NPS analysis used to be a bar chart and a word cloud. Sopact reads every verbatim the moment it lands, attached to the same contact across waves.
NPS analysis used to mean two charts: a promoter and detractor bar, and a word cloud of comments. Both lose the verbatim that explained the score. Sopact reads every comment the moment it lands, attached to the same contact across waves — so the reason behind the number is read while there is still time to act.
By Unmesh Sheth · Founder & CEO, Sopact · Updated May 2026
NPS analysis is the work of reading every verbatim comment that arrives with a Net Promoter Score and connecting it to the same respondent over time. The 0 to 10 rating tells you what; the open-ended answer tells you why and how. Done well, NPS analysis ties the score and the comment to one persistent contact ID — so the reason behind the number is read while the team can still act on it.
A single number on a -100 to 100 scale, calculated as the percentage of promoters (9-10) minus the percentage of detractors (0-6). The arithmetic is the easy part. Two very different programs can produce the same score.
The open-ended follow-up answer to "What is the primary reason for your score?" The verbatim is where the failure mode lives, where the promoter says what they actually value, where the next quarter's churn is usually named in full sentences.
The score and the verbatim sit on the same persistent ID as every prior submission from the same customer. A 4 from a customer who scored 9 last quarter reads very differently from a one-off 4. The trend is the analysis; the static number is not.
Every comment classified, attached to the contact, and surfaced to the right owner the moment the response submits. Not at the end of the quarter when the analyst pulls the data. Not as a sentiment label that strips the wording. The original comment, read in context, while there is still time to act.
The most common failure mode in NPS programs is not a bad score — it is a clear comment that named the failure, sitting in the response set, never seen by the person who could have acted on it. The team meets every Monday to discuss the dashboard. The dashboard shows the trend line. The trend line averaged away the reason.
"Our NPS was steady at 32 for three quarters — we thought the program was working. Then churn ticked up and finance asked what happened. We went back and pulled the verbatims. The reason was written, verbatim, by twelve different customers across those three quarters. The same complaint, in twelve different ways, in the same column we never opened. The score was fine. The comments were screaming."
The failure mode is named by the customer, in the customer's words, the day the response arrives.
For twenty years NPS analysis meant two things: a percentage-promoters-minus-percentage-detractors arithmetic, and a quarterly sample of comments coded by hand. The first was a calculator. The second was a four-week project. Both were treated as the analysis.
Both are gone. The calculator never was the analysis. And the four-week project is now a few seconds — Claude, Power BI, Google's analytics stack will read ten thousand verbatims, classify each one against any rubric a team can write, and return a recommendation faster than a human can scroll the response sheet. The analysis got easy.
So the value moved. It is no longer in the survey interface or the dashboard. It is in the workflow that reads every comment the moment it lands, attaches it to the same contact across waves, and surfaces the failure with the context underneath it — so the team can act before the next wave confirms it.
The rating answers what moved. The verbatim answers why. The persistent contact ID answers since when. The team's codebook answers which failure mode. Run together, on one record, every wave — it stops being a chart and starts being a risk profile.
This is the same locked argument that anchors the qualitative and quantitative analysis pillar and the mixed methods research cluster — expressed here in customer-experience vocabulary. NPS is a small, focused mixed-methods instrument: a quantitative strand (the rating) and a qualitative strand (the verbatim), integrated at the respondent.
Five things happen the moment an NPS response submits. None of them happens in a traditional NPS dashboard, and none of them is optional if the analysis is going to do work the buyer can act on.
Not the same row in a spreadsheet — the same persistent contact ID that holds every prior submission from this customer. Their Q3 score, their Q3 comment, any attached document, any case-note from support — all on the record. The score never enters analysis without its context.
Not a generic positive/negative label. A versioned rubric the team wrote: the named failure modes, the known promoter triggers, the new themes flagged for human review. The same codebook applies to every wave, so a theme rate in Q4 means what it meant in Q1.
If the comment names a known failure mode, the detractor is routed to a named owner — with the original wording attached, the prior-quarter verbatim attached, and any other context already on the record. The owner sees the failure as the customer described it, not as a chart described it.
A 9 last quarter becoming a 4 this quarter is a different story than a 4 from a new customer. The system reads the same contact's trajectory, not just the aggregate average. The promoters drifting toward passive show up before the detractor count does.
When the board asks why churn risk is up, the answer is twelve verbatim comments from named contacts, ordered by date, with the prior-quarter comment from each of them sitting alongside. Not a chart. Not a sentiment percentage. The original words, traceable.
These five moves are the difference between an NPS dashboard and an NPS analysis. The dashboard tells you the number changed. The analysis tells you which customers, in which words, on which date, named the reason.
In 2026, a customer-experience lead can paste a quarter of NPS verbatims into any AI chat and get a fluent summary in seconds. The summary will sound right. It will be wrong in the ways that matter, because the analysis is no longer about reading the comments — it is about reading them with the context underneath.
A fluent summary without context is faster than reading the comments yourself. It is not analysis.
The model is the same. The record the model reads against is what changes the answer.
"Ask Claude to summarize the comments" and "ask Claude to read the comment alongside the same customer's prior comment, score history, and any attached case note" are not the same workflow. The first is a paragraph. The second is a workflow that flags risk while there is still time to act.
A single customer named "C-04812" in the system. Their score and their verbatim across five waves of an NPS program. The dashboard average never moved much. The trajectory on one record told the story two quarters early.
This is not a dashboard chart. It is one customer's record — the only unit at which NPS analysis can name a failure in time to act on it.
The "best NPS analysis tool" question hides three different categories underneath. They optimize for different parts of the workflow. The right tool depends on whether the team is shopping for the form, the chart, or the reading.
| What the team is shopping for | Survey platforms AskNicely, Delighted, Medallia |
Text analytics Thematic, Chattermill, MonkeyLearn |
Sopact |
|---|---|---|---|
| Distribute the survey | Their core strength. Multi-channel sends, response-rate tuning, branching. | Not the product. Assumes the responses already exist. | Capable, but not the reason to buy. |
| Display the score on a dashboard | Their core strength. Trend lines, segment splits, alerting on threshold. | Not the product. | Capable, but the dashboard is downstream of the reading. |
| Surface themes from the verbatims | A late-feature add-on. Word clouds and sentiment labels. | Their core strength. Theme taxonomies, trend tracking, language-model classification. | Done on arrival, against the team's own codebook, traceable to source. |
| Attach the verbatim to the same customer across waves | Limited. Survey-wave-centric data model. Often a fresh export per quarter. | Limited. The text-analytics layer reads comments, not relationships. | The core strength. Persistent contact ID across every wave, every artifact. |
| Read the verbatim against any attached document or transcript | Not designed for it. | Not designed for it. | Built for it. The NPS comment reads alongside the case note, the contract document, the call transcript — all on the same record. |
| Route a detractor with full context to a named owner | Alert by threshold. The owner sees the score. | Not the product. | The workflow. The owner sees the verbatim, the prior-quarter verbatim, and any attached document — on arrival. |
If the team needs the survey channel itself, a survey platform is the right tool. If the team needs theme classification across millions of comments, a text-analytics tool is the right tool. If the team needs the workflow that reads every comment on arrival, in the context of the same contact's prior comments and any attached document — that is where Sopact sits, and nothing on this table sits next to it.
NPS shows up wherever a team needs a quick read on whether a customer, participant, or stakeholder will come back. The dashboard is the same. The cost of not reading the verbatim is different in each context.
A relational NPS program with quarterly waves. The score is steady; the trend says the program is fine. The verbatim from the same five accounts across three quarters names a different story — usage friction, missing capability, lost champion. Reading on arrival means the CS lead sees the second mention, not the twelfth.
A cohort training program with an end-of-program NPS plus follow-ups at 30/90/180 days. The score moves with the cohort; the verbatim names the module, the facilitator, the part that did or did not transfer to the job. The same participant's six-month verbatim sits next to the day-one verbatim — the actual learning outcome reads in their language, not in the post-test number.
A scholarship or grant program with an NPS-style read at the end of each award cycle plus a follow-up six months in. The score signals satisfaction; the verbatim names what changed for the awardee and what is still missing. Each awardee's follow-up sits next to their original application — the program team can read the arc, not just the snapshot.
Your data, your codebook, your contacts. Sixty minutes. No demo accounts.
The same workflow — reading every response together with every other record on the same contact, every wave — goes by different names in different fields. NPS is one of them. The audience changes; the underlying claim does not.
An NPS program run with a persistent contact ID across quarters is a mixed-methods, longitudinal instrument. The vocabulary is different from the academic literature. The work is the same. See the linked clusters for the same argument in adjacent words.
NPS is one feedback instrument. A customer-experience program reads the verbatim today. The same workflow reads a training participant's six-month follow-up, a scholarship awardee's outcomes essay, a grantee's quarterly narrative, an applicant's recommendation letter — all on the same record, every time the customer returns to the page.
When the workflow that reads NPS verbatims is the same workflow that reads every other piece of stakeholder feedback the organization receives — case notes, applications, attached documents, transcripts, exit interviews — the data model stops being "NPS data" and starts being something more useful. It becomes one record per stakeholder, growing across every interaction, read on arrival every time.
That is the thing customers ask about after the NPS demo. Not "how do we read these comments faster?" — that part is solved. The next question is: "What else can we read this way?"
The answer lives at the engine pillar: stakeholder intelligence — the same persistent contact ID, the same read-on-arrival workflow, applied to every kind of stakeholder feedback an organization collects. NPS analysis is the first room. Stakeholder intelligence is the building.
NPS analysis is the work of reading every verbatim comment that arrives with a Net Promoter Score and connecting it to the same respondent over time. The score on its own is a single number on a -100 to 100 scale; the analysis is what tells you why the number moved, what failure the detractor is naming, and what the promoter actually values. Done well, NPS analysis ties the score and the comment to one persistent contact ID and to any prior interaction the same customer has had.
Both. The 0 to 10 rating is quantitative; the open-ended follow-up answer ("What is the primary reason for your score?") is qualitative. Treating NPS as only one or the other is the failure mode of most NPS programs. The rating tells you what; the verbatim tells you why and how. Modern NPS analysis reads both on one record.
Until recently the answer was hand-coding a sample of verbatims, tagging themes, and aggregating into a sentiment chart. AI changed that. A modern NPS analysis workflow reads every verbatim on arrival, classifies it against a versioned rubric the team controls, attaches the result to the same contact across quarters, and surfaces the failure mode while the customer is still on the line. The volume question is now solved; the question is whether the system reads with context.
NPS verbatim analysis is the close reading of the open-ended comment that follows the 0 to 10 score. The verbatim is where the reason lives: why this customer churned, why that promoter is a referral source, what changed since the last wave. Verbatim analysis fails when comments are summarized into a sentiment label and discarded; it succeeds when each comment is read in full, linked to the contact ID, and compared to the same contact's prior comments.
Sentiment analysis assigns a positive/negative/neutral label to a comment. NPS analysis reads the comment in the context of the score the same person gave, the comment they gave last quarter, and any other record attached to the same contact ID. Sentiment analysis is one possible label; NPS analysis is the workflow that decides what to do about the comment, traceable back to its source.
Three reasons. First, the score is an arithmetic compression: percentage of promoters minus percentage of detractors. Two very different programs can produce the same number. Second, the score smooths the trend: a stable 32 NPS over three quarters can hide a churning detractor base offset by a few new promoters. Third, the score does not name a failure: it tells you something moved, not what to fix. The verbatim names the failure; the score only counts it.
It means the open-ended comment is classified, attached to the contact, and surfaced to the right owner the moment the response submits — not at the end of the quarter when the analyst pulls the data, and not summarized into a chart that strips the original wording. Reading on arrival shortens the loop between the customer naming a failure and the team being able to act on it from months to minutes.
The analysis itself got easy. Claude, Power BI, Google's analytics stack can read every comment, classify it against a team's codebook, and produce a recommendation in seconds. That changes where the value lives. It is no longer in the survey interface or the dashboard. It is in the workflow that reads every verbatim on arrival, attaches it to the right contact, and surfaces the failure with the context underneath it — so the team can act before the next wave confirms it.
Yes, when the system uses a persistent contact ID rather than treating each survey wave as a fresh export. A persistent ID means the same customer's score and verbatim in Q4 sit next to their Q3 score and verbatim, on one record. A score that dropped from 9 to 4 between quarters reads very differently than a one-off 4 from a new contact — and the verbatim usually names the change.
The category is changing. Traditional NPS platforms (AskNicely, Delighted, Medallia) optimize for the survey: distribution, response rate, dashboards. Text analytics tools (Thematic, Chattermill, MonkeyLearn) optimize for the verbatim: themes, sentiment, taxonomies. Sopact reads on a different axis: every verbatim, every score, and any attached document live on one persistent contact ID and are read on arrival. The right tool depends on whether the team is shopping for the form, the chart, or the reading.
NPS is a small, focused mixed-methods instrument. The 0 to 10 rating is the quantitative strand; the verbatim is the qualitative strand; the integration is the workflow that reads them on one record. Run across waves with a persistent contact ID, it becomes longitudinal: the same customer's score and reason carried across quarters, so a drift is visible before it becomes a churn. See the mixed methods research and longitudinal design clusters for the broader methodology context.
The most common failure is the verbatim never gets read. The score is rolled up to a dashboard; the comments are exported to a word cloud or a sentiment label; the original wording is never seen by the person who could act on it. The customer described the failure clearly in their own words, and the analysis pipeline strips it out. Modern NPS analysis is built to prevent exactly this loss — read every comment, keep the wording, attach it to the contact.
The workflow this page describes goes by different names in different fields. Three adjacent reads, for the audience that prefers a different vocabulary.
The same argument in general analytical vocabulary — the two kinds of data, read on one record, in the AI era.
The academic-literature term for the same workflow — quant strand, qual strand, integrated at the respondent. Read this if your audience is a funder or an IRB.
What it means to follow the same contact across waves — the missing axis on most NPS programs, and the one that turns a chart into a risk profile.
Your data, your codebook, your contacts. Sixty minutes. We open the verbatims, attach them to the same customers across waves, and walk through what reading on arrival would have changed about the last quarter. No demo accounts. No slideware. Your own records, read live.
No slideware. No demo accounts. Your own records, read live.
