Should I survey or interview customers first?

A sequence that tends to hold up: interview first to discover categories and vocabulary, survey second to size what the interviews found, then interview again to interrogate any surprising survey result. Reversed, the survey encodes your guesses before evidence arrives.

When are surveys better than interviews?

When the categories are already known and the missing fact is prevalence — how common a complaint is, which named option ranks first, whether sentiment has moved between waves. A survey can confirm or refute suspected shapes; it rarely reveals new ones.

How many customer interviews are enough?

No fixed number holds across contexts. Within one reasonably homogeneous segment, continue until new conversations mostly repeat earlier ones — saturation — and treat each new segment as a fresh count. Recruiting from a single channel can fake saturation early.

Do surveys and interviews measure actual behaviour?

No — both collect stated data, which relates to but never equals what people do. For some jobs, behavioural traces such as search activity or the questions buyers put to AI assistants offer a sturdier signal than asking.

Surveys vs interviews for customer research

Guide · Buyer Research & Comparisons · 4 min read · last verified 2026-07-27

Reviewed before publication Editorial board Independent commercial review

In shortSurveys count how many people say a thing; interviews uncover why. Each misleads in its own direction, so this piece maps where each fits, how many interviews are enough, and when neither is the right tool.

Surveys and interviews are both ways of asking customers questions, and the choice between them comes down to the kind of answer you need: a survey counts how many people say a thing; an interview uncovers why a particular person says it. A large share of research failures trace back to using one method for the other's job.

Both, it is worth stating early, collect stated data — what people say, which is related to but never identical with what they do. That limit is shared. Nearly everything else about the two methods diverges.

Two instruments, two kinds of answer

A survey asks fixed questions with mostly fixed options, at scale; its output is a distribution — how answers spread across a population you chose. An interview is an open-ended conversation with room to follow surprise; its output is mechanism — the sequence of a decision, the trigger behind it, the words a buyer reaches for when no options are offered. One tells you how common. The other tells you how it works.

When a survey is the right tool

Surveys fit jobs where the categories are already known and the missing fact is prevalence. How widespread is this complaint among current users? Which of these named options do customers rank first? Has sentiment moved since the last wave? They also reach people who would never accept a call, and they produce comparable results wave over wave — provided the wording holds still.

The precondition is easy to miss: a survey can only confirm or refute shapes you already suspected. Its questions and answer options encode your assumptions before the first response arrives, and nothing a respondent does inside the instrument can escape those choices.

When an interview is the right tool

Interviews fit jobs where you do not yet know the categories. What actually triggered the purchase? What nearly stopped it? Who else joined the decision, and what did they veto? What language does the buyer use for the problem when nobody hands them a vocabulary? Interviews are also the tool for investigating surprises — a churned account, a lost deal, a survey result that refuses to make sense.

Interviewing well is its own discipline, and it is easy to contaminate answers through the way a question is phrased; the craft is covered in How to interview customers without leading them. The method choice comes first, though — technique cannot rescue a method mismatched to the job.

How each misleads, in opposite directions

Surveys flatten. Rich, contradictory situations get forced into the options you offered. Wording can steer answers in ways respondents rarely notice. People tend to satisfice near the end of a long instrument, and the people who respond at all often differ from the people who did not. The result carries a false authority: distributions built from forced choices still produce confident-looking figures.

Interviews amplify. The articulate and the enthusiastic tend to dominate what the team remembers. The interviewer often hears confirmation more readily than contradiction. A handful of conversations gets generalised into 'customers want', and the most vivid quote outlives ten quieter ones that pointed elsewhere.

Neither method is neutral, and the distortions do not cancel out — they mislead in different directions, which is exactly why the two pair well when sequenced deliberately.

How many interviews are enough

Confident sample-size rules circulate widely; we decline to add one, because the defensible answer is conditional. Within a single, reasonably homogeneous segment, continue until new conversations mostly repeat what earlier ones surfaced — the point researchers call saturation — then stop, and treat a new segment as a fresh count, since saturation in one group establishes nothing about another. Watch for false saturation too: recruiting every participant from the same channel produces repetition that reflects the channel, not the market.

A sequence that puts them in order

A pattern that tends to hold up: interview first, to discover the categories, triggers, and vocabulary; survey second, to size what the interviews found; then return to interviews to interrogate whatever the survey made surprising. Run in reverse, the sequence bakes guesses into the instrument — the survey ends up measuring your assumptions with impressive precision.

When neither is the job's tool

Some questions are better answered by behaviour than by asking anyone anything. Awareness is one: what people claim to recognise and what they actually reach for can diverge, and How to measure brand awareness without surveys covers that replacement in full. The questions buyers put to AI assistants form another behavioural trace — a record of demand phrased in the buyer's own words, which Magrios treats as research input alongside, never instead of, talking to customers.

Wherever the findings land — an ideal customer profile, a persona, a positioning document — the standard stays the same: artefacts built on evidence rather than invention, a line drawn sharply in Buyer personas: evidence or fiction.