NPS without lying to yourself

NPS is the most widely used loyalty metric in the world. On paper it is simple: one question, a 0–10 scale, a two-step formula. In practice there are dozens of ways to produce a flattering number that means nothing. Here is how to do it honestly.

What NPS actually is

The question reads: “How likely are you to recommend us to a friend or colleague?” Answers on the 0–10 scale split into three groups:

• Promoters – 9–10. They recommend you unprompted.
• Passives – 7–8. Satisfied, but not active.
• Detractors – 0–6. They may actively talk people out of it.

The formula:

NPS = % promoters − % detractors

The scale runs from −100 to +100. In B2B SaaS, +30 to +50 is normal. Apple once sat around +72. A typical bank lands near −10. Airlines are frequently negative.

Where companies fool themselves

1. They ask the wrong people

The classic: send NPS only to active paying customers and quietly skip everyone who churned last month. The number jumps 20 points and means nothing.

Do this instead: define the full population (everyone who used the product in the last 90 days) and take a random sample of it. Not “active users”, but all of them.

2. They ask at the wrong moment

Right after a successful action, a user gives you a 10. Right after an error, a 0. Average those and you have measured noise.

Do this instead: ask one to two weeks after the key event: onboarding, first purchase, a support conversation. Give the product time to behave normally and the user time to form an opinion.

3. They ignore the sample size

At 30 responses, an NPS of +50 and an NPS of +35 are statistically indistinguishable. The confidence interval is so wide that movement in the metric tells you nothing, and yet the report says “up 5 points!”

Do this instead: calculate the confidence interval. A rough rule of thumb: the margin of error is roughly ±10 / √(N/100). At N=100 that is about ±10 points; at N=1000, about ±3.

4. They never read the comments

An NPS number without an open “why?” question is close to useless. Everything interesting lives in the text: what detractors hate, what promoters praise.

Do this instead: always add a follow-up. In Askyo that is a conditional branch in the graph editor: detractors get “What could we do better?”, promoters get “What specifically won you over?”, passives get “What was missing that would have made you a promoter?”

How it works in Askyo

The NPS template already has:

1. A 0–10 scale (scale-numeric type).
2. Graph logic splitting into three branches by range.
3. A different open question on each branch.
4. A dashboard showing % promoters, % detractors, the absolute NPS, the confidence interval and a word cloud over the comments.

You do not collect the data in a spreadsheet and compute it by hand. The maths happens automatically, sample size included.

An honest-NPS checklist

• The sample is random, not “our best customers”
• Sample size ≥ 200 for significance (or you report the confidence interval)
• The survey triggers 1–2 weeks after the key event
• There is a follow-up asking why they gave that score
• Trends are read as a 90-day rolling average, not month-on-month
• Results are segmented by plan / channel / device instead of averaged into mush

What to use instead of NPS

NPS is not the only loyalty metric. Alternatives worth running:

• CSAT (“How satisfied are you?”) – immediate satisfaction. Good for support.
• CES (“How much effort did that take?”) – a measure of friction. Good for onboarding and complex flows.
• Sean Ellis PMF score (“How would you feel if you could no longer use the product?”) – a measure of product–market fit.

All four exist as ready templates in Askyo. Run them in parallel and you get a picture with depth instead of one flat number.

---

Want to try it? Build a survey in the editor, the NPS template is already there.