How to write NPS questions that actually predict growth

The Net Promoter Score is the most quoted metric in customer experience and the most casually misused. Teams paste “How likely are you to recommend us?” into a form, watch a number tick up or down, and treat it as a verdict. But NPS only predicts growth when the question is written carefully and the follow-up does real work. Here is how to get there.

Ask the standard question — and resist the urge to improve it

The canonical wording is: “How likely are you to recommend [product] to a friend or colleague?” on a 0–10 scale. It is tempting to reword it — “rate your experience”, “how satisfied are you” — but the recommendation framing is what gives NPS its predictive edge. Recommending something stakes a little of your own credibility, so a high score means more than mild satisfaction. Keep the 0–10 scale, too. Collapsing it to a 5-point scale throws away the resolution that separates a passive 7 from a promoter 9.

Anchor it to a real, recent experience

A score is only meaningful if you know what it is a score of. Trigger the survey close to a concrete moment — after onboarding, after a support interaction, after a few weeks of real use — rather than firing a generic “how are we doing?” into the void. Relationship NPS (overall sentiment) and transactional NPS (a specific interaction) answer different questions; decide which you are measuring before you send, because mixing them muddies the trend line.

The follow-up is the actual survey

The number tells you the temperature; the open-ended follow-up tells you the cause, and the cause is what you can act on. Always ask “What is the main reason for your score?” immediately after. The problem is that most people answer these in three words — “it’s fine”, “too slow”, “love it” — which is almost useless. This is where adaptive probing earns its keep: when an answer is thin, a good follow-up asks one specific clarifying question in the moment, turning “too slow” into “exports over 10,000 rows time out”. That is a sentence a product team can do something with.

Segment before you celebrate or panic

A single company-wide NPS hides more than it reveals. The same 30 can be a healthy 55 among power users dragged down by a miserable -10 among trial users who never activated. Cut the score by plan, tenure, and acquisition channel. Growth prediction lives in the segments: a rising promoter share among newly onboarded customers is a leading indicator; a falling score among your longest-tenured accounts is a warning that no headline average will show you.

• Cut by lifecycle stage — trial, newly onboarded, established, at-risk.
• Cut by plan tier to see where value is and isn’t landing.
• Track the promoter and detractor shares, not just the net, so you can see which end is moving.

Close the loop, then measure again

NPS predicts growth only when it changes behavior. Route detractors to someone who follows up, feed the verbatim themes into the roadmap, and re-survey the same cohort after you ship. The score becomes predictive precisely because you have turned it into a loop: measure, learn why, act, measure again. A number you collect and file away predicts nothing. A number attached to a reason and a follow-through predicts quite a lot.

The short version

Use the standard recommendation wording on a 0–10 scale, anchor it to a real moment, and treat the “why” follow-up as the main event — with probing so answers are specific enough to use. Then segment, act, and re-measure. Do that, and NPS stops being a vanity number and starts behaving like the growth signal it was meant to be.