Qualitative analysis turns open-ended text into themes and reasons - the why behind the numbers. The definition, the methods, and how AI changes it.
Qualitative analysis is the practice of reading open-ended text, such as interview answers, open-ends, and case notes, into themes against a codebook so patterns and their evidence are defensible. Sopact calls what it produces the Connected Record: each theme kept on one participant record under a persistent Contact ID, tied to the number that respondent also gave, so the reason is read next to the score.
Data teams describe the same wall a food-bank analyst named on a call: Copilot “produces inconsistent results… the same input can yield different answers on different runs,” which makes the qualitative analysis hard to trust or present as a defensible picture. Open-ends pile up, an eight-to-nine-month analysis cycle drags, and the reason behind a score dies the moment it is collected, because the text sits in one system and the number sits in another.
Key takeaways
Most stacks split the work by design. The rating goes into a survey platform or a spreadsheet, and the open-text goes into a document or a separate coding tool, so the story and the score never reconnect and a low number sits in a column with no reason attached. Reconciling them later is manual, slow, and easy to dispute.
Sopact is record-centric: the open-text is read on arrival against a codebook and tied to the same persistent ID as the numbers, so qualitative analysis is an AND on one Connected Record rather than an either/or across exports. Read the whole survey on survey analysis, or the step-by-step on how to analyze survey data.
Analysts usually compare a handful of names: NVivo, ATLAS.ti, MAXQDA, and Dedoose for manual coding, and Excel for a lighter pass. Each helps a trained researcher code a corpus, and each was built as a workbench separate from where the numbers live, so the qualitative work is a project that finishes months after collection and lands in its own file.
The one test that sorts them: ask the tool to show a single rating with the exact sentence the same respondent wrote to explain it, on one record, and the same view for the prior wave. A separate coding tool answers by making you join two exports by hand. Sopact answers from the Connected Record, because the open-text was read on arrival and tied to the number.
The move that changes an analyst’s quarter is reading each open-text answer against the codebook the moment it lands, so themes and their evidence are ready as responses arrive rather than after a coding marathon. Sopact drafts the coding from the respondent’s own words, quotes the sentence, and keeps it tied to the number, so a human confirms or overrides a draft instead of starting from a blank codebook.
The result is longitudinal and defensible: a participant’s themes across every wave on one persistent ID, each traceable to a sentence and a score. Sopact keeps this on the Connected Record and reads on arrival, so a finding rests on the respondent’s own words rather than a spreadsheet no one can re-check.
A separate coding tool holds qualitative apart from the numbers and finishes months later; the Connected Record reads open text against a codebook on arrival and ties each theme to the number. The difference is whether the reason and the score are one record or two exports.
| The question | Separate coding tool | Connected Record |
|---|---|---|
| Code open text to themes? | Yes: by a trained hand | Yes, against a codebook on arrival |
| Tie a theme to the number? | No: a separate export | Yes: one participant record |
| Same answer on every run? | Depends on the coder | Yes: read against a codebook |
| See the reason in time to act? | No: coded at closeout | Yes: read the moment it lands |
Compare the two families on qualitative vs quantitative, or see the methods in depth on mixed-methods data analysis.
A dropping score is worth understanding while you can still respond to it, not in a report written after the program ends. The value of the open-text behind a number is highest the moment it lands, when the reason for a low rating can still change what happens next. That is the premise of the Loop, Sopact’s method for continuous intelligence: collect clean at the source, with the number and the open-text explaining it on one participant record; analyze on arrival, reading each open-text answer against a codebook the moment it lands and tying it to the number; improve in time, so the reason behind a dropping score surfaces while you can still act.
The Loop is also what makes a mixed-methods finding defensible: every theme traces back to the exact sentence a respondent wrote and the number that respondent also gave, the standard detailed in Loop traceability, so a conclusion rests on the Connected Record rather than a hand-coded spreadsheet no one can re-check.
One method, three moves that never stop
Then the next wave reads a little sharper. Read the method: the Loop methodology →
The fastest way to see the reading gap is to run it on your own data. Export a batch of open-ended responses with their IDs and the ratings the same respondents gave, 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 responses
Here is a batch of open-ended survey responses with each respondent's ID and their rating: [ATTACH]. Read each open-text answer against our codebook as it lands, tag the themes, quote the exact sentence behind each theme, and keep every answer tied to the number the same respondent gave, so I can read the reason next to the score on one record instead of in two separate exports.
Academy walkthrough → Connect the number and the reason
Here is our quantitative data and the open-ended responses on the same participant IDs: [ATTACH]. For each rating, pull the open-text the same respondent wrote that explains it, quote the sentence, and show the number and the reason on one record, so a low score carries the reason a respondent gave rather than sitting in a column with no explanation.
Academy walkthrough → Clean open-ended responses
Here is a raw export of open-ended responses on their participant IDs: [ATTACH]. Flag blanks, duplicates, and off-topic answers, normalize the text so it is analyzable, and keep each cleaned answer tied to its ID and the number that respondent gave, so the open-text is ready to read against a codebook on arrival rather than after a month of hand-cleaning.
Academy walkthrough → Find the drivers behind a score
Here are ratings and the open-ended responses on the same IDs: [ATTACH]. Read the sentiment in each answer, identify the drivers behind the rating with the sentence quoted, and tie each driver to the number, so I can see what is pushing a score up or down from the respondent's own words rather than guessing behind the average.
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 open-text against a codebook on arrival and keeping every answer tied to the number on one Connected Record.
It is reading open-ended text into themes against a codebook so patterns and their evidence are defensible. Sopact keeps the result on the Connected Record — one participant record under a persistent Contact ID where the open-text is tied to the number that respondent gave — so the reason is read next to the score.
Those tools help a trained researcher code a corpus in a workbench separate from the numbers. Sopact is record-centric: it reads open text against a codebook on arrival and keeps each theme on the Connected Record, tied to the same persistent ID as the numbers, so qualitative and quantitative are one record.
Sopact reads each answer against a codebook, so the same input gives the same themes on every run and each theme quotes the sentence behind it. The draft sits on the Connected Record for a human to confirm or override, which is what makes the analysis defensible.
Yes. Because the number and the open-text explaining it sit on one Connected Record, a low rating carries the sentence the respondent wrote to explain it, rather than sitting in a column with no reason attached.
No. Sopact cleans open-ended responses at the source, flagging blanks, duplicates, and off-topic text and keeping each cleaned answer tied to its ID, so the open-text is analyzable on arrival rather than after a month of manual cleanup on the Connected Record.
Not when it is read on arrival. Sopact reads each answer against the codebook the moment it lands, so themes are ready as responses arrive and the reason behind a dropping score surfaces in time to act rather than at the end-of-program report.
No. Sopact drafts the coding from the respondent’s own words with the sentence quoted; a human confirms or overrides it, human-in-the-loop. The Connected Record records what was read and from which answer, so the reasoning is traceable.
It keeps every answer on one persistent ID, so a participant’s themes across waves are read as a trajectory. The record survives the reporting cycle, which is what makes a longitudinal, defensible qualitative view possible on the Connected Record.
Next: compare the two families on qualitative vs quantitative, or read the reason behind an NPS score on NPS verbatim analysis.
