Customer Discovery Synthesis for Founders

Before you reread a single transcript, list the beliefs you went in with, one line each. Cover who has the problem, how often it hits, what they do about it today, who approves the purchase, and what they would pay.

Then note which questions you asked and who heard each one. If you didn't keep an interview guide, rebuild it from the transcripts. Without it, you can't tell a problem nobody has from a question nobody asked.

Take one interview at a time, start to finish, and pull its quotes before you compare it with any other. Teresa Torres, who teaches customer interviewing at Product Talk, says the biggest mistake she sees in synthesis¹ is skipping that step. Condense a 20- to 30-minute interview into a paragraph or two instead, she warns, and you lose its context, nuance and detail.

1. 1

   Copy the exact words

   Paste each useful sentence as it was said. Never tidy a quote into your own phrasing or merge two people's words.

2. 2

   Tag who said it

   Note the person, their role and their segment, so you can count people later and see where each problem lives.

3. 3

   Mark how it came up

   Did they raise it before you mentioned it, agree when you asked, or disagree? Decide now, while the context is in front of you.

4. 4

   Keep the specifics

   Past events, numbers, workarounds and money already spent are evidence. Compliments and predictions about the future are not.

Sort the quotes into themes, on sticky notes or in a spreadsheet, and give each theme a plain name. Nielsen Norman Group, a user experience research firm, says a theme emerges when related findings appear multiple times across participants². So the unit is people, not mentions.

A single owner who talks for 20 minutes about a scheduling spreadsheet that keeps breaking can give you 30 quotes and still be one person. Rank themes by how many people raised each one unprompted, and keep the quote count beside it to see where the two rankings disagree.

Look at who those people are, too. If every unprompted mention comes from firms with more than 10 sites, you have learned something about your buyer as well as the problem.

Agreement is cheap in an interview. Survey researchers call the tendency to agree with a statement put to you acquiescence bias. Pew Research Center notes that it is even more pronounced when an interviewer is present³.

So give each theme four counts: people who raised it unprompted, people who agreed when asked, people who disagreed, and people never asked. The last count is the easiest to forget, because nobody said anything.

Pool the first two columns and overtime looks like 12 of 14 people. Only 4 raised it themselves, so pooling triples the finding. Willingness to pay looks strong at 5 of the 6 people asked, or 83%, yet nobody volunteered it and 8 were never asked. A percentage over all 14 counts those 8 as a no, and one over the 6 hides them.

Go back to the list you wrote first and give each belief one of four verdicts: confirmed, disconfirmed, untested, or not testable with this sample. Watch for untested beliefs hiding behind an enthusiastic yes.

Treat what people say they would pay as your weakest column. Gustaf Alströmer, a Y Combinator group partner, warns that people asked "Will you use our product?" are probably going to say yes, but that's not going to mean anything⁴.

His first test of a valuable problem is whether people already pay for other solutions. So look for money already spent, a budget someone names, or a commitment such as a paid pilot with success criteria both sides sign.

Once a problem is confirmed, make it the problem statement in a product requirements document, with the quotes as evidence.

How many interviews is enough depends on what you need to know. In one health study of 25 in-depth interviews, Monique Hennink and colleagues found the range of issues appeared by interview nine, but understanding them took 16 to 24⁵. The first interview alone produced 53% of the new codes, which helps explain why early calls feel so conclusive.

Aim the next round at what is still moving: themes still gaining detail, untested beliefs, and groups you have barely heard from. Ask about the last time the problem happened, what it cost and what they did. Alströmer's advice is to hold your idea back until the end of the call, if you mention it at all.

Judging product-market fit is a bigger job. NSF's I-Corps program, a seven-week course in customer discovery, expects each team to complete at least 100 interviews⁶ with potential customers and other stakeholders. Competitors' public reviews make a useful cross-check, since nobody prompted them, and a competitor scan counts the complaints that repeat.

Rather not do the counting by hand? The tool at the top of this page reads each interview, counts every theme by who volunteered it and who was asked, and drafts a verdict on each belief. Every theme comes with the quotes behind it.