NPS Survey: Create, Calculate and Actually Learn Something

The Net Promoter Score (NPS) measures willingness to recommend with one standardized question: "How likely is it that you would recommend us to a friend or colleague?", answered on a scale from 0 to 10. The score is the percentage of promoters (9-10) minus the percentage of detractors (0-6) and lands between −100 and +100. Introduced by Fred Reichheld in the Harvard Business Review in 2003, it has become the most widely used loyalty metric in the world — and the most widely misread one.

This guide covers the mechanics — question, formula, timing, benchmarks — and the part most NPS programs skip: turning the number into decisions. If you are building a broader feedback program, start with the customer feedback guide; this article goes deep on one instrument.

NPS = % promoters − % detractors. Worked example: 200 responses, of which 90 promoters (45%), 70 passives (35%) and 40 detractors (20%). NPS = 45 − 20 = +25. Write it as a number, not a percentage. Two properties follow directly from the formula: the score throws away information (a 0 and a 6 count identically), and small samples make it jumpy — one respondent moves the score by 200/n points. The full small-sample math, with confidence intervals, is in our B2B customer satisfaction guide, because it bites hardest when you have 30 clients rather than 30,000.

The recommendation question is standardized on purpose — change the scale or the wording and you lose comparability with your own history and every benchmark. What you should adapt is the object: "recommend our support" measures something different from "recommend our company". Ask about exactly the thing you can act on. And keep the survey at two questions: the rating, then one open follow-up. Every additional question costs response rate — the pulse survey response rate guide documents how quickly grids kill participation.

The classic mistake is running one survey for both jobs. A customer whose parcel arrived damaged yesterday gives you a 3 that says nothing about the relationship — and averaged into a relational score it just adds noise. Post-purchase measurement is its own discipline; our guide to post-purchase customer feedback covers the transactional side in depth.

Short answer: above 0 is positive, above +30 good, above +50 very good, above +70 excellent — as a rough orientation. The honest longer answer: published benchmarks vary wildly by industry, country and survey method. B2B software routinely posts +40 while telecom and insurance sit near +10, and the same company measures higher via in-app prompt than via email. Treat any external benchmark as an anchor at best. The comparison that actually carries decisions is your own trend, measured identically, plus the split by segment: an overall +35 that hides a −10 among your highest-value accounts is not a good score, it is a warning with good lighting.

Here is the contrarian core of this guide: the score is the least valuable output of an NPS survey. The score tells you that something changed; only the open answers tell you what to do about it. Yet the standard implementation — a free-text box labeled "What is the main reason for your rating?" — is exactly where most programs leak insight, because the customers with the strongest opinions skip it, and a one-word answer like "support" is recorded as if it explained anything.

This is where a conversational survey structurally beats a static form. An AI-led interview reacts to the rating: a 4 with "support" gets asked what happened, how long the wait was, and what would have felt acceptable — one question at a time, like a good human interviewer. The result is a reason you can act on instead of a word cloud. That format exists as a ready-made AI interview that sends after every purchase, and the open answers come back summarized by theme with AI instead of as a spreadsheet of fragments.

An NPS program earns its keep in the 48 hours after a detractor answer. The sequence that works: a real person reaches out, references the specific answer, fixes what can be fixed, and reports back. Customers whose concrete problem was resolved and acknowledged routinely end up more loyal than customers who never had a problem — the service-recovery effect. And tell the wider base what changed because of their feedback: response rates climb when people see their answers move something. The same loop logic applies on the employee side, where the eNPS guide walks through it for internal surveys.