A support team reports a CSAT of 92%. A product team reports an NPS of 40. An operations lead asks for the CES on the new self-serve flow. All three sound like they measure the same thing: how happy customers are. They do not. Each one asks a different question at a different moment, and mixing them up leads to comparisons that were never valid in the first place.
This guide separates the three questions, shows how to calculate CSAT with its real denominator, and gives you a tool to pick the right metric for a moment and run your own CSAT numbers.
Short answer: three metrics, three different questions
CSAT, Customer Satisfaction Score, asks how satisfied a customer was with one specific interaction, right after it happened. CES, Customer Effort Score, asks how easy it was to get a specific task done, such as resolving an issue or completing a setup step. NPS, Net Promoter Score, asks how likely a customer is to recommend the business overall, usually on a periodic schedule rather than tied to a single contact.
None of the three is a universal happiness score, and none of them replaces another. A business can run all three at once, as long as each is read as an answer to its own question.
Pick the question first: which metric fits which moment
The question to ask yourself is not "which metric is best" but "what moment am I measuring?" The three metrics map to three different moments.
CSAT: how satisfied was the customer with this interaction?
CSAT fits right after a single support conversation, chat, call or ticket closes. It is about that one interaction and nothing before or after it.
CES: how easy was it to get this done?
CES fits right after a customer finishes a task or a process, such as resolving an issue, completing an onboarding step or finishing a return. Scale conventions differ between teams and tools, so state the scale you used wherever you report a CES figure. Do not imply there is one universal CES scale, because there is not.
NPS: would the customer recommend you?
NPS fits a periodic check on the relationship as a whole, not tied to any single contact. It is typically asked on a schedule, such as quarterly, rather than after every interaction.
Moment to metric, at a glance
| Moment | Metric | What it measures |
|---|---|---|
| Right after one support interaction | CSAT | Satisfaction with that interaction |
| Right after a task or process | CES | Effort for that task |
| Periodic check on the relationship | NPS | Overall advocacy over time |
Use the tool below to pick your own moment and see the question and metric that fits it.
How to calculate CSAT with the right denominators
Once CSAT is the right metric for the moment, the calculation itself has a common trap: which number goes on the bottom of the fraction.
The scale convention this article uses
This guide's worked example uses a 1 to 5 scale, where a response of 4 or 5 counts as satisfied, sometimes called a top two box count. That is one common convention, not the only one. State your own scale and cutoff wherever you report a CSAT number, since the percentage alone does not say how satisfied was defined.
CSAT percent equals satisfied divided by responded
CSAT is the share of responses that counted as satisfied: satisfied responses divided by total responses, multiplied by 100. The denominator is the number of people who responded, not the number of people who were invited to respond.
Response rate percent equals responded divided by invited
Response rate is a separate number: total responses divided by surveys sent, multiplied by 100. The gap between surveys sent and responses received is the no-response count, which is simply surveys sent minus responses.
Worked example: 200 invited, 40 responded, 32 satisfied
The figures below are invented to show the arithmetic. They are not a benchmark and do not describe any real survey.
| Figure | Count | Calculation |
|---|---|---|
| Surveys sent (invited) | 200 | |
| Responses received | 40 | 40 / 200 = 20% response rate |
| Rated satisfied | 32 | 32 / 40 = 80% CSAT |
| No response | 160 | 200 minus 40 |
This month's CSAT is 80%, and its response rate is 20%, and 160 invitations got no response at all. All three numbers can be reported together without contradicting each other, because they answer three different questions.
Two mistakes the numbers invite
The worked example above makes two mistakes easy to spot once you see them written out:
32/200 is NOT CSAT. Dividing satisfied responses by everyone invited, rather than by everyone who responded, understates CSAT and quietly folds response rate into a number that is not meant to carry it.
160 unanswered is NOT 160 unsatisfied. An invitation that got no response tells you nothing about whether that customer was satisfied. Treating silence as dissatisfaction invents a result the survey never produced.
Run your own numbers
The calculator in the tool below takes your own surveys sent, responses and satisfied counts, and returns CSAT, response rate and the no-response count, with both of those warnings attached.
What the score does not tell you
There is no CSAT, CES or NPS number that counts as universally good or bad. What one business calls a strong result, another might read as a problem, depending on the scale used, the cutoff for "satisfied," and how the survey was distributed. This guide does not give a benchmark figure, because a benchmark without a matching definition and sample is not comparable to your own number.
A low response rate also limits how much a CSAT figure can tell you, since a small group of responders may not represent everyone who was invited. That is a reason to report response rate alongside CSAT, not a reason to adjust the CSAT calculation itself. Neither number carries any statistical confidence interval on its own; treat both as a description of the people who responded, not a guaranteed read on everyone invited.
Reading CSAT and NPS next to AI support metrics
If a business is running an AI agent for some portion of its support volume, CSAT, CES and NPS sit alongside other operational numbers, such as containment and response time, rather than replacing them. Our guide on what is next for AI voice agents in enterprises covers CSAT and NPS as part of a wider set of metrics to track against a pre-automation baseline. If you already run an AI agent for chat and want a clearer read on what its own numbers mean, see our guide to containment rate, which works through the same kind of denominator question for a different metric.
Pick your metric and run the numbers
Use the tool below to choose your moment and see the matching question, then, if CSAT is the right metric, enter your own surveys sent, responses and satisfied counts. If you are weighing AI agent options for the support volume behind these numbers, you can also read our guide to choosing AI customer service software, or try an AI agent free and keep measuring your own CSAT the same way you do today.