Solutions · NPS & CSAT Understand NPS and satisfaction scores through the comments behind them.
Compare topics across rating groups and check the matching customer voices.
Why a six – and not a nine? The reasons are in the comment.
NPS is quick to calculate. Analyzing the comments behind it takes time, wave after wave.
The challenge: Comments nobody reads in full
Every wave brings hundreds or thousands of comments. A sample gets read; the rest stays in the spreadsheet.
With deepsight: Every comment with substance counts
Topic analysis assigns every comment with substance to one or more topics – in minutes rather than weeks, even with thousands per wave.
Go to topic analysisThe challenge: Passives stay invisible
People who give a 7 or 8 only count towards the base of the NPS. The report shows detractors and promoters; hardly anyone reads the passives’ reasons.
With deepsight: Passives as a group of their own
NPS analysis treats passives as a group of their own. You see their topics next to those of detractors and promoters.
Go to NPS analysisThe challenge: The loudest voice in the meeting
The most dramatic comment makes it into the management meeting. Whether it speaks for many remains an open question.
With deepsight: Shares with a base, quotes as evidence
For each group you see the share of comments that mention a topic, along with the base – and read those comments verbatim.
Go to group comparisonThe challenge: Re-sorted every quarter
When NPS drops, management wants to know what detractors are now writing about more often. If every wave was categorized from scratch, that is hard to compare.
With deepsight: Every wave on the same basis
Add follow-up waves to a time series and align topics across waves – with your own codebook if you like.
Recurring analysis
Three values that are easy to mix up
Individual rating
One person’s answer on the scale from 0 to 10, for example a 7. It places the person in a group.
NPS
One value for all analyzed responses: the share of promoters minus the share of detractors. It ranges from −100 to +100.
Topic share
The share of a group’s comments that mention a topic. It describes what people talk about, not how strongly it moves the score.
CSAT scales such as 1 to 5 have no fixed groups like NPS. You decide which values to group together – as a separate column in your file.
Which topics do the groups mention?
Select a group. In the analysis, significance indicators help you assess differences between groups; they do not prove a cause.
Illustrative values. n = analyzed responses in the group; topic shares refer to this base. NPS on the same base: +9.
“It often takes 20 minutes before anyone answers the phone. The price increase came without any explanation.”
Select a category to see the matching passage.
What your comments can tell you
- Which topics do detractors mention more often than the rest of the respondents, and how clear is the difference?
- What exactly do promoters highlight – advice, product, availability?
- Do detractors and promoters mention the same topic, but with opposite sentiment – the price, for example?
- Do the reasons differ by region, channel, plan or customer segment?
- What lies behind low CSAT scores after contact with customer service?
- Which topics have come up since the last wave and which are mentioned less often – for example after a change to the bill or the app?
Good to know: You can also put follow-up questions from management directly to the assistant. Its answer separates calculation, evidence and interpretation.
The same NPS can come from very different groups.
An NPS of +10 can mean 40% promoters and 30% detractors – or 20% and 10%.
What you bring
One row per respondent: rating, comment and attributes such as region, channel or plan. deepsight suggests suitable columns for comparisons.
Good to know: We set up a direct connection to your feedback system as a custom project, after a joint assessment.
From export to group comparison
- Prepare the data
Upload the file and filter out empty or content-free comments such as “no comment” up front.
Data quality check - Build topics
deepsight suggests topics, drawing on your study context with key terms such as plan or product names. Or you apply your own codebook.
Topic analysis - Check against the original
Read the comments for each topic verbatim, correct assignments, and merge or split topics. Every change is saved as a version.
Topic refinement - Compare groups
Put topics and sentiment side by side for each rating group and by attributes such as region or channel.
Group comparison
People who don’t comment are missing from the topic shares.
How many people comment can differ from group to group – so always read shares together with their base.
What you pass on to management and departments
What matters with NPS and CSAT comments
What goes into the base
Shares and key figures refer to the analyzed responses, not to all respondents. Anything the data quality check filters out, such as empty comments, doesn’t count toward the base – including for the NPS: with this check, it is the NPS of respondents with a usable comment.
Compared with the rest
Significance indicators compare one group at a time with the rest of the sample – so detractors are compared with passives and promoters combined, not directly with promoters. The base limits are fixed: below 10 responses, values are flagged or hidden, and significance indicators appear from 30 responses per group.
Read NPS and CSAT separately
Likelihood to recommend and satisfaction with a contact measure different things, often among different people. Analyze them as separate ratings instead of combining them into one value.
Set up follow-up waves the same way
Ask in every wave with the same wording and at a comparable point in time, for example right after the contact. You save the study context, objective and glossary as an analysis profile and reuse it for the next wave.
Good to know: Topic shares and group comparisons can be weighted – with a weight column from your file or to target distributions (rim weighting). Weights do not automatically make a sample representative.
Questions about NPS and CSAT comments
Yes. Short comments are assigned to topics too, as long as they say something. For new surveys, you can switch on an optional AI follow-up: if an answer is brief, it asks a follow-up question. Participants can skip it.
You set the priorities, because effort, responsibility and feasibility aren’t in the comments. As a basis, you see how often a topic is mentioned in each group and what customers write about it.
Our anonymization detects personal information such as names, email addresses or phone numbers without a language model and masks it before the analyses run. If translation is switched on, texts are translated before masking. The anonymization does not detect every piece of information – check results before sharing them, especially quotes for departments. For AI features we use Azure OpenAI in the EU; Microsoft does not use your data to train AI models.
Comments are sorted by topic regardless of the scale. As with CSAT, you create groups such as “high effort” and “low effort” as a separate column in your file and compare them just like the NPS groups.
Related to NPS & CSAT
Analyze your NPS comments
See the group comparison in a personal demo, using an example from your field. Afterwards, you’ll receive test access.