Blog
How to run an in-product NPS survey people actually answer
Timing, targeting, and honest scoring for in-product NPS: raise response rates without spamming, and read results without fooling yourself.
In-product NPS fails for predictable reasons: wrong moment, wrong audience, survey fatigue, and scores read without sample context. The fix is not a louder modal. It is disciplined timing, tight targeting, and honest presentation of results.
Timing beats volume
Ask when the user has experienced enough product to have an opinion, and when they are not mid-task. Good moments include after a successful first value action, after several active days, or at a calm account milestone. Bad moments include first paint of the app, during checkout, in the middle of a destructive workflow, or immediately after an error toast.
If you only ask people who just hit a success screen, you bias high. If you only ask after failures, you bias low. Pick a definition of "ready to answer" and stick to it long enough to compare months.
Targeting: who should see it
- Frequency caps: do not resurvey the same person every session. Space invitations in weeks or months, not hours.
- Role filters: an admin and an end user may need different questions; do not average them blindly if their jobs differ.
- Plan or segment: if enterprise and free users live different products, split reporting.
- Exclude recent respondents and people who already left a score for this campaign.
Targeting is how you respect attention. It is also how you keep the sample interpretable.
Design the ask so people finish
One primary question. Optional comment. Mobile-safe hit targets. No dark patterns that trap the dismiss control. A short reason for asking helps more than a long privacy essay in the modal. If you offer a follow-up for detractors, make it optional and short.
In-product widgets should feel native to the app. Hosted links and email still matter for users who rarely open the UI; use them as channels, not as an excuse to spam every address weekly.
Response rates: what "good" looks like without fake benchmarks
There is no universal honest percentage that proves your program works. Rates depend on product involvement, trust, and how often you ask. Track your own baseline: invitations shown, starts, completes, and comment rate. Improve by fixing timing and targeting before you redesign colors for the fifth time.
If almost nobody answers, you are either asking at painful moments or asking too often. If only superfans answer, widen the eligible set carefully and watch for selection bias in the narrative you tell leadership.
Honest scoring
NPS is promoters minus detractors over respondents who answered. Report the denominator. When the sample is thin, suppress the headline score or show a plus/minus margin instead of a naked number that flatters a quiet week. Do not average NPS with CSAT. Do not re-bin scales to manufacture improvement.
Read comments in the same sitting as the score. A stable NPS with rising "too expensive" or "too complex" themes is an early warning, not a reason to celebrate flatness.
Close the loop
Detractors who leave contact information deserve a path to a human. Promoters can be invited to referrals only if that matches your ethics and product. The survey is a listening system; if nobody acts on it, response rates will rot because users learn that answering does nothing.
Bottom line: ask after real product experience, cap frequency, keep the form short, track your own completion funnel, and present NPS with sample honesty. People answer surveys that respect their time and that appear to feed a real product process, not a vanity metric on a slide.