How to analyze NPS results across three layers, the score, segments, and the drivers behind them, so you can explain why the number moved.
NPS analysis is the full read of a Net Promoter survey: calculating the score from the percentage of promoters minus detractors, segmenting it by customer group and time, and — the part most teams skip — theming the open-ended comments into ranked drivers tied to each score. A complete analysis answers three questions: what the number is, who it varies by, and why. The first two are arithmetic; the third is reading.
The gap shows up in almost every NPS program: “we calculate the score every month and chart it, but when leadership asks why it dropped, we are guessing.” The score is easy to produce and impossible to act on alone, because the reason lives in comments nobody has read. Analysis that stops at the number is a thermometer, not a diagnosis.
Key takeaways
A Net Promoter Score is a single figure derived from subtracting the percentage of detractors from the percentage of promoters. It is genuinely useful as a headline: comparable over time, easy to communicate, hard to game. But a thermometer reading of 39 tells you the patient has a fever, not what is causing it. The cause is in the comments, and analysis that never reads them can report the fever forever without ever prescribing anything.
The reason the comments go unread is architectural. A form-centric NPS tool stores the score in one bucket and the comment as loose text, with no way to read the comment against a codebook or to keep it on the respondent across waves. Sopact calls the alternative the Verbatim Thread: every comment read on arrival, themed and cited, and kept on the same persistent contact record as the score. With that in place, “why did the score move” becomes a query over connected records rather than a manual reading marathon, the model behind Sopact’s feedback tools.
NPS analysis moved through three eras. First, a spreadsheet: the score calculated by hand, comments skimmed. Then the NPS dashboard — Delighted, AskNicely, SurveyMonkey and peers — which automated the score, the trend, the segment cuts, and a word cloud, making everything effortless except the reasons. The current era reads the verbatims into ranked drivers on arrival, tied to the score and the person, so the analysis explains itself.
The one test that separates the eras: ask a tool to explain a two-point drop in the score with the ranked drivers behind it, each cited to the comments, split by the segment where the drop happened. A dashboard can show the drop and a word cloud; it cannot show the ranked, cited, segmented reason. If the explanation is a word cloud, the analysis is describing the number, not diagnosing it.
An overall NPS is an average of averages, and averages hide. A score that holds steady at 40 can conceal a new-customer segment collapsing from 55 to 20 while a loyal segment climbs from 30 to 55. Reported as one number, that program looks stable; segmented, it is an emergency in one place and a success in another. Cutting the score by customer type, tenure, region, and channel is what turns a flat line into a map.
Segmentation matters even more for the comments than the scores, because the driver behind a low score in one segment is rarely the driver in another. New customers churn on onboarding; long-tenured customers churn on price or a missing feature. A themed read done per segment keeps those drivers apart, which is the same reason detractor analysis and verbatim analysis insist on reading segments separately.
To explain a score change, compare the driver mix between the two periods on the same segments: which themes rose among detractors, which fell among promoters, and which respondents moved between segments — then quote the comments behind the biggest shifts. A score change is a net effect of many individual movements, and the explanation is in those movements, not in the net figure. This requires keeping the same respondents identifiable across waves, which a form-centric tool cannot do.
The output is a short narrative a leader can trust: the score fell two points because a named onboarding driver rose among new-customer detractors, quoted, while promoter drivers held. Because Sopact reads comments on arrival and keeps them on the Verbatim Thread, that explanation is available the week the scores land, and every claim traces to the comments behind it — the standard how to analyze survey data applies to every open-ended question.
A complete NPS analysis works three layers — the score, the segments, and the drivers — and only the third tells you what to change. Most programs stop at the first two.
| Layer | What it answers | What it cannot do alone |
|---|---|---|
| Score | What the number is, and its trend | Say why it moved or what to fix |
| Segments | Who the number varies by | Name the reason inside a segment |
| Drivers (verbatim) | Why each segment scores as it does | Nothing missing — but only if comments are read |
The driver layer is where Sopact’s reading lives; the ongoing version of all three is continuous feedback.
An NPS number is a lagging summary; the reason a customer gave it is in the comment, and only if someone reads the comment while it still matters. That is the premise of the Loop, Sopact’s method for continuous intelligence: collect clean at the source, analyze the moment a response arrives, improve while there is still time to act. On a feedback program that means a detractor’s reason is themed the day it lands and routed to someone who can close the loop, not filed for a quarterly readout.
The Loop is also what makes an NPS finding defensible: every driver and theme traces to the exact comment it came from, the standard detailed in Loop traceability, so “price is the top detractor driver” is backed by the sentences, not a hunch.
One method, three moves that never stop
Then the cycle runs again, a little sharper each time. Read the method: the Loop methodology →
The fastest way to feel the three layers is to run your own data through them. Export scores, segments, and comments, then paste the prompts below into Sopact Sense’s Assistant, or reason through them with your team. The arrow above each links the Academy walkthrough with the expected output and tips.
Academy walkthrough → Analyze open-ended survey responses
Here are our NPS comments with each respondent's score: [ATTACH]. Theme every comment against our codebook, cite the sentence behind each theme, and show the top five themes among detractors, passives, and promoters separately.
Academy walkthrough → Analyze sentiment and its drivers
For these NPS verbatims: [ATTACH], score sentiment per comment, then show the specific drivers behind the detractor comments — the recurring reasons, each with a quoted example — so I know what to fix, not just the score.
Academy walkthrough → Connect quant and qual data
Here are our NPS scores and open-ended comments on the same respondent IDs: [ATTACH]. Show which themes explain the detractor scores, quote a comment for each, and list the specific respondents worth a personal follow-up.
Academy walkthrough → Analyze NPS across waves
Here are two waves of NPS from the same respondents on the same IDs: [ATTACH]. Track each respondent's movement between waves, show who moved from detractor to promoter and who slipped, and surface the comments that explain the biggest moves.
Each walkthrough is short and practical: what to do, the prompt to run, the output to expect, and the tips that keep it reliable.
Watch: reading feedback on arrival and tying every comment to the person who gave it.
Subtract the percentage of detractors (0–6) from the percentage of promoters (9–10); passives (7–8) count toward the base but not the score, giving a result between minus 100 and plus 100. Sopact keeps the calculation tied to the comments on the Verbatim Thread, so the number always has its reasons attached.
Work three layers: calculate and trend the score, segment it by customer group and time, and theme the open-ended comments into ranked drivers tied to each score. The first two are arithmetic; the third is the read that tells you what to change. Sopact runs the driver layer on arrival rather than leaving it for a quarterly marathon.
A score change is the net of many individual movements, so the explanation is in the driver mix: which themes rose among detractors, which fell among promoters, and who moved between segments. Sopact answers this as a query because it keeps comments and respondents connected across waves on the Verbatim Thread.
Because an overall score is an average that hides: a flat number can mask one segment collapsing while another rises. Cutting NPS by customer type, tenure, region, and channel turns a flat line into a map, and theming comments per segment keeps the drivers from being averaged into mush.
It depends entirely on industry and context, which is why comparing your score to a benchmark is less useful than comparing it to your own prior waves and reading the drivers behind the movement. Sopact focuses NPS analysis on the reasons you can act on rather than a single benchmark number.
You need every comment read, but not by hand. Sopact reads each comment against your codebook on arrival, themes it, and cites the sentence, so the ranked drivers are available immediately and nothing is skipped. Manual reading is what makes teams fall back on word clouds.
Verbatim analysis is the driver layer specifically — reading and ranking the comments. NPS analysis is the full read: the score, the segments, and the drivers together. Sopact treats them as one connected workflow on the Verbatim Thread rather than three disconnected exports.
Next: go deep on the driver layer in NPS verbatim analysis, or work the bottom of the scale in NPS detractors.
