Track in 2026
Four metrics worth tracking, one worth ignoring, and how to tell a working agent from customers giving up.
What Is a Good Ticket Deflection Rate? Benchmarks for 2026
Published benchmarks range from 15% to over 80% for the same metric. Here is why they disagree, and the number a first-year deployment actually hits.
12 AI Customer Service Mistakes (and How to Fix Each One) in 2026
Best-practice lists are usually a vendor describing its own feature set. These are the failure modes teams actually report, and what prevents each one.
AI Resolution Rate Claims Explained - Why 93% and 30% Both Mean Nothing (2026)
Vendors publish autonomy rates between 30% and 93%. The range is that wide because they are not measuring the same thing.
SOURCES
WRITTEN BY AR · UPDATED 2026-08-04
I read the fine print. Vendor pricing pages, billing definitions, terms, funding filings and acquisition notices — then I do the arithmetic nobody publishes: what a platform actually costs at your volume, what its headline metric is really counting, and who owns it now. I do not run benchmarks, and no page here pretends otherwise.