Enter counts for each score or paste a list. The satisfied cutoff is a setting, and the score, the interval and the sample size needed update as the numbers change.
- CSAT
- –
- Respondents
- –
- Average score
- –
- 95% interval
- –
- Respondents needed
- –
- Second survey CSAT
- –
- Change in points
- –
- Larger than chance at 95%
- –
How is CSAT worked out?
A respondent answers one question on a fixed scale, how satisfied they were with a purchase, a support reply or a release. CSAT counts everyone at or above the satisfied cutoff and divides by everyone who answered.
CSAT = satisfied respondents / all respondents × 100
Eighty people answer a post purchase survey on a 1 to 5 scale and fifty two pick 4 or 5. CSAT is 52 divided by 80, times 100, which is 65%.
Which scores count as satisfied?
The cutoff defaults to 4 on a 1 to 5 scale, 6 on a 1 to 7 scale and 8 on a 1 to 10 scale, and a team can move it. Once a team picks a cutoff it has to keep using that same cutoff on every later survey, because changing it changes the score without anything about the customers changing.
Take those same eighty responses, spread as 20 fives, 32 fours, 18 threes, 6 twos and 4 ones. Counting 4 and 5 as satisfied gives 52 respondents and a CSAT of 65%. Counting only 5 as satisfied gives 20 respondents and a CSAT of 25%. Nothing about a single customer changed between those two numbers, only the cutoff.
What does the interval mean?
A CSAT is built from a sample, not from every customer, so the true figure sits somewhere in a range around the number shown. The calculator above uses the Wilson score interval, which stays inside 0 to 100 and holds up at small samples and at shares near 0 or 100, unlike a plain normal approximation.
With p the satisfied share, n the respondent count and z = 1.96:
centre = (p + z²/2n) / (1 + z²/n)
half width = z × √(p(1−p)/n + z²/4n²) / (1 + z²/n)
Forty replies at a 65% CSAT carry a 95% interval of about 49.5% to 77.9%, a range of roughly 28 points on a score that looks like one firm number. The interval narrows as replies come in, but only with the square root of the count.
How many replies does a team need?
For a margin of plus or minus 5 or plus or minus 10 points, at a few shares:
| CSAT | ±10 points | ±5 points |
|---|---|---|
| 95% | 19 | 73 |
| 85% | 49 | 196 |
| 75% | 73 | 289 |
| 65% | 88 | 350 |
| 50% (most cautious) | 97 | 385 |
Halving the margin takes roughly four times the respondents, because the margin shrinks with the square root of the count. The calculator above recomputes the figure for whatever share and margin are currently set.
When is a change between two surveys real?
Two surveys each carry their own interval, so a score that moved also carries a combined uncertainty, wider than either survey's interval on its own.
With p1 and p2 the satisfied shares and n1 and n2 the respondent counts, the standard error of the change is √(p1(1−p1)/n1 + p2(1−p2)/n2) × 100, and the change is larger than chance at 95% when it exceeds 1.96 times that standard error. A first survey of 150 replies at a 65.3% CSAT followed by a second survey of 150 replies at 80% is a change of 14.7 points against a threshold of about 9.9 points, so that change is larger than chance. The second survey fields in the calculator run this same test on the counts entered.
How does CSAT differ from NPS?
CSAT asks about one interaction and reports the share who were satisfied with it. NPS asks one relationship question and subtracts detractors from promoters, so it tracks loyalty over time rather than a single moment.
What can CSAT not tell a team?
A CSAT says how many people were satisfied, not what the dissatisfied ones wanted instead, and the cutoff throws away everyone who answered in the middle without saying whether they sat close to satisfied or close to nothing at all.
A customer feedback rating is a temperature reading on its own. The specific requests sitting behind a falling score belong on a feature request board, where a customer writes the thing they want and other customers vote for it, so what comes out is a named, ranked list rather than another score.