Customer discovery interviews with an AI interviewer
To run customer discovery interviews with an AI interviewer, start from the decision you need to make, write a brief with the goal and topics, and invite your own customers by link. The interviewer asks about real past behavior rather than opinions about your idea, and you get transcripts, summaries and quotes with speaker and time, optionally with findings measured against your research question.
Start from a decision, not a list of questions
Customer discovery interviews pay off when they change a decision: which customer problem to solve, why people choose or reject your offer, or what to clarify before you change a product or service. Write that decision down before anything else. It becomes the research question, and it tells you when you have heard enough.
A good research question is narrow and concrete, for example: “Why customers cancel their subscription in the second month.”
Write the brief instead of a script
With an AI interviewer, the brief replaces the interview script. You do not decide the order of questions; the interviewer picks each next question from what the customer says. The brief tells it what matters:
- Goal: what you need to understand, in one or two sentences.
- Topics: the areas the conversation must cover, such as the moment they started looking, what they tried before, and what made them stay or leave.
- What counts as enough: the evidence that would answer your question.
- Supporting material: documents the interviewer may use to prepare. Only permitted material is used.
Ask about the past, not the future
People are poor at predicting what they would do and generous with compliments about your idea. The approach popularized by The Mom Test is simple: ask about what customers actually did. Narratia’s interviewer is built around this habit. It asks for a real case and is designed to avoid leading questions.
| Instead of | Ask |
|---|---|
| “Would you use a feature that…?” | “The last time you needed this, what did you do?” |
| “Do you like our product?” | “Why did you choose it? What were you using before?” |
| “How much would you pay for this?” | “What does this cost you today, in money or time?” |
| “Why do customers cancel?” | “Walk me through the week you decided to cancel.” |
Invite your own customers
Customer discovery works best with the people you already serve, including the ones who left. Send each person the interview link by email or messenger. They answer by voice or text in English, Spanish or Russian, and they read the participant notice before the conversation starts: an AI runs it, what is recorded, who receives the answers and how long they are kept.
Choose participants on purpose. Recent buyers, people who cancelled and people who evaluated you and chose something else will each tell you a different part of the story.
What the conversation looks like
In a demo interview with a test persona, a coffee shop owner explained that catering inquiries arrive through four channels and nobody owns checking them. The interviewer asked her to walk through one missed inquiry, then reflected back what it heard. When it assumed all three late inquiries were lost, she corrected it:
only the one i know for sure we lost. the other two i did get back to eventually, just late
The interviewer acknowledged the mistake and updated its picture, and the correction became part of the result instead of being smoothed over. This is what an adaptive interview adds to customer discovery: the details, the numbers people are unsure of and the corrections all stay in the record.
Read the results with the quotes attached
You choose the result when you set up the study: transcripts and summaries for your own analysis, or the same plus findings measured against your research question. Either way, findings rest on verbatim quotes with speaker and time.
- Look for repeated triggers, workarounds and the exact words customers use for their problem.
- Check any surprising finding against the transcript before you act on it.
- Treat material marked as preliminary as unconfirmed.
- If the interviews do not answer your question, Narratia reports a partial result rather than an answer it cannot support. That is a signal to run another round, not a failure.
Decide what happens next
The interviews inform your decision; they do not make it. Typical next steps are deciding which questions to check next, which part of the offer to clarify, and whom to interview next and about what. A second round with a sharper brief often answers what the first round only raised.
Common mistakes
- Pitching your idea during the interview, which invites compliments instead of facts.
- Asking about hypothetical features instead of past behavior.
- Interviewing only happy customers.
- Writing summaries without the quotes that support them.
- Stopping at the first plausible answer.
Weighing whether interviews are the right method at all? See AI interviewer vs survey vs human interviewer, or read how an AI interviewer works.
Frequently asked questions
How many customer interviews do I need?
There is no fixed number. Keep going until new conversations stop changing the picture. Narratia assesses whether the research question is adequately answered and reports a partial result when evidence is missing.
Should I describe my product idea during the interview?
No. Ask about what customers did in the past. Describing your idea invites polite compliments, which tell you little about real behavior.
Can I interview customers in Spanish?
Yes. Spanish speakers talk to Camila, English speakers to Sam and Russian speakers to Anya, by voice or text through a link.
Who should I interview?
Your own customers and users: recent buyers, people who cancelled and people who chose an alternative. Narratia works with your own contacts invited by link or messenger; other ways of finding participants are agreed separately.
What do I get from the interviews?
Transcripts and summaries with quotes tagged by speaker and time, and optionally findings measured against your research question.