Net Promoter Score
Date: 2026-08-16
One question, scored on a scale that throws away most of its own data. It’s near-universal, weakly predictive, and genuinely useful for one thing that isn’t the number — the free-text box underneath it.
What it is
Net Promoter Score (NPS) asks how likely someone is to recommend you, 0–10, then buckets and subtracts:
0–6 detractors
7–8 passives ← discarded entirely
9–10 promoters
NPS = %promoters − %detractors range −100 to +100
The problems, plainly
It discards information. A 7 and a 0 are both detractors; a 6 becoming a 7 doesn’t move the score at all. The bucketing throws away most of the signal in the responses you collected.
The same score means different things. 40% promoters and 0% detractors gives +40. So does 60% promoters and 20% detractors. Very different businesses, identical number.
It’s a stated intention, not a behaviour. People are poor at predicting their own future actions, and “would recommend” is not “did recommend” — Voice of Customer Data.
Response rates are low and biased. People with strong feelings answer. The middle — usually most of your customers — doesn’t, so you’re measuring the tails and calling it the population — Selection Bias, Survivorship Bias.
Predictive claims are contested. The strong version — that NPS predicts growth better than alternatives — has been challenged repeatedly. [CHECK: the academic position before citing any predictive claim.]
Why it survives
Not nothing:
- It’s comparable. Everyone uses it, so it travels across organisations and into board packs
- It’s one question, so response rates beat longer surveys
- It’s a trend line. Absolute value means little; movement over time on a consistent instrument means something
- It’s easy to explain, which is a real property for a metric that has to be understood by people who won’t read a definition
The part that’s actually useful
The follow-up question. “What’s the main reason for your score?”
The free text is worth more than the number. It’s unprompted, from real customers, at scale, and it identifies problems no dashboard will:
score reason
3 "delivery took nine days and I couldn't track it"
4 "the size guide was wrong, had to return"
6 "fine, just expensive compared to elsewhere"
Three actionable findings — logistics, product content, positioning — from three responses. Code the free text systematically and it becomes a standing input to prioritisation — Qualitative Coding.
Using it sensibly
- Track the trend, never the absolute against a benchmark. Scores vary by industry, country and question wording enough to make external comparison meaningless — Benchmarking
- Report the distribution, not just the net. The full 0–10 histogram carries what the bucketing discards
- Segment by cohort and by recent experience. NPS after a return is a different measurement from NPS after a smooth delivery, and the mix moves the blended score
- Never use it as a test metric. Response rates are too low and the measurement too noisy to detect anything a test could produce — Metric Sensitivity
- Close the loop. Contacting detractors who left contact details recovers some of them and is often worth more than the score
Alternatives worth knowing
- CSAT — satisfaction with a specific interaction. More actionable, less strategic
- System Usability Scale — for usability specifically, with better psychometric grounding
- Actual referral rate — measured behaviour rather than stated intention. Harder to instrument and immune to every objection above — Referral Programmes
In plain terms: treat NPS as a cheap, noisy trend line with an excellent free-text field attached. Don’t let it become a target, and don’t let anyone set an NPS goal without asking what behaviour would have to change to hit it.