By the Wavefield Research team · Published Nov 8, 2026 · Updated Nov 10, 2026
A CSAT survey asks customers to rate their satisfaction with a specific experience, usually on a five-point scale from very dissatisfied to very satisfied. The CSAT score is the percentage choosing the top two points. It is sent right after the experience it measures: a purchase, a support ticket, an onboarding step.
Below: the standard question and its scale, the formula with a worked example, a question bank by touchpoint, where CSAT ends and NPS begins, and a calculator that prints the two numbers most CSAT dashboards skip: the margin of error, and the mean that the top-2-box collapse throws away.
The standard item is one line: “How satisfied were you with [the experience]?” answered on a labeled five-point scale from very dissatisfied to very satisfied. The brackets are where the craft lives: name the specific experience (this delivery, that support ticket), because satisfaction with something specific is answerable, and satisfaction in general is a mood.
Variants exist: seven points, 1 to 10, stars, smiley faces, a plain yes/no. They all produce a number, and they stop being comparable the moment you mix them. Pick one scale, label every point with words rather than numbering the endpoints, and keep it identical across touchpoints and waves. The case for fully labeled five-point scales, and what happens to data quality without the labels, is covered in the Likert scale guide.
The second question, when you ask one, is open: “What is the main reason for your score?” One rating plus one why is the whole instrument. A satisfaction questionnaire that runs ten items is a customer feedback survey wearing the wrong name, and it will earn a feedback survey's response rate instead of a CSAT's.
CSAT = satisfied responses (the 4s and 5s) divided by total responses, times 100. Worked example: 200 responses, of which 90 rated 5 and 50 rated 4. Satisfied responses are 140, so CSAT = 140 / 200 × 100 = 70%. Neutral and dissatisfied responses count in the denominator only; that is by design, and leaving them out inflates the score.
That top-2-box collapse is the score's convenience and its cost. When MeasuringU examined top-box scoring of rating scales, the verdict was that it “loses information about both precision and variability,” and that even rather large changes can be masked once a scale is reduced to two buckets. Concretely: a 70% built mostly from 5s and a 70% built mostly from 4s read identically on a dashboard while describing different customer bases, and a base sliding from delighted to merely satisfied moves the mean long before it moves the score. This is why the calculator above prints the mean next to the CSAT.
The score is also a percentage estimated from a sample, so it carries a margin of error like any other survey statistic: at 200 responses, a 70% CSAT runs about ±6 points at 95% confidence. Month-to-month wiggles inside the margin are noise, and the response counts needed for tighter margins are what the sample size calculator computes.
Treat the bank below as a CSAT survey template: pick the touchpoint, take its wording, and keep the five-point scale identical everywhere so the touchpoints compare. Each CSAT survey example here holds to the same rule, which is that it names the experience and asks nothing else. The CSAT questions you add beyond these should survive the test every transactional item faces: would a customer answer it thirty seconds after the experience, on a phone, without resenting you?
Within hours of the ticket closing
Separate the agent from the outcome; they move independently, and a kind agent delivering a bad policy deserves a different fix than a rude one delivering a good one.
At checkout, then again when the order arrives
Ask per stage. A single end-to-end rating blends the carrier's sins with the product's virtues, and you cannot fix a blend.
On a schedule, not a trigger
Overall product satisfaction is relational, so schedule it quarterly or twice a year, and benchmark it separately from support CSAT: the two have different drivers and different ceilings.
At the end of onboarding; ahead of the renewal window
The renewal-window reading is the one finance believes. Always pair it with the open-ended why; a renewal-risk score without a stated reason is an alarm with no address.
The CSAT vs NPS question has a short answer: CSAT measures satisfaction with a specific experience, just after it happens; NPS measures the strength of the whole relationship, on a schedule. One is a thermometer held to a moment, the other a quarterly checkup. Search for NPS vs CSAT and most results end with “use both,” which is right as far as it goes: use both for different jobs, and never average CSAT and NPS into one happiness number, because a strong relationship survives a bad delivery and the two scores are allowed to disagree.
CES, the customer effort score, is the third sibling: it asks how easy a task was, and after support interactions it predicts churn better than satisfaction does. The working division of labor: CSAT for transactions, CES for support flows, NPS for trajectory. The NPS mechanics, including the asymmetric cutoffs and the margin its dashboards skip, live in the NPS calculator; the program design across all three instruments is covered in customer feedback surveys, done properly.
Vendor benchmark tables disagree with each other by whole points, and the reasons are structural. Survey mode moves scores (an in-app prompt and an email sample different moods). Response bias moves them further: the delighted and the furious answer, the quiet middle does not. And the companies pooled into a table measure at different moments with different scales. A benchmark collected differently from your survey is a different instrument that happens to share a name.
The careful version of cross-industry comparison does exist. The American Customer Satisfaction Index has measured satisfaction across the U.S. economy since 1994 on a designed national sample, and the Journal of Marketing paper that established it (Fornell and colleagues, 1996) describes a market-based performance measure built from modeled expectations, perceived quality, and value, estimated econometrically. Reading it is a useful calibration: rigorous satisfaction measurement is a designed study, and an in-house CSAT is not that. It does not need to be. It needs to be consistent.
So the working answer: a good CSAT score is better than your last one, by more than the margin of error, measured the same way. Fix the scale, fix the trigger timing, and judge movements across two or three waves rather than reacting to any single reading. Teams that want the always-on stream version of this, and what it trades away against a designed study, should read customer satisfaction survey software before buying either.
Customer satisfaction. On a business dashboard the CSAT meaning is always the customer satisfaction score: the percentage of respondents choosing the top two points of a satisfaction scale, which is why you will also see it called top-2-box satisfaction. (For anyone asking what is CSAT in the other sense: South Korea's college entrance exam goes by the same acronym, and if that is the one you were looking for, we can only offer our sympathy.)
The percentage of survey respondents who chose 4 or 5 on the five-point satisfaction question. If 140 of 200 respondents picked the top two boxes, the CSAT score is 70. The number only carries meaning next to its base size and margin of error, which is what the calculator above reports, and it hides distribution shifts that the mean rating catches.
Most advice on how to improve CSAT aims at the survey instead of the product: begging for 5s, gating unhappy customers away from the form, or surveying only after wins. That raises the number and corrupts the measurement. The honest loop is slower: read the open-ended why answers, code the recurring complaints, fix the biggest one, and let the score catch up on its own schedule.
On a five-point scale, divide the count of 4 and 5 ratings by the total responses and multiply by 100: 140 satisfied responses out of 200 is a CSAT of 70%. The same top-2-box rule applies on other scales, which is exactly why mixing scales breaks comparability: the top two boxes of a seven-point scale are a stricter cut than the top two of a five-point one.
There is no defensible universal threshold. Scores in the 80s are commonly read as strong and sustained scores below the 60s as a warning in most consumer categories, but survey mode, timing, and response bias move scores by more than the gaps teams celebrate. The comparison that means something is your own trend, measured identically each wave and read against the margin of error.
Both, for different jobs. CSAT measures satisfaction with a specific experience right after it happens; NPS measures the overall relationship on a periodic schedule. Use CSAT (or CES) transactionally and NPS as the trended headline, and never average them into one happiness number: a strong relationship survives a bad delivery, and the two scores are allowed to disagree.
One rating plus one open-ended follow-up asking why; two ratings when you genuinely need to separate the agent from the outcome. Transactional surveys live or die on response rate, and the people who abandon longer surveys are not a random sample of your customers, so every added question both costs completes and biases the ones you keep.
Within hours of the experience it measures. Delayed surveys measure memory rather than experience, and response rates fall with every day of distance. The exception is product-level satisfaction, which is relational: schedule it periodically instead of triggering it, and keep the schedule fixed so waves compare.
Related: NPS calculator with margin of error · NPS survey questions · customer feedback surveys · customer satisfaction survey software · the Likert scale, in depth
Brief the agent and field a real CSAT study: weighted results, margins on honest bases, significance between waves. From $99 per study, no subscription.
Start a study