Customer interviews

An interview collects evidence about what people did. It does not collect their opinions about your ideas. Every rule on this page follows from that difference.

The core rule

Ask about what happened, never about what someone would do. A specific occasion that a person recalls is evidence. A preference, a prediction and a feature request are not evidence, however confidently the person offers them.

Build each question from the last answer. When you read the next question from your list instead of following what the person just said, you start to collect opinions again.

Methods

Pick the method from what you want to learn, not from the one you know best. Write down which method you picked and why before you write any questions.

When more than one method fits, run the cheapest one that could change your mind first.

Switch interviews

A switch interview reconstructs one real decision: a purchase, a cancellation, a tool swap, or a workaround the person built by hand. Fix exactly one switch before you start. Reject hypotheticals and averages. There is no average switch.

Walk the timeline backwards from the moment the person committed:

A switch happens when push and pull beat anxiety and habit. Record a quote for each of the four forces:

People rarely volunteer anxiety and habit, and a product can often act on them. Ask directly: "What nearly stopped you?" Mark every force that you inferred instead of heard. An inferred force is a question for the next interview, not a finding.

Finish with a job story in one sentence: When [situation], I want [motivation], so that [outcome]. The situation is a circumstance, not a demographic. The motivation is progress, not a feature.

Contextual inquiry

Watch the work in the place where it happens, on real tasks. Act as an apprentice, not an auditor. Interrupt rarely, and only to ask what just happened. Never suggest a better way.

The findings are in what people do without comment: the second spreadsheet, the message to a colleague, the field they always leave blank, the step they redo. Count every wait and every workaround. At the end, play back what you saw and let the person correct you. The corrections are often the most useful result.

Watching beats asking when you want to know how work is done. People cannot describe workarounds they no longer notice.

Critical incident

When you cannot watch, ask for one specific occasion in order: "Walk me through the last time that happened." Stay on that day. Ask where they were, what they had just done, who they contacted, and what they did when it failed. When the person moves to the present tense or says "usually", take them back to the occasion.

Problem interviews

A problem interview tests whether a problem is real before anybody builds for it. Ask about the last time the problem affected them, what it cost, and what they did instead. Then ask what they have already tried and what they paid for it in money, time or workarounds.

Never pitch, and ignore compliments. "That sounds useful" tells you almost nothing. A person who built a spreadsheet to work around the problem has told you more than an hour of enthusiasm.

Recruiting

• Recruit by behavior, not by title. "Registered as a supplier in the last month" finds evidence. "Procurement manager" finds opinions.
• Talk to the people who do the work. A buyer or a manager can describe the organization's story. You need the story of the person who chose or who struggled.
• Keep a standing pipeline. Ask customer-facing colleagues to book one or two conversations every week. A team that recruits only when it has a question interviews too late.
• Include people who left or never started. People who cancelled or stalled explain the forces that held them back.

Interview count

One interview is a story, three are a coincidence, and ten are a pattern. Ten to fifteen interviews usually cover the patterns in one customer segment. When you share a finding, say how many interviews support it.

Interview every week, in small numbers, instead of once a quarter in a large batch. Teresa Torres calls this continuous interviewing. Weekly contact keeps the team's assumptions close to current evidence.

Guide template

Keep the guide to one page. Answer every question by recalling an event. If a question can be answered with "it depends", it is a topic, not a question. Put the easiest recall first, and mark the questions you would cut if the call runs short.

Interview guide: <topic>
Method: <switch | contextual inquiry | critical incident | problem>, because <reason>
Recruited for: <the behavior, e.g. "registered as a supplier in the last 30 days">
Recording: permission asked at the start

1. Warm-up: "What is your role, and what did you work on this morning?"
2. The occasion: "Tell me about the last time you <did the thing>."
3. Before: "What were you doing just before that? What made you start?"
4. The hard part: "Where did it slow down or go wrong? What did you do then?"
5. Workarounds: "What have you tried to make it easier? What did that cost?"
6. Anxiety and habit (switch only): "What nearly stopped you?"
7. Close: "What else happened that week that I should know about?"
   (cut if short: 5, 7)

After the call, within one hour:
- Job story: When ___, I want ___, so that ___.
- Quotes: <verbatim, with timestamps>
- Heard vs inferred: <mark each finding>
- Surprises: <what contradicted what we believed>

Record the interview with permission. You cannot follow the last answer and take good notes at the same time.

Synthesis

Write up each interview within an hour, while you still remember it. Keep the person's words. For each interview, write:

1. The job story.
2. The opportunities: needs, pains and desires, each with the quote it came from.
3. What you heard and what you inferred, marked separately.
4. What surprised you.

Across interviews, group the opportunities and count how many people raised each one. The opportunity solution tree is the structure for this step.

Watch for these signs of weak evidence:

Prepare and review interviews with your agent

Ask your agent to pick the method, draft the guide, or review your transcript for opinion questions and inferred forces. Roko Platform gives the agent the method table, the rules and the signs of weak evidence on this page over MCP.

We want to learn why suppliers abandon registration halfway. Pick the interview method, say why, and draft a one-page guide. Then list the recruiting criteria as behaviors.

Review the interview transcript in notes/interview-07.md. List every question that asked for an opinion or a prediction, every force I inferred rather than heard, and the job story.

Connect over MCP covers setting up the agent. One interview produces one person's story. To structure many stories, continue with opportunity solution trees.