You check out of a hotel in Denver and the emailed folio asks how satisfied you were with the stay. A support rep closes your ticket and the chat window asks how much effort it took to get there. Once a year your bank asks whether you would send a friend its way. Same subject, three different numbers, and people lump all of them together as customer satisfaction.
Measuring customer satisfaction nearly always comes down to one thing: asking, and asking the same way every single time. Three metrics have taken over that job. The NPS (Net Promoter Score) is about recommendations, the CSAT (Customer Satisfaction Score) is about how one experience went and the CES (Customer Effort Score) is about how much work one task took.
You measure customer satisfaction by asking the same short question after every important interaction, rolling the answers into a metric such as NPS, CSAT or CES and tracking that number month over month. At empirio.ai, an online survey tool made in Germany, sharing by link is free. By the end of this guide you will know which of the three belongs in your program.
📌 Key takeaways
- Every one of the three boils down to a single core question.
- NPS tracks loyalty, CSAT tracks satisfaction, CES tracks effort.
- CSAT and CES zoom in on single contacts, the NPS on the relationship.
- Your last score beats any benchmark you read in a report.
- Two metrics together tell you far more than one alone.
Create a survey for free
With empirio.ai you can create a modern online survey in minutes — free to start.
- AI-built survey
- Adjust by drag & drop
- Real-time analysis
How to measure customer satisfaction: the four routes
There are four ways to find out how satisfied your customers are, and exactly one of them hands you a number you can line up quarter after quarter. The other three still pull their weight, because they explain what is behind that number.
| Route | What you get | What stays hidden |
|---|---|---|
| Survey with a metric | a score that holds up over time | the reasoning behind the score |
| Open-ended question | reasons in the customer’s own words | a number you can trend |
| Complaints and online reviews | the loudest problems | the quietly unhappy |
| Behavior data such as renewals | what customers really do | the reason why |
The metric is the frame, the open-ended question fills it in. Complaints and reviews show up uninvited, so they never represent the whole customer base, only its two loudest edges. Behavior data is the most honest of the four and also the most silent, because a subscription that quietly lapses never states a reason.
All three metrics in this guide sit in the first route. Where they part ways is in what they ask, when they ask it and how the answer gets calculated.
What NPS, CSAT and CES have in common
On the surface the three questions have little in common. Underneath, they share a design: a short, fixed question that a large number of customers can answer almost without thinking, which yields a number you can watch over time. Five traits follow from that:
- One question each: every metric rests on a single question with a set scale.
- A “why” underneath: the open follow-up is what turns a score into an action item.
- No changes between rounds: same wording, same scale, same timing, or the trend means nothing.
- History beats benchmarks: last quarter’s score is a fairer comparison than another company’s.
- No built-in explanation: none of them tells you by itself why it went up or down.
Employees get their own spin on the NPS, too: the Employee Net Promoter Score asks staff whether they’d recommend their employer to people looking for a job.
Note
One more abbreviation shows up next to NPS, CSAT and CES: the Customer Satisfaction Index, or CSI. It is not a fixed formula but a weighted average. Several satisfaction questions get scored on their own and are then combined according to how much weight each topic carries for customers. If those weights were never measured, they are guesses, and a guessed weight is what usually sinks the index in practice.
NPS, CSAT and CES side by side
Question, scale, math and timing are where the three part ways. Two tables keep those levels apart: the first covers what you ask and when, the second what comes out of it.
Question and timing
The question sets the ceiling on what a metric can pick up at all, and the timing sets how sharp the answer is. Both get decided once and then left alone.
| Metric | Standard question | Best moment to ask |
|---|---|---|
| NPS | How likely are you to recommend us to a friend? | on a set schedule, for example quarterly |
| CSAT | Overall, how satisfied are you with our service? | right after a purchase, visit or delivery |
| CES | How easy was it to get your issue resolved? | right after a checkout, return or support ticket |
The NPS points forward and covers the entire relationship, while CSAT and CES look back at one specific event. That is what drives the cadence: a recommendation question after every single interaction would be pointless, and a satisfaction question once a year would be just as useless.
Math and range
Each calculation fits in two lines, and still the three results live on scales that cannot be held against one another. Every number stays inside its own range.
| Metric | How it is calculated | Possible results |
|---|---|---|
| NPS | % promoters (9 and 10) minus % detractors (0 through 6) | minus 100 up to plus 100 |
| CSAT | 4s and 5s divided by all responses, times 100 | 0 up to 100 percent |
| CES | total points divided by the number of responses | 1.0 up to 7.0 |
The scales belong to the math and are not open to taste: 0 through 10 for the NPS, 1 through 5 for the CSAT, 1 through 7 for the CES. Our free calculators do the arithmetic, so you can work out your NPS, work out your CSAT and work out your CES in seconds.
Strengths, blind spots and where they come from
Each metric is strong at its own job and weak at everything else. One at a time:
- The NPS delivers a single loyalty number and will not tell you where the problem sits. Fred Reichheld introduced it in the Harvard Business Review in December 2003 in The One Number You Need to Grow.
- The CSAT pinpoints the touchpoint that is letting you down and says little about loyalty. It has no single author, because the satisfaction question is older than the acronym.
- The CES surfaces friction in a process and ignores the product itself. It traces back to Stop Trying to Delight Your Customers by Dixon, Freeman and Toman in the Harvard Business Review of July 2010.
When to use NPS, CSAT or CES
You rarely have to choose just one. Each metric answers its own question and earns its keep at a different stage:
- Go with the NPS if you want a long-running read on how loyal your customer base is.
- Go with the CSAT if you need a verdict on one touchpoint, say a delivery, a consultation or an event.
- Go with the CES if a process needs fixing and you have to find the spot where customers stall.
Many teams run them as a set: CES and CSAT right after key interactions, the NPS every few months for the overall trend. If the NPS slips, the other two frequently reveal where the trouble began. Our guide to customer surveys shows how to plan a program like that.
B2B is a different game from B2C
In B2B you have fewer accounts but several contacts inside each one, and only one of them signs the renewal. The average matters less there than who actually responded: the procurement lead and the person in the tool every day often rate the same service very differently. B2C runs the other way, where a single response says almost nothing and only volume and trend do. An NPS from eight responses is worth a phone call in B2B; in B2C it is just a number.
Create a survey for free
With empirio.ai you can create a modern online survey in minutes — free to start.
- AI-built survey
- Adjust by drag & drop
- Real-time analysis
What a tool has to do for the measurement to hold
A tool for satisfaction measurement needs four abilities: build the scale exactly the way the metric defines it, repeat the identical survey as many times as you want, count responses group by group and make the free text readable. Everything past that is convenience.
The first one sounds obvious and is where most programs go wrong. A tool that can only build the NPS scale from 1 to 10 does not give you an NPS, it gives you a ten-point scale whose results line up with nothing anyone has published. So check before the first round that 0 to 10, 1 to 5 and 1 to 7 exist as separate question types.
Repeat the survey, then compare the waves
A satisfaction program only pays off on the second round. In practice that means building the survey once, reusing it for each wave and putting the results next to each other without touching a question. At empirio.ai the comparison mode starts with Lite, cross tabulations and filters start with Plus, and evaluating free text with AI also starts with Plus (as of October 2026). Those limits are worth saying out loud: the free plan gives you the results and the charts, not the comparison between two waves and not the AI read on open answers. The free plan with 50 participants per month covers a first round, and you can put an online survey together without paying before you commit to a metric.
When the second wave comes due and you want both rounds in one view, open a free account and keep them in the same project.
Privacy: the contract your vendor has to sign
A customer survey collects personal information as soon as a business stands behind it. Under the California Consumer Privacy Act as amended by the CPRA (Cal. Civ. Code § 1798.100 et seq.), a vendor that processes that information on your behalf is a service provider, and the statute defines one as a person that processes personal information on behalf of a business pursuant to a written contract. At empirio.ai the data processing agreement is included in the free plan, so the paperwork is ready before the first invitation goes out.
Common mistakes when measuring customer satisfaction
The four mistakes below share one trait: none of them shows up until the second round, and by then the first round is already wasted.
The wording quietly changes between rounds
A metric only means something next to its own history. Change the wording, the scale or the timing between two waves and you own two measurements instead of one trend. Write the question once, without rushing it, have two colleagues read it back, then lock it down.
Asking so often that people stop replying
Survey fatigue comes less from long questionnaires than from being asked again with nothing to show for last time. Set a cap on how often one person gets asked per year, and put one line in the invitation about what the last round changed. People who see a consequence have a reason to answer again.
Borrowed benchmarks instead of your own history
Industry numbers in published reports come from other samples, other scales and other moments. An NPS of 30 can be excellent in one market and a warning sign in the next. Compare against your own last score first, and against anything else only after that.
You measured, and then nothing changed
A metric that never turns into a task is just overhead. After each wave, take the reason that shows up most often in the open answers, turn it into exactly one change and check at the next round whether it moved the number.
Careful
The finer you slice the results, the faster a single answer points back at a person. Break a report down by location, plan type and age bracket and you can end up with two respondents in a cell, both of whom recognize themselves. Set a minimum number of responses per subgroup before you start reporting, and state that threshold in the invitation.
Conclusion
Picking between NPS, CSAT and CES is the smaller call. The bigger one is whether you ask the same question for years and whether every round ends in a change. Start with the metric that matches your most common customer interaction, and add a second only once the first has a clean run of history behind it.
A score with nothing to compare it to is a snapshot, not a measurement.
What to read next
- Responses already in and waiting to be analyzed? Evaluating a questionnaire
- Response rate too low for comfort? Tips for online surveys
- Running the same check inside your own team? Employee satisfaction surveys
Ready to ask your customers for an NPS, CSAT or CES?
empirio.ai gets your customer satisfaction survey live in a few minutes and shows the analysis as soon as the first responses land.