Guide
Customer interview analysis for product teams without a researcher.
Write down the question the round was meant to answer. After each interview, note what surprised you. Then go through the transcripts and keep the sentences about what people actually did, group them by what sits underneath, count people and companies per group, and write one page that answers the question with quotes.
Updated

Debrief, ten minutes after
- Three things that surprised me
- The best sentence, word for word
- What they do today instead
How is an interview round different from everyday calls?
Everyday customer calls are a stream: you collect what comes up and count it over months. That is covered in How to analyse customer feedback from your calls.
An interview round is a project. You talk to a handful of people because you want to answer one question, usually in a week or two. And you asked every question yourself, so almost everything they said was an answer to you. That changes what counts as evidence: the useful split is no longer asked or unasked, but what people did against what they say they would do.
Step 1: What question are you answering?
Write it down before the first interview, in one sentence. "How do teams decide what to fix after a customer complains?" is a question. "Learn about our users" is not.
Everything later gets measured against it. Without it, analysis turns into a list of everything anyone said, and the loudest interview wins.
Step 2: What do you do straight after each interview?
Spend ten minutes, the same day, before the next one. Memory is best right after the call and fades fast, and the surprises are what you will forget first.
Interview: [company, role, date] Three things that surprised me: 1. 2. 3. The best sentence, word for word: "..." What they do today instead: What I would ask next time:
Keep the recording or transcript too. The debrief is for what you noticed. The transcript is for what was actually said.
Step 3: Which sentences count as evidence?
Go through each transcript and keep sentences in the person's own words, with who said them. Not every sentence weighs the same:
| Kind | Made-up example | Weight |
|---|---|---|
| What they did | "Last quarter we rebuilt the report twice because nobody trusted it." | Strong |
| What they do instead | "Mia keeps a spreadsheet that fixes the export every Friday." | Strong |
| What it cost them | "That's most of a Monday morning for two people." | Strong |
| What they say they would do | "We'd definitely pay for something that fixed that." | Weak |
| Opinions and compliments | "That sounds really useful." | Ignore |
People are kind in interviews, and bad at predicting their own behaviour. Rob Fitzpatrick's book The Mom Test is the classic on asking questions that avoid both problems. When you analyse, the same rule applies: past behaviour and workarounds outweigh any amount of "I would".
Also note the questions that got a shrug. A problem nobody had a workaround for is often a problem nobody really has.
Step 4: How do you group and count?
Put each kept sentence on a card or a row and group them by the problem underneath, not by the question that produced them. Sticky notes on a wall work. So does a spreadsheet with a column for the group.
Then count, per group, how many people and how many companies said something in it. With eight interviews the numbers are small, and that is fine: "six of eight people, from five of six companies" is still more honest than "users want".
Count people, not sentences. One talkative person can fill a group on their own, and on a wall of sticky notes it looks like a pattern.
Step 5: How do you write it up?
One page that answers the question from step 1. Lead with the answer, then the evidence, then what you did not expect.
# [The question this round set out to answer] ## Short answer [Two or three sentences. Yes, no, or "not the way we thought".] ## What we saw [Group]: [N] of [N] people, [N] of [N] companies "[Quote]" [role, company] "[Quote]" [role, company] [Group]: ... ## What surprised us [Anything that changes the question itself.] ## What we still don't know [And how we would find out.] ## What we suggest next [A brief, another round, or stop.]
If the answer is "this is worth building", the next document is a product brief, and the quotes move straight into its evidence half.
How many interviews are enough?
Enough to stop being surprised. Keep a running list of groups as you go. When the last two or three interviews fit into groups you already have and add nothing new, the round has done its job.
If every interview still adds a new group, your question is probably too broad, or you are talking to people who are too different from each other. Narrow one or the other before adding more interviews.
Should an assistant do the analysis?
An assistant is useful for a first pass: pulling candidate quotes, or checking whether a group shows up in an interview you have not reread. Give it the question from step 1 and ask for exact sentences with who said them.
Do not hand it the judgement. Assistants summarise, and a summary smooths away exactly the details that matter: the hedge, the workaround mentioned in passing, the difference between "we did" and "we would". Check every quote against the transcript before it goes into the write-up. If your interviews are in Granola, Use the Granola MCP to read your customer calls has prompts that keep the exact words.
What happens after the round?
The round answers one question, then the transcripts usually sit in a folder. But the same problems keep coming up on sales, onboarding and success calls for months afterwards, and that is where you find out whether the round was right.
Ordagrann reads the customer calls you record in Granola, groups every problem and request into themes, counts how many companies raised each one, and turns a theme into a one-page brief your team can build from. Interviews and everyday calls feed the same themes, so what you heard in the round keeps being counted after it ends.
Works with Granola Business. Any other transcript can be pasted in.