Where we are
Session 1 produced a brief built entirely out of what you and your colleagues believe. Tonight is about finding the parts of it that are wrong, cheaply, before they become a screen.
★ What you walk out with tonight. A pre-registered guess you can be judged against, a five-question discussion guide with no leading questions in it, and three design decisions traced back to specific things specific people said.
LiveWhy five, and how to say that honestly2 min▶
Five conversations will not tell you what percentage of your customers think anything. They are not a sample and pretending otherwise is how research loses its credibility with the people who understand statistics.
What five conversations do reliably give you is the shape of a problem, and the vocabulary people use for it. You are looking for the thing you did not know existed, not for a proportion. The third person saying "I did not know pause was a thing" is not 60 percent evidence. It is one discovery, and one discovery is enough to change a design.
Say it that way out loud and nobody argues: "This is qualitative. It tells us what is going on and what to call it. If we need a number, that is a survey or a funnel query, and it is a different piece of work."
The pre-registered guess
6 min liveBefore the first conversation, write down what you expect to hear. Seal it. It costs four minutes and it is the difference between research and confirmation.
LiveQuestions that do not lead, and the one that does most of the work5 min▶
The single most reliable question in product research: "Tell me about the last time you tried to do that." It is about a specific event, in the past, that actually happened. Everything else people tell you is a theory about themselves.
| Do not ask | Ask instead | Why |
|---|---|---|
| "Would you use a pause button?" | "Tell me about the last time you wanted to stop a box." | Nobody knows what they would use. Everybody remembers what they did |
| "Is the page confusing?" | "Show me where you would go to pause. Talk me through it." | The first invites politeness. The second produces behaviour you can watch |
| "How important is flexibility to you?" | "What made you sign up, and what nearly stopped you?" | Importance ratings are a way of agreeing with the person asking |
| "Do you like this design?" | "What would you do next on this screen?" | Liking is not using, and people will not tell you they dislike your work |
And the two most valuable seconds in any interview: saying nothing after they finish an answer. People fill silence with the part they were not sure they should say, and that part is usually the finding. Count to three before your next question.
"We are thinking about adding a pause feature - would that be useful to you? Do you find the current page easy to use?"
"Yeah, pause would be great." "The page is fine I think." You now have two agreeable sentences, zero information, and a strong sense that you were right all along. The interview took four minutes and confirmed a decision that had already been made before you sat down.
"Tell me about the last time you wanted to stop a box arriving. What did you actually do?"
"I was going to Spain for a month. I went into the account page, I could not see anything about pausing, and I did not want a box sitting outside my door for four weeks, so I cancelled and I thought I would just sign up again after. Then I did not, because honestly by then I had bought coffee somewhere else."
Quotes to insights to decisions, in three columns
6 min liveA transcript is not a finding. The work is the move from what someone said to what you will do about it, and it has exactly one rule: every decision must trace back to a quote you can point at.
Self-studyWhen what they say and what they did disagree4 min▶
It happens constantly and it is not lying. People report their intentions accurately and their behaviour badly.
- Behaviour wins on "what happened". The funnel says 31 percent of cancellations opened Options first. That is a fact. If an interviewee says "I went straight to cancel" and the log says otherwise, the log is right about the click and they are right about how it felt.
- Words win on "why". No dashboard will ever tell you they were worried about a parcel on the doorstep for a month. That only comes out of a conversation, and it is the thing that changes the copy.
- Disagreement is a finding, not an error. "They say the price is the problem, but the funnel says they never reach the price screen" is the most useful sentence a researcher can produce, and it usually means the stated reason is a socially acceptable stand-in for a different one.
The practical rule: use the log for the what, and the interview for the why, and treat any contradiction as the next question rather than as noise.
Six quotes to three decisions
★ 22 min · everyone buildsReal-shaped quotes from Daybreak's five conversations. Your job is the middle column, and then to notice which of your session 1 beliefs just died.
Predict. Three sentences, written before anything else, about what you expect to hear. Put a date on it. This is the only step that makes the rest falsifiable.
Recruit. Five people who did the thing recently. For Daybreak: three who cancelled in the last month, two who paused successfully. Recent matters more than representative, because memory decays fast and specifics are what you are after.
Ask. Five questions, twenty minutes, all about the last real time. Record it if they agree. Count to three before every follow-up.
Cluster. One quote per line. Group by what the person was trying to do, never by which feature they mentioned - feature grouping smuggles your existing roadmap into the analysis.
Decide. Each cluster becomes one insight sentence and one testable decision. Then go back to your prediction and mark what was wrong, out loud, in the write-up.
Read the six quotes below and write the middle column for each. One sentence per quote, and it must say something about the person rather than about the interface.
Quote 1. "I assumed Options was notification settings." P3, and P5 said almost exactly the same thing.
Quote 2. "I did not want a box sitting outside my door for four weeks." P3. Nobody mentioned money.
Quote 3. "I thought I would just sign up again after." P1, P3, P4. All three cancelled. None of them came back.
Quote 4. "I have got three bags in the cupboard, I just drink less than I thought." P2. This person does not want a pause. They want it to arrive less often, permanently.
Quote 5. "I did pause it once. I never got told when it was starting again, so I forgot and then a box turned up." P4. A successful pause that still produced an unhappy person.
Quote 6. "Honestly I would have just emailed you but I could not find an address either." P5. The escape hatch failed too, which is why the funnel shows a cancellation.
Now cluster them by intent, not by feature. You should end up with roughly three groups: "I want a break", "I want less coffee", and "I want to know what happens next". Notice that quote 4 does not belong in a pause feature at all.
Mark your prediction. Session 1 assumed this was one problem - people cannot find pause. It is at least two, and the second one - "I drink less than I thought" - needs a frequency change, not a pause. Write that down as a belief you gave up.
If you take one thing from tonight. Cluster by what the person was trying to do, never by the feature they mentioned. Feature clustering produces the roadmap you already had, with quotes attached to make it look researched.
Before session 3
2 hr over three days- Write your three predictions. Dated, sealed, before anything else. Four minutes.
- Talk to three people. Not five if five is unrealistic - three real conversations beats five imaginary ones. All questions about the last real time.
- Do the three columns. Quote, insight, decision. Every decision points at a quote.
- Mark what you got wrong. Explicitly, in writing. If nothing was wrong, look hard at whether you led them.
- Find one place where the log and the words disagree. That contradiction is your best question for the next round.
Three questions
Clear all three and session 2 gets stamped in your passport.
1 · Why write down what you expect to hear before the interviews?
Without a sealed prediction, hindsight quietly rewrites what you believed, and five conversations become an expensive ceremony. The row where the prediction was wrong is the entire value of the exercise - and if no row is wrong, suspect the questions.
2 · Which question is most likely to produce something you can design with?
Nobody knows what they would use, and importance ratings are mostly a way of agreeing with the person asking. A specific past event produces behaviour, obstacles, workarounds and the person's own vocabulary - all of which you can design with.
3 · The funnel says people opened Options before cancelling. An interviewee says they went straight to cancel. What do you do?
People report intentions accurately and behaviour badly, and neither source is lying. The log owns the click; the conversation owns the reason. A contradiction usually means the stated reason is standing in for a different one, which is worth chasing rather than resolving away.
What this session teaches, and where it came from
Daybreak's quotes are written for teaching and are labelled as such in the course's source map. The method is real; the participants are not.
Session 2 cheat sheet · pin this
You now know it is at least two problems, not one. Session 3 gives the screen a structure - task flows before screens, the cost of every level of nesting, and deliberately ugly wireframes, because a polished mock gets you feedback on the polish. Plus the empty, loading and error trio, which on a screen like this one is not the edge case, it is Tuesday.