NPS Segmentation, Revisited: How to Find Out Which Customers Are Unhappy (and Why)

Tsiki Naftaly

Oct 08, 2026

Cheerful dashboard showing a friendly NPS: fine score while one small red bar on a segment chart waves a tiny white flag

Back in 2020 I wrote A Guide to Segmented NPS for Consumer Products. To this day it's my most-read story on Medium, which tells you either how much people care about NPS or how little they care about my other posts. I choose to believe the first one.

That post was mostly about the why: an overall score only gives you a limited view, and connected products give you the data to see much more. This post is the how. You have NPS responses, you have device and app data, and you have maybe an hour on a Thursday afternoon. Where do you start?

A 30-second refresher

NPS comes from one question: "How likely are you to recommend us to a friend or colleague?" answered on a 0 to 10 scale. Promoters score 9 or 10, passives 7 or 8, detractors 0 to 6. Your score is the percentage of promoters minus the percentage of detractors (Bain & Company). Bain also notes you can track it for customer segments, not just the whole company. That second part is the whole game.

Why "fine" is the most dangerous NPS

An overall NPS is an average, and averages are very good at keeping secrets.

Say, for illustration (made-up numbers), your overall NPS is +32. Nobody panics. But when you split by firmware, most devices sit around +40 while the ones on version 4.2 are at -5. Those customers are not "fine." They're writing reviews. The healthy majority is just loud enough to drown them out.

That's NPS segmentation in one paragraph: stop asking "how are we doing?" and start asking "how are these people doing?"

Step 0: Put the score next to the data

Before you can cut anything, each NPS response needs to sit next to who answered it and what they're running. At minimum, join on a user ID:

The score (and the comment, please keep the comment)
Device model and firmware version
Phone OS and app version
Setup date
A few key feature usage flags
Sales channel and region

A spreadsheet is enough to start. Fancy can come later.

Step 1: Cut these segments first

For a connected product, here's the order I'd go in, roughly from "most likely to explain a sudden problem" to "most likely to explain a slow one":

Firmware version. The usual suspect when "it worked fine last month."
Phone OS. The classic Android vs. iOS gap. Go one level deeper to OS version if you can.
App version. Sometimes the device is fine and the app is the troublemaker.
Device model or hardware revision. Older models age differently, and new ones have their own surprises.
Time since setup. First 30 days vs. six months in. Low scores early usually point at onboarding. Low scores later point at reliability or "the novelty wore off."
Feature usage. People who use feature X vs. people who don't. This is where you find what actually makes people happy.
Sales channel. Online marketplace, retail, or direct. Different expectations, different return policies, sometimes different support paths.
Region. Language, local support hours, even typical home networks can differ.

One rule: one cut at a time. Resist the eight-way pivot table. It will look impressive and tell you nothing.

Step 2: Read the gaps between NPS detractors and promoters

Finding a difference is easy. Finding a real one takes a little discipline.

Look at the mix, not just the score. Two segments can share the same NPS with very different shapes. Lots of passives is a different problem than a pile of promoters canceled out by a pile of detractors.
Check how many responses you have. A segment with a handful of answers swinging wildly is a rumor, not a finding. Wait for more data or widen the time window.
Weigh the size of the segment. A disaster in a tiny slice of users and a mild dip in a huge slice deserve different urgency. Both deserve attention.
Watch for hidden overlaps. If firmware 4.2 only ships on your oldest model, is it the firmware or the hardware? Once a single cut shows a gap, then cross it with one more dimension.
Read the comments inside the segment. Comments are mildly interesting in bulk. Filtered to one unhappy segment, they practically write the bug report for you. This is where the "why" lives.
Line it up with your release dates. If the gap appeared the week a version shipped, you have a strong lead.

Illustrative bar chart of NPS by firmware version with one bar dipping below zero under a magnifying glass — Found you, 4.2. (Illustrative data only. No firmware was harmed in the making of this chart.)

Step 3: Do something about it

A finding nobody acts on is just a very nice chart. Here's what I'd do once a gap shows up.

Close the loop with the detractors in that segment. Not a generic "sorry you're unhappy" email. Something specific: "We know some devices on 4.2 have had trouble with X. Here's what we're doing, and here's what helps in the meantime." As I wrote in the original post, a 0 and a 6 are not the same customer, so don't treat them the same. The 0 might need a human. The 6 might just need the fix.

Fix it, then re-measure the same cut. After the fix ships, survey the same segment again and compare against the same split. Keep a simple changelog of releases next to your NPS timeline so the next person (probably future you) can see cause and effect.

Turn promoters into reviewers. If you're wondering how to improve your NPS score where it shows up in public (reviews, not just dashboards), start here. Your happiest segments (say, active users on the latest firmware who use your best feature) are the right people to ask for a review. Asking everyone at once means asking your unhappiest segment too, which is a bold strategy.

Spread what your promoters love. If one feature shows up again and again among promoters, help other users discover it. Just remember that a correlation is a hint, not a proof. Nudge, then measure whether the nudged group actually gets happier.

Hand-drawn four-step loop diagram where a small grumpy detractor turns into a smiling promoter — The NPS glow-up, in four easy laps.

The one-hour version

If you only have an hour this week:

Export your last 90 days of NPS responses with user IDs.
Join them with firmware version (or phone OS if that's easier).
Calculate NPS per group and the promoter, passive, and detractor mix.
Read the comments from the lowest-scoring group.
Write down one thing you'll fix and one group you'll contact.

That's it. No new tool required.

(If you'd rather not do the joining by hand, this is exactly why we built Segmented NPS at Copilot.cx. It brings together user, app, and device data so you can identify common characteristics among promoters and detractors. But a spreadsheet will absolutely get you started.)