TL;DR:
Clean each interview into short, separate observations first, then group them by meaning, not by who said them.
Report counts as "in N of 8 interviews", counting people once, and attach one short quote to each finding.
A finding is a claim plus evidence plus what it suggests. A list of what people said is not yet a finding.
You have eight interviews, eight pages of notes in different shapes, and a meeting where someone will ask "so what did we learn?" Reading the notes again will not answer that. A repeatable pass will: split the notes into small pieces, group the pieces, count the groups by person, and write each theme as a claim with a quote behind it.
When you need this
You need this after any set of conversations where people talk freely: customer interviews, usability sessions, stakeholder chats. It is not for surveys with fixed answers, where you can simply total the responses.
What to set up first
One document or spreadsheet for the pass. A table works well: one row per observation.
Interviews labeled P1 to P8, never by name, with the real names kept in a separate private list.
The research question at the top, so you can drop observations that do not bear on it.
A rule that every observation keeps its source: P-number and, if you have it, the timestamp or page.
Step by step
Split notes into one observation per row: Each row says one thing a person did, said or struggled with. "Took 20 minutes to find the export button and then gave up" is one row. If a sentence contains two ideas, make two rows.
Keep exact words as quotes: Mark which rows are verbatim quotes and which are your paraphrase. Never tidy a quote into something the person did not say.
Group by meaning: Read down the rows and put similar ones next to each other. Name each group with a short plain phrase. Move rows freely; the first grouping is rarely the last.
Count people, not mentions: For each group, count how many different interviews have at least one row in it. Someone who complained about the same thing five times is still one person.
Write each finding as a claim: A claim names the group, gives the count, and offers one quote. Then add what it suggests, kept separate from what you saw.
Look for what is missing and what disagrees: Note any interview that contradicts a theme. A theme that holds in 5 of 8 has 3 interviews worth reading again.
Worked example: eight customer interviews about a small scheduling app
The app and the interviews are invented for this example. After splitting the notes you have 41 observation rows, grouped into three themes. You mark, for each interview, whether it has at least one row in each theme.
Interview | Slow first setup | Wants export to a spreadsheet | Pricing unclear |
|---|---|---|---|
P1 | Y | Y | |
P2 | Y | Y | |
P3 | Y | ||
P4 | Y | Y | |
P5 | Y | Y | |
P6 | Y | Y | |
P7 | Y | Y | |
P8 | Y |
In a spreadsheet with this table in A1:D9 (headers in row 1, interviews in rows 2 to 9), count each theme with COUNTIF:
B10: =COUNTIF(B2:B9,"Y")
C10: =COUNTIF(C2:C9,"Y")
D10: =COUNTIF(D2:D9,"Y")Check by hand. Slow first setup: P1, P2, P4, P6, P7 is 5. Export: P1, P3, P4, P5, P6, P8 is 6. Pricing: P2, P5, P7 is 3. So the formulas return 5, 6 and 3.
Writing the findings
Now each count becomes a sentence a colleague can act on, with the denominator always visible:
Export was the most common request: 6 of 8 interviews. "I end up retyping everything into my own sheet every Friday." (P3) Suggests: export is worth a closer look before new features.
First setup felt slow in 5 of 8 interviews. "I thought it would take two minutes and it was nearly half an hour." (P6) Suggests: watch someone do setup and time each stage.
Pricing was unclear to 3 of 8. "I could not tell what happens after the trial." (P7) With only 3, treat this as a lead to check, not a conclusion.
Both P1 and P4 raised setup and export, which is one reason to look at them again as a pair. And P3 and P8 raised only export, so the themes are not all one group of unhappy people.
Mistakes to avoid
Percentages from a tiny group. Writing
62.5%for 5 of 8 people suggests precision you do not have. "5 of 8" is honest and just as clear.Counting mentions. Eight complaints from one person is one interview.
Quotes as decoration. A quote must support the claim next to it. If you picked it because it sounds good, replace it.
Mixing observation and idea. "People find setup slow" is a finding. "Add a wizard" is a suggestion. Keep them in separate sentences.
Leading questions in the source. If you asked "Wouldn't export be useful?", agreement tells you little. Mark those rows so you discount them.
Doing it faster
A reusable table template saves the most time: one row per observation, columns for interview, quote or paraphrase, and group. If you read your notes out loud to a voice assistant such as DuckyHelper, keep in mind that grouping is a judgment call you should make yourself; use any tool to tidy, never to decide what the interviews meant.
Frequently asked questions
How many interviews do I need before I can call something a theme?
There is no magic number. Two interviews is a hint, and a theme that appears in most of your interviews is stronger. Always show the count out of the total so readers can judge.
Should I keep the quotes exactly as spoken?
Yes. Trim only filler words, mark any trimming with an ellipsis, and keep the speaker label.
Get started
