A CSAT calculator that runs both formulas on the same responses.

The two CSAT formulas, and why they disagree

Customer Satisfaction Score has two formulas in common use and they answer different questions. Top-2-box counts the share of respondents who picked either of the two highest options, usually 4 and 5 on a five-point scale: satisfied responses divided by total responses, times 100. Mean-based CSAT averages every rating instead, then divides that average by the scale maximum to express it as a percentage.

In a spreadsheet with raw ratings in column A, top-2-box is COUNTIF(A:A,">=4")/COUNT(A:A)*100 and mean-based is AVERAGE(A:A)/5*100. Both work unchanged in Excel and Google Sheets. On a seven-point scale, change the 4 to a 6 and the 5 to a 7.

The two move apart whenever opinion is split. A set of 100 responses that is half fives and half ones gives a top-2-box CSAT of 50% and a mean of 3.00, which rescales to 60%: ten points of disagreement from the same answers. Neither is wrong. Top-2-box measures enthusiasm, mean-based measures central tendency, and a report that names only one of them hides the fact that the other exists.

The gap does not always run the same way. Ratings that pile up on 4 push top-2-box above mean-based, because a 4 counts in full toward top-2-box and only 80% toward the mean. Ratings that split to both ends push mean-based above top-2-box. Pick the formula before you see the number, write it next to the score, and keep it there.

A worked CSAT calculation on four response shapes

Each row below is 100 responses to one of the four survey templates above, scored by both formulas. Counts are listed highest rating first, so 62, 21, 9, 4, 4 means 62 people picked 5, 21 picked 4, nine picked 3, four picked 2 and four picked 1. The gap column is mean-based minus top-2-box, in percentage points. Enter any row into the calculator and you get the same numbers back.

The spread is widest on the seven-point scale. Onboarding CSAT below is an ordinary, healthy distribution, and the two formulas report 58.0% and 79.1% for it: 21 points apart, decided entirely by which formula the team picked. Top-2-box on a seven-point scale means 6 and 7, so every 5 counts as a failure, while the mean treats a 5 as 71% of a perfect answer.

Which one to report follows from what the number is for. Top-2-box when you need a threshold to act on, because it answers how many people were actually happy. Mean-based when you are tracking movement on a stable scale, because it uses every rating instead of collapsing them into satisfied or not. On a two-point thumbs scale the choice disappears and mean-based becomes useless: with only 1 and 2 available it can never fall below 50%, however bad the week was.

What counts as a good CSAT score

There is no single good CSAT score, and the pages that publish bands like 90 to 100 exceptional, 70 to 89 good rarely say where the bands came from. Three sources are worth knowing because they are published, dated and attributable. Two of them are not measured the way this calculator measures, and that difference matters more than the numbers do.

The American Customer Satisfaction Index put the United States national score at 76.1 for the second quarter of 2026, down from 76.7 in the first quarter, in a release dated 11 August 2026. It is built from roughly 200,000 interviews a year across more than 400 companies in more than 40 industries. Its 2026 sector studies, each with its own fieldwork window, run from 80 for audio streaming and specialty retail, through 79 for online retail, 77 for wireless service and for hotels, 76 for airlines and 75 for social media, down to 73 for internet service providers.

ACSI is a 0 to 100 index, not a percentage of satisfied customers, and it is not produced by the arithmetic on this page. It comes from a multi-question model weighting expectations, perceived quality and perceived value, drawn from a national sample rather than from your customers at your moment. Reading a 76% top-2-box as level with the national 76.1 compares two different instruments. That confusion is the origin of most of the unsourced CSAT bands in circulation: they are index-shaped numbers wearing a percentage sign.

For a benchmark measured the way this calculator measures, Retently's 2026 CSAT benchmarks, published 8 April 2026, report a dominant range of 65% to 80% across industries, with financial services at 81%, ecommerce and retail at 77% and internet and software services at 64%. Retently publishes no sample size or methodology beside those figures, which is a reason to treat them as orientation rather than as a target. Zendesk's guidance, last updated 21 November 2025, calls 75% to 85% good and above 90% exemplary, and presents itself as a rule of thumb rather than as measured data.

Your own previous quarter beats all of them, provided it was collected on the same scale, at the same trigger, with a comparable number of responses. That is three conditions, and a benchmark comparison that fails any one of them is decoration.

How much of your CSAT is sampling noise

A CSAT of 80% from 50 responses and a CSAT of 80% from 2,000 responses are not the same finding, and no calculator on the first page of results says so. The 95% interval on a proportion is 1.96 times the square root of p times one minus p, divided by n. At p = 0.80 and n = 50 that is plus or minus 11.1 points, so the honest reading of that 80% is somewhere between 69% and 91%.

The interval shrinks with the square root of the sample, which means four times the responses to halve it. The same 80% is plus or minus 5.5 points at 200 responses and plus or minus 2.5 points at 1,000. It also widens as the score approaches 50%: at p = 0.50 and n = 50 the interval is plus or minus 13.9 points, not 11.1, so a mediocre score is measured less precisely than a good one.

The practical consequence is the one most CSAT dashboards get wrong. A move from 80% to 84% on 50 responses a month is not an improvement, and treating it as one is how a team spends a quarter chasing a number that never moved. The calculator prints this interval next to the top-2-box score at whatever sample size you entered.

Where the responses come from

Arithmetic is the easy half. The number is mostly decided before any calculator runs, by the moment you asked at, how many people answered, and whether the question was worded the same way last quarter. A CSAT that moved because the trigger changed from ticket closed to ticket closed and not reopened within 48 hours has measured the trigger, not satisfaction.

That collection half is what Sleekplan does. A survey fires on a path or a journey, checkout, settings, the end of onboarding, on a scale you set once, and cohort throttling stops the same person being asked twice in a quarter, which is the most common way a CSAT sample quietly stops being representative. An optional open question captures the reason next to the rating, and the dashboard reports rolling 30, 90 and 365-day scores instead of one survey round.

Because the responses land in the same workspace as the feedback board, a 2 out of 5 is one click from whatever that customer has been asking for, and a detractor can page a Slack channel within seconds rather than surfacing in a quarterly review. That also removes the sample size problem above: a trigger that fires on every resolved ticket collects hundreds of responses a quarter without anyone launching a campaign.