Enter the number of satisfied responses and the total. You get CSAT as a percentage, the margin of error, and the range the true figure plausibly sits in.
Satisfied normally means the top two points of a five point scale. Some teams use only the top point, which produces a much lower figure for the same customers. Whichever you choose has to stay fixed, or the trend becomes meaningless.
CSAT asks how satisfied someone was with a specific interaction: a delivery, a support ticket, a purchase. NPS asks whether they would recommend you at all. The first is about a moment; the second is about the relationship.
That difference explains why the two numbers rarely agree. In the illustration above, the same 412 customers produce a CSAT of 77 percent and an NPS of +21. Most of the satisfied customers are passives on the NPS question: happy with the order, not moved enough to recommend anyone.
Use CSAT to find broken touchpoints, because it is specific enough to point at one. Use NPS to track the overall relationship. The NPS calculator handles the other half.
CSAT is a single proportion, which makes it less noisy than NPS for the same sample, but not free of noise. The worked example, 318 satisfied out of 412, carries a margin of error of about 4 points.
| Responses | Margin of error near 77% |
|---|---|
| 50 | ±11.7 |
| 100 | ±8.2 |
| 250 | ±5.2 |
| 412 | ±4.1 |
| 750 | ±3.0 |
| 1,500 | ±2.1 |
For per-agent or per-channel CSAT the samples get small very quickly. An agent with 40 rated tickets a month has a margin of error around 13 points, which means ranking agents on monthly CSAT is ranking them largely on luck.
In the example, 412 of 2,580 surveyed customers responded, a 16 percent response rate. The margin of error assumes those 412 represent the whole group. They usually do not.
People who had a notably good or notably bad experience answer more often than people who had an ordinary one. That pulls the result toward the extremes and away from the middle, and no amount of extra responses fixes it, because the bias is in who answers, not how many.
Two practical checks help. Compare the order values, product mix or tenure of responders against non-responders; if they differ much, the score is describing a subgroup. And watch the response rate itself over time. A falling response rate with a steady score often means the unhappy middle has stopped bothering to reply.
Satisfied is not the same as loyal. Most satisfied customers never recommend anyone.
The first trap is changing the scale or the cut-off. A five point scale with top-two as satisfied and a seven point scale with top-three will produce different figures for identical customers. Fix the method before comparing periods.
The second is surveying only after resolved tickets. If the survey fires when a support ticket closes, customers whose problems were never resolved may never receive it. The score looks excellent and excludes exactly the people who would lower it.
The third is treating CSAT as a loyalty measure. Satisfied customers still leave, especially when a competitor is cheaper or more convenient. Satisfaction is necessary for retention, not sufficient for it. Check the link directly with the retention rate calculator.
The last is small-sample league tables. Ranking teams, products or agents on CSAT from a few dozen responses each rewards luck and punishes variance, and the people at the bottom of the table notice.
Divide the number of satisfied responses, usually the top two points on a five point scale, by the total number of responses, then multiply by 100.
CSAT measures satisfaction with a specific interaction. NPS measures willingness to recommend the business overall. The same customers usually score much higher on CSAT than NPS, because satisfied is not the same as loyal.
It depends on the scale, the cut-off and what is being surveyed. Compare against your own history using the same method, rather than against published figures that may have been measured differently.
Because the people who respond are rarely typical. Very happy and very unhappy customers reply more, so a small response rate can make the score describe a subgroup rather than your customers.
We use Google Analytics to count visits, only if you allow it. Nothing you type into a calculator is ever sent. Privacy policy