Interview synthesis
Interview synthesis finds the themes across a set of interviews and shows the evidence under each one: the quotes it came from, by number, and how many people said it. The strength of a theme is counted in code from the quotes attached to it, so it cannot be talked up.
- Asks for
- The interviews — notes or transcripts, two at a minimum and ideally five or more
- What the round was for, so the themes answer the question you asked
- Gives back
- Themes across the set, strongest first, each with its evidence by number
- How many people each theme rests on, counted from the quotes
- What contradicts a theme, where somebody said the opposite
- What the round did not settle, and what you would have to ask to settle it
- Plan
- Pro and Max
What it gives you
Theme 2 — The weekly report is abandoned, not disliked. 4 of 6 participants. Evidence: #3, #11, #17, #24. Three of the four said it unprompted. #11 "I gave up on it in March. Now I just paste the Jira filter into Slack." #17 "It is a good doc. I have not written one since the reorg." Against: #19 "We still do it every Friday, it takes me ten minutes." (1 participant.) Not settled: nobody was asked what happens when the report is skipped.
How it works
One interview is an anecdote; five are a finding, if you can show the join. The usual way of getting from one to the other is a wall of sticky notes and an afternoon, and the output is a set of themes whose support nobody can reconstruct a month later when somebody asks how many people actually said that.
This keeps the join. Every piece of evidence is numbered as it goes in, every theme names the numbers that support it, and nothing can appear in a theme that is not traceable to a quote somebody actually said. If a theme has two sources it says two, and it says so whether or not it is the theme you were hoping for.
The counting is deliberately not the model’s job. How many people a theme rests on, and how many of them said it unprompted, are arithmetic over the quotes attached to it — worked out in code, after the writing, so the number in the heading and the quotes underneath it cannot disagree. That is the one claim on the page you can check line by line.
Who it is for
A PM who has finished a round of discovery and has to tell the team what they learned by Thursday, with the evidence to back it.
Where it stops
It synthesises what you give it and will not reach past it: with two interviews it says the themes rest on two, rather than writing five confident findings. It also cannot tell you whether your sample was the right one — six interviews with people who all came from the same customer list produce a strong, consistent, possibly worthless answer.
Questions people ask
How many interviews do I need before synthesis is worth doing?
Five or more gives you themes with enough support to act on, and the usual rule of thumb is that new themes stop appearing somewhere around six to eight for one type of user. Two or three still works and is often worth doing early, but the output will say the themes rest on two or three people and will not pretend otherwise. A theme with one source is reported as a quote, not a finding.
Can it invent a quote that nobody said?
No. Every quote is numbered when it goes in and a theme may only cite the numbers it actually has, so a sentence in the output traces back to a line in your source or it is not there. Where the synthesis is drawing an inference rather than quoting, it is written as an inference and names the quotes it is inferring from, so you can disagree with the reasoning rather than the transcript.
What if two interviews contradict each other?
The contradiction is reported rather than averaged away, under the theme it concerns, with the quote that disagrees and how many people it represents. This matters more than it sounds: a theme four people support and one flatly contradicts is a different thing from a theme five people agree on, and a synthesis that quietly drops the fifth is the one that leads a team into building for an audience that was never unanimous.
Run it on your own work
Interview synthesis needs an account, because it writes against what you give it and remembers what you have told it before. It is part of Pro and Max. Or see what the plans cost first.