Interview the last real purchase, not the imagined next one

A customer interview becomes unreliable the moment the founder asks a stranger to predict the future of a product they have not used. People are generous with hypothetical approval. Their calendars, expense records, workarounds, and abandoned evaluations are less polite and more useful.

The job is to reconstruct a recent decision, not to collect reactions to a pitch. The source chapter's distinction between problem evidence and payment evidence is available at validation chapter.

Recruit around an event

Define a qualifying event: bought a tool in the category, renewed one, replaced one, built a workaround, requested budget and lost, or decided the problem could wait. Put a time boundary on it that the participant can remember.

Do not fill the sample with friendly users. Include at least one person who chose a competitor, one who stopped evaluating, one who lacks budget authority, and one who lives with the problem without buying. GOV.UK guidance says research should cover how users currently complete the task, the frustrations they encounter, and a broad range of users rather than only a typical case (GOV.UK discovery research). This is public-service guidance; the commercial interpretation is ours.

Follow the purchase chronology

Ask for the last real episode and move in order:

1. What happened that made the issue noticeable?
2. What did you do first?
3. Which alternatives entered the list, including doing nothing?
4. Who joined the decision, and what could each person stop?
5. What information, trial, or reassurance changed the direction?
6. What was paid, approved, or postponed?
7. What happened after purchase?

Ask to see ordinary artifacts when appropriate and consented: a redacted comparison sheet, calendar invite, procurement checklist, or old spreadsheet. Contextual-research guidance notes that observing real tasks with ordinary documents and devices can reveal barriers that an interview misses (GOV.UK contextual research).

Fictional interview plan

Hypothetical plan, not a transcript: A founder exploring scheduling software for mobile veterinary practices recruits six practice managers: two recent buyers, one competitor buyer, one team using a shared calendar, one failed evaluator, and one practice that says scheduling is not a priority.

The 35-minute guide spends five minutes on role and context, twenty minutes reconstructing the latest scheduling change, five minutes examining a redacted workflow artifact, and five minutes asking who else should be interviewed. It never demos the proposed product. Evidence is logged as observation, participant interpretation, or founder inference.

Record evidence without upgrading it

Write notes in three columns: what happened, what the participant says it meant, and what you infer. GOV.UK's analysis guidance recommends recording what was seen or heard before interpretation (GOV.UK research analysis). A buyer's explanation is still retrospective; the invoice or workflow does not reveal every motive. The separation simply prevents a founder hunch from entering the file as customer fact.

Decision artifact

After every five interviews, fill a one-page purchase map:

• Trigger and urgency
• Existing workaround
• Alternatives considered
• User, champion, approver, and payer
• Decision criteria and disqualifiers
• Time and money actually committed
• Reason for no decision
• Evidence that contradicts the current thesis

Advance only when a narrow group shows a repeated trigger, a consequential workaround, and a reachable buying path.

Limits

Interviews are selected memories, not transaction logs. Incentives, recall, status, and the founder's wording can distort them. A few recent purchases can clarify a mechanism but cannot estimate market prevalence. Protect personal and commercially sensitive material, obtain consent, and do not retain documents you do not need.