Drawing (The Right) Conclusions
After all your hard work, focus on drawing conclusions you can count on. Avoid two of the most common mistakes teams make after good discovery work: concluding too early that you understand the pattern, and selectively using evidence that confirms what you already believe while discounting evidence that challenges it
Qualitative, Not Quantitative
Discovery interviews generate qualitative data, not quantitative data. You are not trying to achieve statistical significance — you are looking for patterns, themes, and signals that are worth investigating further. This is important to understand because it cuts both ways:
- You don't need 50 interviews to identify a real pattern
- You also can't make confident market decisions from two interviews, no matter how compelling they were
The Rule of Seven
A useful rule of thumb: when you hear the same thing from approximately seven people with meaningfully different backgrounds — different roles, different institutions, different geographies — you are probably close to something real. Seven is not a magic number, but it reflects the point at which a pattern becomes hard to explain away as coincidence or selection bias.
If you're hearing something from three people who all work at the same institution, know each other, and were referred to you by the same contact — that's still effectively a sample of one cluster, not three independent data points.
A Simple Synthesis Process
The right way to draw conclusions is to separate data collection from interpretation, and interpretation from conclusion. Moving too quickly between these stages is where most errors happen.
Gather all raw notes before discussing conclusions
Do not debrief as a team until everyone has independently reviewed the full set of interview notes. Group discussion before individual review allows the most vocal team member to anchor everyone else's interpretation.
Tag each observation by theme independently
Each team member tags raw observations by theme on their own before comparing notes. Themes should emerge from the data — not be imposed on it. If you find yourself forcing an observation into a category, that is a signal.
Count occurrences across independent sources
For each theme, count how many independent sources raised it — unprompted, in their own words. Weight independent sources more heavily than clusters. One theme raised by seven unconnected people is stronger evidence than the same theme raised by seven people from the same institution.
Rank themes by frequency and independence
Produce a ranked list of themes before drawing any conclusions. This forces the data to speak before your interpretation does. Themes that appear frequently across independent sources deserve the most weight — regardless of whether they are the ones you hoped to find.
State conclusions as hypotheses, not facts
Discovery data supports hypotheses — it does not confirm them. "The data suggests that X is a significant problem for Y" is the right framing. "We confirmed that X is a problem" is not. The distinction matters when you present to mentors, funders, or program directors who will push back on overconfident claims.
Identify the most challenging finding and address it explicitly
Before finalizing your synthesis, identify the single finding that most challenges your current direction. State it clearly. Explain how you are accounting for it. If you cannot account for it, it belongs in your synthesis as an open question — not buried or omitted.
Extracting Lessons Without Bias
The analysis phase is where confirmation bias does the most damage. By the time you've finished interviews, you have emotional investment in what you found. Some findings will feel more important than they are because they confirmed what you hoped. Others will be minimized because they were inconvenient.
To counteract this:
- Tag before you analyze. Sort raw findings by theme before you draw conclusions. Don't start with the conclusion and work backward.
- Count disconfirming evidence. For every finding that supports your hypothesis, actively look for contradictions in your data.
- Report frequencies honestly. "Most people we spoke with" means something different from "one person mentioned" — be precise about how many.
- Involve someone skeptical. Have a team member or mentor review your synthesis with the explicit brief to challenge your conclusions.
The first trap is premature conclusions — deciding you understand the pattern before you have enough data to support it. The second is confirmation bias — weighting evidence that confirms what you already believe and discounting evidence that challenges it. Both let your prior beliefs do work that the data should be doing.
A team conducts three interviews. All three are positive — the interviewees confirm the problem is real and express interest in a solution. The team stops interviewing and begins building. What they don't know: all three were referred by the same clinical champion, work at the same institution, and share the same workflow. When they reach their fourth, fifth, and sixth interviews — at community hospitals with different staffing models — the problem barely registers. Three interviews felt like a pattern. They were a cluster.
A team conducts 10 interviews. Eight people say the problem is minor or manageable. Two people — both enthusiastic early adopters at well-funded academic centers — say the problem is acute and they would pay to solve it. The team builds their pitch around the two. The two are real. They are just not the market.