How do you synthesize customer interviews into insights?

Synthesis is deciding what you just heard while it is still true. It happens three times: after each conversation, before memory fades; weekly, when patterns can outrank anecdotes; and once at the end, when the problem analysis matrix tells you which cluster deserves a statement. Stop interviewing when you can predict the next answer. Never stop listening.

What is a problem analysis matrix and how do you use it?

Synthesis is deciding what you just heard while it is still true. It happens three times: after each conversation, before memory fades; weekly, when patterns can outrank anecdotes; and once at the end, when the problem analysis matrix tells you which cluster deserves a statement. Stop interviewing when you can predict the next answer. Never stop listening.

When should a startup stop doing customer interviews?

Synthesis is deciding what you just heard while it is still true. It happens three times: after each conversation, before memory fades; weekly, when patterns can outrank anecdotes; and once at the end, when the problem analysis matrix tells you which cluster deserves a statement. Stop interviewing when you can predict the next answer. Never stop listening.

Synthesis · Product Sprint · Product Thinking

§The question this chapter answers: what did we actually learn, as opposed to what do we remember hearing?

§After ten conversations you have thirty pages of notes and a strong feeling. The feeling is the danger. It is usually the last conversation, or the most articulate person, or the one who agreed with you, and it is about to become your unless you do the work this chapter describes.

§ 6.1Three moments#

§After each conversation. Before you leave the building, before you check your phone. Write the quotes that struck you, verbatim. Write what surprised you. Name the themes you heard. Then open your research plan and mark which assumptions moved and in which direction. This takes fifteen minutes and is the most valuable fifteen minutes of the day, because by tomorrow you will remember what they said and not how they said it, and the how was the finding.

§Weekly. Read every page from the week together. Look for what repeats. Look, especially, for what contradicts: two people from the same segment describing the same problem in opposite ways is not noise, it is usually a segment boundary you have not drawn yet. Update the interview shape for next week; some questions have stopped teaching you anything and should go.

§Once, at the end. When you can predict the answers, gather everything and build the matrix below.

§ 6.2Patterns over anecdotes#

§The method is old and still the best: every observation on its own note, every note on the wall, similar notes grouped in silence, groups named. Erika Hall's version fits a team of one to five in an afternoon. The rule that matters is the silence. If the team talks while grouping, the loudest person's grouping wins, and the exercise exists precisely to stop that.

§Group by problem, never by solution. The most common mistake at this step is a cluster called "needs a mobile app," which is not a problem anyone has. Keep asking of every group: what is the underlying thing going wrong here? A group you cannot name as something that goes wrong for a person is a solution in disguise.

§ 6.3The problem analysis matrix#

§When the groups have names, score them. Four columns, one row per cluster.

Problem cluster	Frequency	Pain	Current solution	Will pay
Reorder happens too late	Daily	High	Count at closing	Yes, said so unprompted
Supplier prices vary	Weekly	Medium	Spreadsheet	Maybe
Staff waste on prep	Weekly	Low	None	No

§Frequency and pain come from the stories: how often did it come up, and how did people describe the cost. Current solution comes from the workarounds: "none" is a warning, not an opportunity, because it often means nobody cares enough to try. Will pay comes only from : someone said so, or someone already pays for a worse version. Be brutal here. Teams routinely build for the cluster that scored high on frequency and low on everything else, because frequency is the easiest thing to hear.

§ 6.4The learning wall#

§Keep one place, physical or shared, where four things accumulate across the whole : key quotes, surprising findings, patterns, and open questions. Update it after every conversation. Read it before every conversation. Its value is not the content, which lives in the pages. Its value is that the whole team, including the person who did not attend the interview, sees the same evidence and is corrected by it. Gary Klein's research on how insights form points at exactly this: they come from a contradiction noticed, an anomaly that will not fit, and the wall is where anomalies become visible.

§ 6.5When to stop#

§Stop interviewing a segment when three things are true. You can predict what the next person will say. You have stopped learning anything new for two or three conversations in a row. Your core assumptions have each been either supported or contradicted by more than one person.

§Then stop interviewing, and do not stop listening. Every company that has stayed close to its market has kept a steady cadence of conversations long after the first version shipped. The sprint ends; the habit does not.

§ 6.6What you leave with#

§A learning wall the whole team has read. A problem analysis matrix with every cluster scored. A short list, two or three clusters at most, that scored high on more than frequency. And an honest note of the contradictions you could not resolve, because those are what the problem statement in the next chapter has to survive.

6. Synthesis