NPS vs CSAT vs CES: which number answers your question
NPS measures the relationship and predicted advocacy. CSAT measures whether one specific experience was satisfactory. CES measures how much work the customer had to do. They are three different questions asked at three different moments, and they are not interchangeable.
The short version
If you want to know whether a support interaction went well, ask CSAT or CES right after it. If you want to track how people feel about you as a company over time, ask NPS on a schedule that has nothing to do with individual interactions. Sending NPS after a support ticket measures the ticket, then gets reported as if it measured the relationship.
All three are summary numbers. None of them tells you what to fix — that comes from the free-text follow-up you attach to the score, which is the part most teams skip and later wish they had.
Side by side
| Metric | What it actually measures | When it's the right one |
|---|---|---|
| NPS | Stated likelihood of recommending you — a proxy for the overall relationship | Quarterly or twice-yearly relationship tracking, same sample, same wording |
| CSAT | Satisfaction with one specific thing that just happened | Right after a purchase, delivery, onboarding step, or resolved ticket |
| CES | How much effort the customer had to spend to get something done | After a task that should have been easy: signup, checkout, a support request |
All three are ordinal scales. Their absolute values are only comparable to your own history, measured the same way.
How each one is calculated
Get the formula right before you compare anything to anything. All three are commonly reported wrong.
NPS — percent promoters minus percent detractors
Question: "How likely are you to recommend us to a friend or colleague?" on 0–10. Promoters answer 9–10, passives 7–8, detractors 0–6. NPS = %promoters − %detractors. Worked example: of 200 responses, 120 promoters (60%), 50 passives (25%), 30 detractors (15%) gives 60 − 15 = +45. Passives are excluded from both ends but still counted in the total, which is why a wall of sevens and eights produces a score near zero. The result runs from −100 to +100 and is written as a number, not a percentage.
CSAT — percent of responses that are satisfied
Question: "How satisfied were you with [the specific thing]?" usually on a 1–5 scale from very dissatisfied to very satisfied. CSAT = satisfied responses ÷ total responses × 100, where satisfied conventionally means the top two points, 4 and 5 ("top-two-box"). Worked example: 160 of 200 responses chose 4 or 5, so CSAT = 80%. Always publish the scale length and which points you counted — a CSAT calculated on the top point only is a different measure wearing the same name.
CES — the average of the effort scale
Question: either "How much effort did you personally have to put forth to handle your request?" or the agreement form, "[Company] made it easy for me to handle my issue," typically on a 1–7 scale. CES = the mean of all responses. Worked example: 200 responses summing to 1,080 gives a CES of 5.4 out of 7. Record which wording you used, because the two versions run in opposite directions: on the effort wording low is good, on the agreement wording high is good.
What the number cannot do for you
It cannot be compared to another company's number with any confidence. Scores move with industry, with how the question is delivered, with whether respondents believe they are talking to a person, and with who bothers to answer. A cross-company league table of NPS is mostly measuring survey design.
It cannot survive a small sample. Because NPS is a difference between two percentages, it is unstable at low response counts: with 20 responses, three people moving from 8 to 9 shifts the score by 15 points and nothing has actually changed. Look at the response count before you react to a movement, and prefer trends across several periods over any single reading.
It cannot tell you why. Everyone who answers a score question has a reason, and the reason is what you can act on. Add one open follow-up — "What is the main reason for your score?" — and read the detractor answers first. That single question tends to be worth more than the metric it is attached to.
It is also not neutral about who replies. People at the extremes are more motivated to answer than the contented middle, so any score built on voluntary responses skews toward strong feeling. That is tolerable if it is consistent, which is another argument for keeping wording, timing, and sampling identical across waves.
Running any of the three properly
- Pick the metric that matches the decision you will make, not the one your competitor publishes.
- Keep the question wording identical across periods, word for word.
- Keep the timing consistent — "48 hours after resolution" every time, not whenever someone remembers.
- Attach one open follow-up asking for the main reason behind the score.
- Record the response count next to every score you report.
- State the scale and the calculation in the report itself, so nobody re-derives it differently next quarter.
- Close the loop with detractors where you can. A survey that visibly changes nothing trains people to stop answering.
Building these in Zunoform
CSAT and CES are ordinary rating questions: pick a scale of 5 to 10 points, label the two ends, and you are done. Rating questions are on the free plan, which includes 500 responses a month, so a monthly CSAT program costs nothing to run.
NPS is one of the three question types that need a paid plan to publish, along with e-signature and payment. If you want the 0–10 promoter-and-detractor question specifically, that is the $14.99 Pro plan. If you only want the underlying data, a 0–10 rating question collects the same answers and you can bucket them yourself.
For the arithmetic, responses can be pushed to Google Sheets on the free plan. NPS in a spreadsheet is two COUNTIFs — one for scores of 9 and above, one for 6 and below — each divided by the total, then subtracted.
Templates to start from
Each one is editable, and none of them require a paid plan to publish.
A five-question satisfaction check after any customer interaction.
Post-appointment feedback on punctuality, the technician, and the fix.
One week in: setup ease, sticking points, and what the first week should explain better.
One reason question that branches into the right follow-up — features, price, or competitor.
Questions people ask
Is NPS a percentage?
No. It is the difference between two percentages, so it runs from −100 to +100 and is written as a plain number. Reporting "our NPS is 45%" is a category error that will confuse anyone who knows the metric.
What is a good NPS score?
There is no defensible universal answer. Scores vary so much by industry, survey channel, and sampling that cross-company comparisons are mostly noise. The comparison worth making is against your own previous score, using identical wording, timing, and audience.
Why don't passives count toward NPS?
By design. Scores of 7 and 8 are treated as satisfied but not committed, so they are excluded from both the promoter and detractor percentages while still counting in the total. That means a large group of passives pulls the score toward zero rather than lifting it.
How do I calculate NPS in a spreadsheet?
Count responses of 9 or 10 and divide by the total to get the promoter percentage, count responses of 0 to 6 and divide by the total to get the detractor percentage, then subtract the second from the first. Multiply by 100 if your percentages are stored as decimals.
If I can only run one, which should it be?
Run the one attached to a decision you have already agreed to make. CES after signup if you are working on activation, CSAT after support if you are working on support quality, NPS if you genuinely intend to track the relationship over years. A metric nobody has planned to act on is a chore with a chart.
Keep reading
Better forms.
Better data.
Build your first form in under 60 seconds. Free forever for personal use, no credit card required.