NPS Calculator — Net Promoter Score (-100 to +100)

The discarded middle is the whole story

Respondents rate 0 to 10. Scores of 9 and 10 are promoters, 7 and 8 are passives, and 0 through 6 are detractors. The score is ((promoters − detractors) ÷ total) × 100, giving a whole number between −100 and +100.

Note what that formula does: passives count toward the denominator but never toward the numerator. They are included in the total that percentages are computed from, and then discarded from the score itself. This single property explains most of NPS's strange behaviour, and it is worth understanding before you act on a number.

Incidentally, this is also why "our NPS is 40%" is a category error. It is not a percentage and not an average — it is a difference between two percentages, which is why it can be negative.

Identical scores describe opposite realities

Survey 100 customers and suppose every one of them picks 7 or 8. Promoters 0, detractors 0, so NPS is 0.

Now suppose 50 pick 10 and 50 pick 0, with nobody in between. Promoters 50%, detractors 50%, so NPS is 0 again.

The first describes total indifference — nobody dislikes you and nobody would recommend you. The second describes a genuinely polarising product that half your customers love and half actively dislike. These call for completely different responses, and the headline number cannot tell them apart. This is the clearest demonstration of why the score alone is not enough.

Which boundaries actually matter

With 100 respondents, each person is worth one point of NPS. What moves the score is crossing a band boundary, and only that:

The last row is the important one. Genuinely improving someone's experience from a 3 to a 6, or from a 7 to an 8, registers as nothing whatsoever. The scale rewards boundary crossings and is blind to everything else, which is why a real improvement in sentiment can show up as a flat score.

Small samples are mostly noise

Each respondent is worth 100 ÷ n points. At 100 responses that is 1 point per person; at 20 it is 5, so a single customer moving from promoter to detractor swings the score by 10 points.

Month-to-month movement on a small sample is therefore largely a random walk, and treating it as signal means reacting to nothing. Always report the response count beside the score — a score without an n is not interpretable.

Selection bias

The people who answer surveys are not a random sample of your customers. Users who have just contacted support are over-represented, as are the highly engaged. Churned customers mostly do not answer at all, which systematically removes your least satisfied users from the calculation — so the score tends to flatter you in precisely the situation where you would most want a warning.

What to do instead

Report the full distribution rather than only the headline number, since that is what separates the two zero-scoring cases above. Watch direction across several quarters rather than reacting month to month. And treat the free-text comments as the actual output — the score is a tracking number, while the reasons are the information you can act on.

The tool takes counts of promoters, passives and detractors, computes the score, and bands the result from Negative through to World-class. It also shows industry medians for reference — roughly 30 for SaaS, 25 for B2C software and 35 for retail. Treat those as medians rather than targets: benchmarks shift with sector and, more than people expect, with survey methodology, so a score is only meaningful in its own trend and distribution.

A score is a symptom rather than a diagnosis. Where a satisfaction figure moves, the revenue effect usually shows up in retention first, which the churn rate calculator measures directly.