To analyse NPS verbatim comments at scale: set up a consistent coding framework before reading, separate promoter and detractor themes (they rarely overlap), tag responses by customer segment and tenure, and run the analysis across the full dataset rather than reading top-to-bottom. Tools like Skimle can process 500+ NPS verbatims in minutes and surface themes by segment, including the detractor-specific patterns that manual reading typically misses.
The number at the top of your NPS dashboard tells you whether things got better or worse. The verbatim comments underneath tell you why, and what to actually do about it. Most teams spend 90% of their time on the score and treat the comments as an afterthought.
That is backwards.
According to benchmarks across survey platforms, around 10% of NPS respondents provide open-text verbatim comments after rating. In a B2B context with a response rate of 12–25%, that means a company with 5,000 accounts might receive 60–150 verbatim comments per survey wave: enough to be meaningful, but too many to read carefully and not enough to dismiss. At 500 responses, manual reading becomes a serious problem. At 2,000, it is impractical without tooling.
This guide covers how to extract real insight from NPS open text at any scale, from a first-time analyst trying to make sense of 80 comments to a CX team processing quarterly surveys with thousands of responses.
For context on related open-text analysis challenges, see our guide on how to analyse open-text responses at scale without losing your mind.
Why NPS scores alone mislead
An NPS of 42 tells you that 42 percentage points more of your customers are promoters than detractors. It does not tell you what promoters love, what is driving detractors away, or whether the gap between the two groups has a simple fix or a structural problem.
Two companies can have identical NPS scores for completely different reasons. One has enthusiastic promoters who recommend on every call, held back by a vocal detractor segment frustrated by pricing. The other has quiet, satisfied customers and a detractor group that churned within three months and rated 0 on the way out. The response is different in each case, but the score looks the same.
The verbatim comments contain the differentiation. Promoters tell you what is working ("the analysis speed is unlike anything we've used before", "the team we work with is exceptional"). Detractors tell you what is not ("integrations are too limited", "the price jumped 30% and nobody explained why"). Passives often surface latent needs that neither extreme expresses.
The challenge is that you cannot just read them. You need a system.
Why manual reading does not scale past a few dozen comments
Reading 30 NPS verbatims in a spreadsheet is perfectly manageable. Reading 300 starts to become unreliable. Reading 3,000 without a framework produces findings that reflect what the last 50 comments said, not what the full dataset says.
Manual reading has several structural weaknesses at scale:
Recency and salience bias. Readers remember the most recent and most emotionally salient comments. A strongly worded detractor comment sticks in memory even if it represents 2% of responses. A quiet majority of positive comments about onboarding fades because positive sentiment is harder to encode in memory.
Inconsistency across analysts. Two analysts reading the same 200 comments independently will apply different labels, group similar themes differently, and reach different priority rankings. Research on manual content analysis consistently finds that without a predefined coding framework, inter-rater reliability is low.
No segment visibility. Reading comments in a flat list (ordered by date or score) gives no visibility into whether enterprise customers say something different from SMB, or whether new customers have a different experience from long-tenured ones. Those differences are often the most actionable finding.
No frequency tracking. A manual read might note that "pricing" was mentioned but cannot easily tell you that pricing was mentioned by 34% of detractors, concentrated in accounts with contract values below €20k, and mostly related to a specific change made in Q1.
Manual reading remains useful for initial sense-making and for reading individual responses in context. It does not scale as the primary analysis method.
How to set up a coding framework for NPS verbatims
A coding framework for NPS verbatim analysis is a predefined set of categories that you apply consistently across responses. Setting it up before you start reading prevents the framework from being shaped by whichever responses you happened to read first.
Step 1: Define your categories from your known hypotheses. Before looking at the data, list the dimensions that likely drive NPS in your business: product features, pricing, support quality, onboarding experience, integration reliability, account management, etc. These become your starting categories. You can always add new ones inductively once you start reading, but starting with hypotheses prevents the framework from being an accident.
Step 2: Separate promoter, detractor, and passive coding. Promoter verbatims and detractor verbatims tend to talk about different things. Promoters focus on what they love; detractors focus on what is frustrating them enough to not recommend. Passives are the most interesting for growth: they are one good experience away from becoming promoters. Code each group separately to see the contrast.
Step 3: Add a sentiment flag within each category. A comment about "pricing" from a promoter ("the pricing is fair for the value") is different from a comment about "pricing" from a detractor ("the pricing went up without warning"). If your coding only captures the topic, you lose the sentiment dimension. Add positive/negative/neutral flags to each coded segment.
Step 4: Capture verbatim length and specificity. Short verbatims ("great product") carry less analytical weight than detailed ones ("the analysis view is excellent but the export to Excel drops the category hierarchy"). Weight your conclusions accordingly.
For teams that want a predefined category analysis workflow, Skimle lets you specify your NPS coding framework and apply it consistently across all uploaded responses, with every coded passage linked to the source verbatim.
How to find themes across segments
The most actionable NPS verbatim findings almost always come from segment analysis, not the aggregate picture.
What do enterprise customers say vs SMB? Enterprise customers often have different pain points from small accounts. They may find onboarding too self-service, want dedicated support, or need integrations that smaller customers never use. Conflating enterprise and SMB verbatims into a single "pricing is a concern" finding masks the fact that only accounts above €50k ACV mentioned pricing positively (as a sign of value), while all the pricing complaints came from accounts below €10k.
What do new customers say vs long-tenured ones? New customers (under 6 months) who are detractors are a warning sign: something is breaking in the onboarding or early experience. Long-tenured detractors who were once promoters are a churn signal: something changed in the product or relationship that eroded their satisfaction. These are different problems with different owners.
What does the shift look like over time? NPS verbatim analysis by wave (Q1 vs Q2 vs Q3) can surface the exact themes that emerged after a product change, a price increase, or a support team restructure. Without wave-level coding, you see that the score dropped but not why.
To do this analysis well, you need metadata. Every NPS response should carry fields for at minimum: customer segment (enterprise/mid-market/SMB), tenure band, geography, and contract value band. Without metadata, segment analysis is guesswork.
Skimle's metadata analysis lets you tag each uploaded document with these attributes and then slice findings by any combination of them. The discovering themes using metadata variables guide explains the approach in detail.
For a broader view of how to use NPS findings alongside other customer data, see our voice-of-customer research guide.
Manual reading vs Excel vs dedicated tool: a comparison
| Approach | Works at | Speed | Segment analysis | Traceability | Cost |
|---|---|---|---|---|---|
| Manual reading | Up to ~30 responses | Slow (1–2 days for 100) | None without pre-sorting | Only if notes taken | Free but analyst time-intensive |
| Excel pivot analysis | 30–300 responses | Medium (half day for 200) | Possible with pre-tagged data | Partial (original text in adjacent column) | Free but requires careful setup |
| Dedicated AI tool (e.g. Skimle) | 30–5,000+ responses | Fast (minutes for 500) | Full (slice by any metadata field) | Every theme links to source verbatims | Tool cost, but 10–50x analyst time saving |
Manual reading is the right default for fewer than 30 responses or when you need to read carefully for unusual context. Excel works when responses are pre-tagged with segment data and you need a frequency count. At 300+ responses, or whenever segment analysis matters, a dedicated tool pays for itself within one survey wave.
How to close the loop: turning NPS verbatim themes into actions
Analysis without action is just a report. The discipline of closing the loop, connecting NPS verbatim themes to specific changes, is what separates CX programmes that move the score from those that track it.
Map themes to owners. Every theme that comes out of your NPS verbatim analysis should have a named owner in the business. "Pricing concerns" belongs to Revenue or Pricing. "Onboarding too slow" belongs to Customer Success or Implementation. "Missing integration" belongs to Product. Without named owners, themes become talking points rather than commitments.
Distinguish structural from transactional issues. Some NPS verbatim themes are structural (a product gap, a pricing model that does not match value delivered) and require a roadmap decision. Others are transactional (a specific bad onboarding experience, a support ticket that took too long) and can be resolved at the account level. Structural themes go to the quarterly business review; transactional ones go to the account team this week.
Run the analysis by wave and report the delta. The most compelling presentation to leadership is not "here are this quarter's themes" but "here is how themes shifted from last quarter, and here is what changed in the business during that period." That narrative turns NPS verbatims from a retrospective metric into a forward-looking feedback system.
Report to detractors. Some CX programmes follow up individually with detractors, using the verbatim as the opening for a conversation. When you can say "I saw your comment about the integration reliability. We have made a change, here's what it is" the response is almost always positive regardless of whether the change fully solves the problem.
6 common mistakes in NPS verbatim analysis
Reading verbatims without coding. Reading creates familiarity but not structure. Without coding, you cannot tell whether a view is held by most respondents or expressed loudly by a minority of eloquent complainers.
Reporting top themes without checking detractor-specific themes. The top theme in aggregate NPS verbatims is often positive (because promoters write more and more enthusiastically). If you report top themes without separating promoters from detractors, your "most mentioned" finding will often be something you are doing well, not something that is driving churn.
Using the score to filter which verbatims you read. Reading only 0-3 (extreme detractor) verbatims and 9-10 (promoter) verbatims skips the passives, who often contain the most nuanced and actionable feedback.
No metadata on responses. Without customer segment data attached to each response, you cannot do segment analysis. This information needs to be appended to the survey data before analysis, not looked up afterwards.
Treating every wave as a standalone study. NPS verbatim themes are most valuable as a time series. Running each wave independently, without comparing to previous waves, means you cannot detect whether a theme is new, worsening, or resolving.
Confusing volume with severity. A theme mentioned by 15% of detractors may represent a manageable issue. A theme mentioned by 3% of detractors that correlates with your highest-value accounts is a revenue risk. Volume and severity are different signals.
How Skimle handles NPS verbatim analysis
Skimle is designed for exactly this use case: processing batches of open-text feedback (interview transcripts, survey responses, NPS verbatims) and surfacing structured themes with full traceability.
The workflow for NPS verbatims: export your NPS responses with metadata (score, customer segment, tenure band, contract value) from your NPS platform (Delighted, Medallia, Qualtrics, SurveyMonkey, or a custom CRM export). Import the verbatims into Skimle, tagging each document with its metadata fields. Run an analysis, either inductive (let the AI discover the themes) or predefined (apply your existing NPS coding framework). Review the themes in the categories view, slice by metadata in the statistics view, and export findings to the spreadsheet view for reporting.
Every theme links directly to the verbatim passages that generated it. If a stakeholder asks "which specific comments are behind the 'pricing' finding?", you can show them in seconds.
For market researchers and customer insights teams, see our customer and market researchers use-case page for more on how Skimle fits into a broader insights workflow.
For a comparison of survey tools, see our Typeform vs SurveyMonkey vs Google Forms vs Skimle comparison.
Frequently asked questions
How many NPS verbatim responses do I need for reliable analysis?
Themes start to become reliable at around 30 verbatim responses. Below that, a few strong comments can dominate the picture. At 100 responses, most core themes will be visible. At 300+, you can reliably do segment analysis. At 1,000+, wave-on-wave tracking and trend detection become the most valuable outputs. The limiting factor is usually not sample size but whether you have metadata attached to each response for segment analysis.
Should I analyse promoter and detractor verbatims separately?
Yes. Promoter and detractor verbatims address different aspects of the customer experience, and mixing them in a single analysis produces findings that average out the most important contrasts. Analyse each group separately first, then look at the full picture. Passive verbatims deserve their own read: they often surface latent needs that neither extreme expresses.
How do I handle very short NPS verbatims like "good" or "nothing to add"?
Short verbatims carry limited analytical value and should not be forced into a coding framework. Flag them as low-information and exclude them from frequency counts. If short verbatims dominate your dataset (more than 30% of responses are one or two words), that is a survey design problem. Your open-text prompt may be too generic. Prompts like "What is the one thing that would most improve your experience?" consistently produce more useful verbatims than "Do you have any other comments?"
What is the difference between NPS verbatim analysis and text sentiment analysis?
Sentiment analysis classifies text as positive, negative, or neutral. NPS verbatim analysis goes further: it identifies specific themes (what the comment is about), assigns sentiment to each theme, and aggregates across segments and waves. Sentiment analysis alone tells you that a comment is negative; thematic analysis tells you that the comment is negative about pricing, from an enterprise customer with more than two years of tenure, and that 23% of similar customers say the same thing. The latter is actionable; the former is a metric.
Ready to analyse your NPS verbatims at scale? Try Skimle for free and run your first NPS verbatim analysis. Upload your open-text responses, get structured themes by segment, and trace every finding back to the source comment.
Related reading: See our guides on how to analyse NPS verbatim comments, how to analyse customer interviews at scale, and qualitative consumer insights research.
About the authors
Henri Schildt is a Professor of Strategy at Aalto University School of Business and co-founder of Skimle. He has published over a dozen peer-reviewed articles using qualitative methods, including work in Academy of Management Journal, Organisation Science, and Strategic Management Journal. His research focuses on organisational strategy, innovation, and qualitative methodology. Google Scholar profile
Olli Salo is a former Partner at McKinsey & Company where he spent 18 years helping clients understand the markets and themselves, develop winning strategies and improve their operating models. He has done over 1000 client interviews and published over 10 articles on McKinsey.com and beyond. LinkedIn profile
