How to Measure Customer Satisfaction in B2B SaaS

Two people can enter the same answers into a CSAT calculator and get different scores.

We tried it. Two neutral answers and three very satisfied answers gave us 60 in SurveyMonkey's public calculator. In SmartSurvey's, we got 80.00. You need to know how each tool works before you use its score.

If you're deciding how to measure customer satisfaction, start with the task you want to improve. Ask a question about that task, then work out a score from all valid answers. Use the same question, scale and math each time. Read the comments and give someone the job of acting on them.

1. Choose the experience and the question

Start with a choice your team can act on. “Did onboarding help admins set up their workspace?” gives you a place to look. “Are customers happy?” leaves much more to guess.

Customer satisfaction is how people feel about a product, service or task. In B2B SaaS, an admin may struggle with setup while a daily user likes the app. The person who pays may have a third view. Show those roles in your report.

You could use CSAT to ask how happy someone was with onboarding. Use CES to ask how easy it was to link two apps. You can ask both after the same task and learn different things.

If your CES report uses a mean, add the valid scores and divide by the count of valid answers. Check the scale's two ends. Does a high score mean easy or hard? Qualtrics' CES guide covers asking after a task and then asking why the person chose that score.

Our CSAT vs NPS guide can help you choose the question and scale. Once you have those, set up a way to collect, check and act on the answers.

2. Collect feedback from everyone who had that experience

Choose who gets the survey before you see any scores. An onboarding survey might go to everyone who reached the end of setup that month. Include people who struggled. If you ask only your best users, you leave those problems out.

Choose a channel that reaches those people:

• In-app: useful after a task, but it misses people who stopped opening the app.
• Email: can reach users, admins and budget owners outside the app. Use a sender they know. Say which task you're asking about.
• Post-support: name the issue they asked about. A closed ticket does not tell you if they think the issue is fixed.

Ask while the task is still fresh in their mind. Add a “What led you to that answer?” field that they can skip. Keep the invite brief. Check that other teams are not sending the same person more surveys at once.

Count invites and valid answers on their own. You do not know how someone feels if they did not reply. Keep blank answers out of the neutral group. Set a rule for repeat answers and forms left half done. Use that rule each time.

Decide whether each person or each account gets equal weight. A large account with lots of replies can drive the mean for all users. Show how many accounts and which roles replied, so your team can see whose views shaped the score.

3. Check what your CSAT calculation counts

For a five-point scale from very dissatisfied to very satisfied, you can use this formula:

Satisfied-response CSAT (%) =
(Satisfied + Very satisfied answers) / All valid answers × 100

Keep neutral answers in the total at the bottom of the formula. A mean rating uses different math, so label it as a mean. Our tests show why you need more than “CSAT” on a report.

How we tested the public calculators

On November 5, 2026, we opened SurveyMonkey's CSAT calculator and SmartSurvey's CSAT calculator. We put the same made-up counts in all five boxes. The boxes ran from very dissatisfied to very satisfied. We let the scores and shares update, then wrote down what each tool showed.

We moved two answers from neutral to satisfied, with the total still at five. Then we set all boxes to zero. A separate reviewer ran the same tests. We made up the counts to test the math. We did not collect answers from real customers.

In the first row, SurveyMonkey counted three positive answers out of five. SmartSurvey showed the same split: 40% neutral and 60% very satisfied. Yet its score was 80.00. The two tools agreed on which answers they had received. They used different math to turn those answers into a score.

In the second row, we moved both neutral answers to satisfied. All five answers now passed the positive cut-off, so SurveyMonkey reached 100. SmartSurvey rose by ten points to 90.00. Its score still gave different weight to satisfied and very satisfied. The first tool had put both answers in the same positive group.

We could match SmartSurvey's results by giving the five choices zero to four points. In the first row, two neutral answers would earn four points. Three very satisfied answers would earn twelve. That gives sixteen points out of twenty, or 80%. This is how we read the tested results. SmartSurvey's calculation docs also describe a mean based on points that start at zero. Its survey reports let you choose that method or a share of satisfied answers.

Switching from the first tool's math to the second would raise the first test's score by twenty points. No answers would change. Make that same switch in our second test and the score would drop by ten points. There is no fixed number to add or subtract to make these two scores match.

Before you compare a dashboard with a spreadsheet, put known counts through both. Check the mix of answers as well as the final score. If the mix is the same but the scores differ, check the math settings first. We tested these public tools on one date. Paid reports may have other settings.

4. Keep passives in NPS and record rounding

Bain's NPS definition takes the share of detractors away from the share of promoters. Work out both shares from all valid answers. Keep passives in that total.

NPS = 100 × (Promoters − Detractors) /
(Promoters + Passives + Detractors)

As part of the same November 5 test, we put three sets of made-up counts in SurveyMonkey's public NPS calculator. A separate reviewer ran them too. We wrote down the shares of promoters and detractors, plus the final score. Rounding those shares first can change the result.

In the first two tests, we kept the same promoter and detractor counts. We took out thirty passive answers. That cut the total from one hundred to seventy, and the score rose from 50 to 72. No one changed their answer. We changed which answers we counted. Leaving valid passives out of a real report would create that same false gain.

The third test let us check rounding. Using the raw counts gives 26.6667. Round that to a whole number at the end and you get 27. The tool showed 53% promoters minus 27% detractors, which gave 26. The page uses this method in its own worked example too. That one-point gap came from when the tool rounded. It did not tell us that customers felt worse.

For the seventy-answer test, the tool showed 72. Using the raw counts gives 71.4286 before rounding. So even with the same total answer count, scores can differ if you round at different steps. Keep the raw counts for each group. Someone else can then check how you got your score.

When you change a reporting tool, test both methods on the same set of answers first. If NPS moves by one point, check rounding and which answers you kept. A new product release may have nothing to do with it.

Treat an empty calculator as missing data

With no answers, SurveyMonkey's NPS and CSAT tools showed 0. SmartSurvey's CSAT tool showed 0.00. We had cleared all the counts, yet each tool still showed a score. No one had answered.

After those first tests, we ran two extra NPS checks with made-up counts. We wanted to see which real answer sets could also give a score of zero. A separate reviewer repeated both checks.

We entered fifty promoters, no passives and fifty detractors. With one hundred answers, the tool showed 50% for each of those two groups and an NPS of 0. Next we entered no promoters, one hundred passives and no detractors. The tool again showed an NPS of 0, this time with 0% promoters and 0% detractors.

Those two valid scores had the same number of answers behind them. The first split the answers between promoters and detractors. The second put every answer in the passive group. The same headline zero hid that difference. For the first report, read comments from both groups. For the second, ask what led people to choose a seven or eight.

The empty test had a third meaning: we had heard from no one. Its displayed zero shared a number with both valid results, but had no answers to inspect. Show the count and group shares beside NPS. When the count is zero, use “No responses” in the report rather than plotting a score.

5. Compare like-for-like results and investigate the comments

Use equal completed periods. Keep invite rules, channel, question, scale and math the same. If you change a rule, mark it on the chart. You may need to start a new trend from that point.

Show the answer count for each group. In a small group, one changed answer can move the score a lot. Check who replied this time. Did you hear from more admins, new users or people due to renew?

Read the comments beside the scores. Look for a problem your team can check, such as failed setup steps or people having to explain an issue twice. Keep what the customer said apart from your guess about what caused it.

Use correlation analysis to find which factors move with satisfaction only if you have enough answers to support it. An association does not prove cause. It also does not show that you can predict what one person will do.

Use other measures to investigate, with their limits intact:

• Churn: customers lost divided by customers at the start of the same period. People may leave due to price, a closed business or changed needs as well as poor service.
• CLV: an estimate of what an account is worth over time. Check the business model and costs behind it. A high value does not prove the customer is happy.
• Support: time to reply and time to fix an issue describe the service you gave. Ask the customer whether it helped.
• Usage: define what counts as active and how often users need the tool. Low daily use can be normal for an app that makes monthly reports.

Keep customer advocacy KPIs apart from survey scores. Count reviews and references that people completed. Saying they would recommend you does not count as taking that action.

6. Assign follow-up and keep publication permission separate

Give each recurring problem an owner, a next step and a date to check back. Reply to the person when an issue is still open. Say what your team will check. Return with what you found, even if you cannot make the change they want.

A customer may also offer a story worth sharing. Our customer testimonial guide covers how to ask and get their consent. A private survey answer alone does not give you the right to publish it.

If you use our older guides on turning NPS promoters into advocates or the best time to ask for reviews, check who gets the invite. For public reviews, do not use a high survey score to screen out people who might be critical. Check the review site's current rules before you send invites or offer rewards.

For a wider customer advocacy program, track consent and completed tasks through your customer advocacy metrics. The NPS to Review Conversion Calculator helps you plan. Replace its assumed rates with your own results before you report how well a campaign worked. Use the customer advocacy maturity quiz to choose which part of the program needs work next.

HighAdvocacy gives you hosted campaign pages and an in-app widget to collect reviews, social posts and testimonials. Your team can review what customers send. The Library keeps their words, source and consent details for reuse. Keep survey math in your survey or reporting tool. Use HighAdvocacy to collect customer stories through a campaign.

Copy this record into your next report before sharing its headline score:

Experience measured:
Exact question and answer choices:
Eligible people, account roles and invitation channel:
Reporting window and timezone:
Invitations / valid answers / incomplete or excluded answers:
Calculation, positive answers and rounding rule:
Score, response distribution and changes since last period:
Customer-reported reasons to investigate:
Action owner / next step / follow-up date: