What the agent's passport says
The limitation of the CustDev Agent is formulated directly: the sample is not equal to market statistics. The extended version sounds even stricter: the agent does not prove the market size, does not turn a small sample into a statistically significant one and does not replace the researcher's conversation with users. This is not a footnote in small print — it is the boundary around which the whole process is built.
What a small sample honestly gives
Five to ten interviews are a normal input for the research agent. From them it extracts:
- Observations — the situations, jobs and obstacles that people talk about.
- Direct quotes — separated from the researcher's interpretations.
- Patterns — recurring signals, with the exceptions shown explicitly.
- Contradictions — the places where respondents disagree with each other.
- Hypotheses — formulations for the next test, not finished conclusions.
The chain: pattern → hypothesis → test
The agent's process is built as four stages: Preparation (checking the goal, the segment and the sufficiency of the materials), Coding (extracting observations, situations, jobs and obstacles), Synthesis (grouping the recurring signals with the exceptions shown) and Artifact (an opportunity map with links to the source fragments). The output is a research report with a map of themes, quotes, contradictions, hypotheses and a verification plan.
The key phrase here is "verification plan". A pattern from five interviews does not become a claim about the market; it becomes a hypothesis that has a next step: a new interview, an observation or a product test.
Where the conclusions must remain limited
Four conclusions cannot be drawn from a small sample:
- "The market is such-and-such" — interviews do not prove the market size.
- "The majority of customers think…" — five people are not the majority.
- "The effect is significant" — statistical significance is not reached on such a sample.
- "This is confirmed by users" — only what these specific people said in these specific interviews is confirmed.
The quality of the conclusion depends on the quality of the materials and on the formulation of the research goal — this too is fixed in the passport. A poorly formulated goal produces a map that answers the wrong question.
A practical rule
If after five interviews the team wants to make a big product decision — first turn the patterns into hypotheses and plan the verification. If the decisions are small (what to ask in the next series of interviews, which segments to check) — the opportunity map is enough right now. Honest research differs from selling research not by the absence of conclusions but by the explicit boundary of what the conclusion can claim. About how the opportunity map is built — in the article "The opportunity map".
Questions and answers
How many interviews are enough to start?
Enough to structure the observations and hypotheses; not enough for claims about the market. The agent checks the sufficiency of the materials for the stated goal at the preparation stage.
Can the agent run the interviews itself?
No: the passport states directly that the agent does not replace the researcher's conversation with users. It works with the interviews and notes that already exist.
What to do with contradictions in the answers?
They are not smoothed over: the synthesis shows the exceptions explicitly, and the contradictions go into the report as a separate block — as material for hypotheses.
Why are five interviews too few for statistics?
A small sample does not prove the market size: the agent's passport states directly that the sample is not equal to market statistics.
What can you get from five interviews?
Patterns, quotes, contradictions and hypotheses for the next checks.
How does the agent mark uncertain conclusions?
It separates facts from hypotheses and shows the strength of evidence for each conclusion.
What to do after the first interviews?
Turn the patterns into hypotheses and check them: the interviews are the input for the next study or test.
Does the agent replace the interviewer?
No: it does not replace the researcher's conversation with users, it structures the collected materials.
Which materials does the agent accept?
Interviews, notes, the goal and the segment; the quality of the conclusion depends on the quality of the materials.
How to increase confidence in the conclusions?
Expand the sample and check the hypotheses; the agent's conclusions are material for the researcher's decision, not statistical proof.