A customer interview is a deliberate, structured conversation with a single customer to understand their needs, behaviour, and reasoning in their own words. It belongs to qualitative research, which trades breadth for depth: instead of a thin signal from thousands, you get a rich picture from a handful. Done well, it answers questions no survey can.
Why interviews, not just surveys
Surveys and metrics are superb at measurement. They tell you what is happening and how often. But they are poor at explaining motivation, because a closed question can only return the options you already thought of. When you genuinely do not know why customers behave a certain way — why they abandon onboarding, why a feature goes unused, why they chose you over a competitor — an interview lets the customer surprise you with an answer you never would have put on a survey. The richest research programmes use both: quantitative methods to find the patterns, qualitative interviews to understand them.
How to run one well
Recruit the right people: talk to customers who actually have the problem you are studying, not whoever is easiest to reach. Keep the sample purposeful, not random — five interviews with the right segment beat fifty with the wrong one. Prepare a loose guide of themes rather than a rigid script, so the conversation can breathe. During the interview, your job is to listen: the customer should be doing most of the talking. Ask open questions, stay silent long enough for them to fill the gap, and chase specifics with "why" and "can you give me an example?".
The golden rule is to anchor everything in real past behaviour. "Walk me through the last time this happened" produces honest, concrete data. "Would you use this?" produces flattery. Rob Fitzpatrick's "The Mom Test" built a whole method around this distinction: ask about the customer's life and past actions, never about your idea, because people lie to spare your feelings — not maliciously, just kindly.
Common Misconceptions
Most people think
"I need a large, representative sample for interviews to count."
Actually
Interviews are not meant to be statistically representative. Their job is
to explain and to surface the unknown. A handful of well-chosen
conversations can reveal a problem that a thousand-person survey would
miss, because the survey never thought to ask.
Most people think
"If customers say they would buy it, that validates the idea."
Actually
Stated intentions are weak evidence. People routinely say they will buy
things they never purchase. What they have actually done in the past — paid
for, worked around, complained about — is far more reliable.
Common mistakes
The most damaging mistake is leading the witness. Pitching your solution, asking yes/no questions, and nodding along until you hear what you wanted turns the interview into a mirror. A second mistake is talking too much: every minute you spend explaining is a minute the customer is not revealing something. A third is treating a few interviews as proof — they generate hypotheses and understanding, which you then test at scale with quantitative methods. Interviews tell you what might be true and why; surveys tell you how common it is.
There is also a newer way to make a stack of interviews readable at a glance. The bottleneck with qualitative research has always been that someone has to actually read it — and no executive is going to sit through forty transcripts. So once the interviews are transcribed, AI can code each one against a fixed set of analysis questions — did they mention price, what job were they hiring the product for, where did they get stuck — and, for each answer, pull out the exact transcript fragments that support it. Every interview becomes one structured record. The payoff is an executive reading of the qualitative: a filterable, quantitative view across every conversation that a leader can scan in minutes, while anyone who wants the detail can still click straight through to the real quotes behind any finding. The structure lets you count and compare; the attached fragments keep you honest, because every number traces back to something a customer actually said. It is coding, just faster — and with a human reviewing, the rigour stays intact.
In practice, the two feed each other in a loop. Your survey and feedback data flag a pattern worth understanding. Because tools like Dayalogs keep the open comments attached to scores and segments, you can see which customers raised a recurring theme and reach out to exactly those people for an interview. The conversation then explains the pattern, and the fix you ship gets measured by the same metrics that flagged it. Used this way, interviews are not a separate activity bolted onto research — they are the part that turns a number into an understanding.