Solutions · Customer surveys Look at satisfaction scores and open-ended reasons together.
Organize open-ended reasons by topic, compare customer groups and check the statements behind every result against the original wording.
Why a two and not a five? The reasons are in the comments.
Often hundreds or thousands of them. Between those comments and a finished report lies a lot of manual work.
The challenge: Coding by hand
Reading comments, assigning codes, keeping tally sheets. In large studies, often only a sample gets coded.
With deepsight: Every response, not just a sample
Topic analysis assigns every open-ended response – using its own suggestions or your codebook. You review the result instead of coding response by response.
Go to topic analysisThe challenge: Code frames that drift between waves
Whoever codes shapes the result. When the team or the code frame changes, waves no longer line up cleanly.
With deepsight: The same categories in every wave
With your codebook, every wave refers to the same categories – no matter who runs the analysis. Changes to the topic system are kept as versions.
Refine topicsThe challenge: Hunting for quotes for management
A fitting quote for every slide: filtering raw data, searching comments, copying.
With deepsight: Evidence right where the result is
For every topic, you see the assigned responses verbatim. You find suitable quotes at the result, not in the raw file.
Go to reportingThe challenge: Export after export
Questionnaire in one tool, tables in the next, charts rebuilt in PowerPoint. Every correction starts from scratch.
With deepsight: One project from response to report
Whether you run the survey in deepsight cloud or upload exports from your own tool: from fieldwork or upload onward, analysis, deep dives and reporting stay in one project – with no exports in between.
Go to deepsight cloud
Two ways into the analysis
Your responses are already in
- Upload data
An Excel or CSV file with ratings, open-ended responses and attributes such as line of business or customer group.
- Check topics
Review suggested categories or your codebook against the original responses.
Refine topics - Compare groups
Compare topics by rating and customer group, then export the results.
Share results as a report
From the rating all the way to the report
An insurer surveys customers after their claim has been settled: one satisfaction question, one open-ended question about the reason, plus line of business and contact channel.
How satisfied were you with how your claim was handled?
“The agent on the phone was very friendly, but the payout took seven weeks and I had to ask about the status several times.”
Categories in this response
In the report: share of each category for auto claims, with sample responses.
Illustrative example. Select a category to see the matching passage.
Questions the analysis helps you answer
- Which topics do dissatisfied customers mention most often, and which do satisfied ones mention?
- How do the reasons differ between customer groups, lines of business or channels?
- Which responses lie behind a notable score?
- Which topics are new since the last wave, and which are mentioned less often?
- Does your codebook still cover what customers write?
What you bring
Each open-ended response stays linked to everything else the respondent told you.
“Friendly, but too slow” is praise and criticism at once.
From upload to topic review
What deepsight prepares and where you make the call.
- Screen responses
Empty, unintelligible and content-free responses such as “no comment” are assigned to categories and filtered out before the analysis. The check is optional.
Data quality check - Mask personal details
Our own solution detects personal information without a language model and masks it before the analysis. Not every detail is detected.
Anonymization - Assign topics
deepsight assigns the responses to matching topics – from its own suggestions or from your codebook.
Topic analysis - Sentiment per topic
Sentiment is assessed for each topic, not just for the response as a whole. That keeps praise and criticism in one comment apart.
Sentiment analysis
What you end up with
Good to know: Wave comparison is included in every paid plan – see the plan comparison. If your result tables already exist in Excel, for example from your analysis software, slide factory turns them into slides in your own template.
Behind every topic share are the responses, word for word.
From result to original response – for checks, follow-up questions and quotes.
Backing up your results
Check against the original responses
For every category, you see the assigned responses verbatim. Check samples for each category before you share results.
Every change as a version
Rename, merge, split, remove: each change is saved as a topic version, and the previous state is kept.
The same settings for every wave
You save the study context, objective and glossary as an analysis profile. It is versioned and can be reused for follow-up waves.
Putting differences in context
A topic share shows what a group talks about, not how strongly it affects the rating. If you wish, significance indicators flag groups that stand out statistically, tested from 30 cases per group; values with a base below 10 are flagged or hidden. You can weight results by a weight column or to target distributions (rim weighting); weights do not automatically make a sample representative.
Good to know: A new AI run can produce different results. Completed runs are kept as saved results and can be reopened later.
Questions about analyzing customer surveys
Yes. Use your own codebook – including parent and child categories – to analyze open-ended responses, either strictly by your categories or with additional suggested topics. You upload it as an Excel file (.xlsx) or CSV, optionally with descriptions and parent categories. You do not need to train a model for this. On request, deepsight checks the responses in the residual “other” category against your categories again; select at least five of them and it suggests a new category.
That depends on the research question, the data and the category system, which is why we do not quote a blanket accuracy figure. You correct individual assignments as needed. For methodological validation, compare the results with a manually coded subsample, for example in the Excel export.
That depends on the volume, the research question and how much review is needed. The assignment itself runs automatically – minutes rather than weeks, compared with manual coding. Most of your time goes into reviewing and refining the topics.
Each analyzed open-ended response counts as one text: 1,000 people with three open-ended questions make 3,000 texts if all responses are analyzed. Closed-ended responses such as ratings and quantitative tables do not count. A new completed analysis run can count the same texts again. Allowances per plan: see our pricing page.
You decide which kinds of information are masked, such as email addresses or phone numbers; the names category is always switched on. The anonymization does not automatically detect indirect references, such as a role combined with a location – you mask these by hand, with versioning. For AI features we use Azure OpenAI in the EU; Microsoft does not use your data to train AI models. Learn more under Security.
With the Assistant you ask questions about your analysis in your own words, for example which topics come up more often for auto claims. The answer separates calculation, evidence and interpretation; check it before adding it to a report.
Related topics
Analyze your customer survey
In a personal demo, we show you the path from rating to report. Afterwards, you receive test access. We agree on the scope of the test with you – for example, a completed wave whose results you already know.