Customer satisfaction survey with CSAT and NPS

Why do satisfaction surveys so often tell you nothing?

You send customers a survey, a few of them answer, and you end up with an average of 4.2. Is that good? What should you change on Monday? Most surveys fail at this point, not because people did not answer, but because the questions were not built to lead anywhere.

Part of the confusion is the two numbers everyone mentions. CSAT and NPS sound like rival brands of the same thing, and guides mix them up. Yotpo's article comparing reviews with NPS, last updated on 9 August 2023, describes the NPS scale as one through 10. Bain & Company's own Net Promoter System site describes a zero to ten scale. That gap matters, because 0 is a real answer, and it counts as a detractor.

What is the difference between CSAT and NPS?

CSAT asks how satisfied someone is with one thing, usually on a 1 to 5 scale, and the usual convention is to report the share who chose 4 or 5. Nobody owns it, so check how any CSAT figure you compare against was calculated. It works best right after a single moment: a delivery, a support chat, a first week with the product.

NPS asks how likely someone is to recommend you to a friend or colleague, from 0 to 10. Bain's page on measuring the score puts 9 and 10 as promoters, 7 and 8 as passives, and 0 to 6 as detractors. The score is the percentage of promoters minus the percentage of detractors, so it runs from −100 to 100. It is about the whole relationship, and it is most useful as a trend measured the same way every few months.

Do not expect NPS to predict your growth on its own. Timothy Keiningham and colleagues tested that claim in the Journal of Marketing in 2007, using 21 companies and more than 15,500 interviews from the Norwegian Customer Satisfaction Barometer, and did not find NPS clearly better than other satisfaction and loyalty measures. Use it as a signal, not a forecast.

What happened when we scored the same answers both ways?

We built the form further down, then sent it 20 answers from made-up customers through a real browser to Formgong's own server code, run locally. The people, their answers and their example.com addresses were invented for the test, and we chose the answers to look like a small shop with mostly happy buyers and a few problems.

The same 20 people produced a CSAT of 75% and an NPS of −11, and both numbers are correct. We picked the answers, so the numbers themselves prove nothing about real shops. What they show is how far apart the two scores can sit: 9 of the 14 satisfied people who answered both questions gave 8 or less on recommending, and a satisfied customer who scores 8 counts for nothing in NPS. If you reported only one of the two numbers, you would tell a different story.

Two smaller results matter when you score your own answers. The person who skipped the second question sent no value for it at all, so each question needs its own count: here 20 for CSAT and 19 for NPS. And with 19 answers, one person moving from a 6 to a 9 shifts the NPS by about 11 points, so wait for more answers before reading much into a change.

Which questions should a customer satisfaction survey ask?

Three is enough for most small businesses. The satisfaction question names the thing you are measuring, such as “How satisfied are you with your order?”, because a vague “How satisfied are you with us?” mixes the product, the delivery and the price into one number. The recommend question uses Bain's wording so your score is comparable over time. The third asks for the main reason, and it is the one you will act on, since a number tells you something is wrong but only words tell you what.

Leave email optional and say why you want it, so people who want a reply can get one without everyone feeling watched. Then send the survey close to the moment you are asking about. A survey about delivery sent three weeks later measures memory, not delivery. If you want more structure than this, our customer survey template adds where people found you and how often they buy, and our form design guide covers why fewer fields get more answers.

What does a working survey form look like?

This is the form we tested. It is plain HTML that posts to Formgong, so it works on any site; replace fk_your_access_key with your own key, or point the form at another backend. Each part is there for a reason. The satisfaction radios are required, so nobody sends a blank survey by accident, while the recommend question is optional, because a forced answer from someone who does not care is noise. The hidden botcheck field catches bots that fill in every input.

If you only want the recommend question, the NPS survey template is a two-step version of it. The customer feedback template swaps the scores for what worked and what to fix, and the survey form generator builds other question types.

<form id="survey" action="https://formgong.com/submit" method="POST">
  <input type="hidden" name="access_key" value="fk_your_access_key">
  <input type="hidden" name="_subject" value="Customer satisfaction survey">

  <fieldset>
    <legend>How satisfied are you with your order?</legend>
    <label><input type="radio" name="satisfaction" value="1" required> 1 Very unsatisfied</label>
    <label><input type="radio" name="satisfaction" value="2" required> 2 Unsatisfied</label>
    <label><input type="radio" name="satisfaction" value="3" required> 3 Neutral</label>
    <label><input type="radio" name="satisfaction" value="4" required> 4 Satisfied</label>
    <label><input type="radio" name="satisfaction" value="5" required> 5 Very satisfied</label>
  </fieldset>

  <fieldset>
    <legend>How likely are you to recommend us to a friend or colleague? (0 = not at all likely, 10 = extremely likely)</legend>
    <label><input type="radio" name="recommend" value="0"> 0</label>
    <label><input type="radio" name="recommend" value="1"> 1</label>
    <label><input type="radio" name="recommend" value="2"> 2</label>
    <label><input type="radio" name="recommend" value="3"> 3</label>
    <label><input type="radio" name="recommend" value="4"> 4</label>
    <label><input type="radio" name="recommend" value="5"> 5</label>
    <label><input type="radio" name="recommend" value="6"> 6</label>
    <label><input type="radio" name="recommend" value="7"> 7</label>
    <label><input type="radio" name="recommend" value="8"> 8</label>
    <label><input type="radio" name="recommend" value="9"> 9</label>
    <label><input type="radio" name="recommend" value="10"> 10</label>
  </fieldset>

  <label for="survey-reason">What is the main reason for your answers?</label>
  <textarea id="survey-reason" name="reason" rows="3" maxlength="2000"></textarea>

  <label for="survey-email">Your email, if we may reply (optional)</label>
  <input id="survey-email" name="email" type="email" autocomplete="email">

  <div aria-hidden="true" style="position:absolute;left:-9999px"><input name="botcheck" tabindex="-1" autocomplete="off"></div>
  <button type="submit">Send</button>
</form>

How do you score the answers?

Download the answers as a CSV from your form's submissions in the dashboard and open it in a spreadsheet. For CSAT, count the 4s and 5s and divide by the number of people who answered that question. For NPS, take the share of 9s and 10s, take away the share of 0 to 6, and use the number who answered that question as the base. Write the base next to every score, as in “CSAT 75% of 20”, so nobody compares 75% of 20 with 75% of 2,000.

Then read the reasons, sorted by score. The detractors' words often point at one or two fixable things, like slow support replies or packaging, and those are worth more than a decimal point on either score.

Can Formgong score the survey for you?

For the form on your website, no: Formgong stores and emails each answer, and you score them from the CSV as above. For online shops there is a second option. The review emails Formgong sends after an order can carry a short survey, including a 0 to 10 NPS question, and the results page shows the promoter, passive and detractor split using Bain's 9 to 10 and 0 to 6 bands, with one response per order email so nobody answers twice. That last part is the real difference from a public form, where one unhappy person can send ten answers. Yotpo's page comparing reviews with NPS, read on 3 November 2026, offers its reviews to fill the gaps NPS leaves and describes no NPS survey of its own. The review emails docs describe the survey block.

Did your scores tell different stories too?

If you run CSAT and NPS side by side, we would like to know whether they agree for your business. Write to support@formgong.com, and tell us if anything in the form above did not work on your site.