Six conversations are a beginning, not a market size.

Choose a question a conversation can answer

A founder may have time for six customer conversations before deciding whether to improve an offer. That can be useful work, provided the question fits the method. Ask how a buyer completed a recent task, what caused uncertainty, or which information was missing. Do not ask the conversations to establish the proportion of all businesses that will buy. GOV.UK’s user research guidance recommends small rounds for qualitative methods and makes participant numbers dependent on the method. The practical lesson is to match evidence to a decision. Six is an illustrative planning choice here, not a universal rule for research sufficiency.

Recruit for relevant differences

Write a recruitment brief before inviting familiar contacts. Define the task or purchase experience that makes someone relevant, then deliberately include differences that might change the answer. A small property-services study could seek a person who bought, a person who deferred and a person who chose an alternative. Those are examples of contrast, not a representative sample design. Keep a simple record of whom the study could not reach. If every participant comes from the founder’s professional circle, say so. Convenience is sometimes unavoidable, but hidden convenience becomes false confidence when the findings are presented to management.

Ask about a specific recent episode

Build the conversation around a particular episode. Ask the participant to describe what happened, what they tried first, who else was involved and how the decision ended. A relevant document or screen can help them reconstruct the sequence, provided they choose what to share and sensitive details are protected. Ask neutral follow-up questions instead of praising answers that support your idea. A person who says they love an imagined service has not demonstrated a purchase. Recent behaviour, workarounds and actual decision constraints usually provide a more useful starting point for designing the next test than an abstract statement of enthusiasm.

Separate the observation from the interpretation

Keep three separate columns in the notes: what was observed or reported, the team’s interpretation, and the decision it might affect. For example, a buyer reports calling twice to confirm delivery. The observation is the repeated call; the interpretation might be uncertainty about timing; the proposed change could be a clearer confirmation message. That interpretation still needs checking. Record whether a statement came from a participant, an interviewer’s inference or another source. Obtain appropriate permission for recording and quotations, minimize personal information, and share only what the team needs. An AI summary should not convert a tentative interpretation into a factual finding.

Look actively for the awkward case

After a round, search for the case that does not fit the emerging explanation. If some buyers needed reassurance and another wanted speed, the answer may depend on urgency, experience or a different buying role. Keep those possibilities visible rather than averaging them into an invented universal customer. Reporting that four of six participants raised an issue is a description of this group, not evidence that two-thirds of the market has that issue. Show the participant mix and the unanswered questions alongside any count. Invite a colleague who did not conduct the interviews to challenge the proposed explanation.

Make the next test proportionate

Choose a next step that responds to the remaining uncertainty. A clearer page can be tested with another small group; a claim about market prevalence requires a different research design; willingness to pay may need a properly scoped offer and observed buyer decisions. Keep the first change reversible where possible. Write a short learning note containing the question, recruitment approach, evidence, limitations and next action. This turns a handful of conversations into disciplined progress. The objective is not to make a small study sound large. It is to learn enough to choose the next useful question without overstating what the evidence can support.