In-Depth Interview Guide

Line-art diagram of a friendly in-depth interview: two people seated across a table, the participant gesturing in a green top, the researcher taking notes in a pale-blue shirt, with a small recorder, a note sheet and a green notebook on the table between them. — The in-depth interview is a controlled, patient conversation. The instruments on the table between the two chairs are almost the entire toolkit.

A survey tells you what changed. An interview, done well, tells you why, how it felt, and what the person would do about it. This guide shows the structure and the probing technique behind that depth. It is built to be adapted to any subject, a technology, a service, a transition.

Five levels down

Most interviews stop at level one or two. The Lab's standard is to reach level four or five on the questions that matter, because that is where the actionable insight lives.

Level	What it captures	Example question	What a survey gets
1, Fact	What happened	"What did you do?"	✓ surveys reach here
2, Behaviour	How they did it	"Walk me through it."	✓ partly
3, Reasoning	Why they did it that way	"What made you choose that?"	✗
4, Meaning	What it means to them	"What did that change for you?"	✗
5, Tension	What they'd trade, fear, or fix	"What's the part you'd never accept?"	✗

The craft is not in the question list. It is in the follow-up, staying on a thread until it reaches level four or five rather than moving to the next item.

The funnel: wide, then deep, then wide again

Opening (5 min), establish ease and consent. Plain-language consent, permission to record, and a genuinely easy first question. The goal is for the person to start talking in their own words before any probing begins.

"Before anything else, tell me a bit about an ordinary day for you, and where [the thing we're studying] fits into it."

Orientation (10 min), map their world. Establish the context the rest of the interview depends on. Who is around them, what the alternatives were, what normal looked like before.

Core (25-35 min), go deep on two or three threads. This is where depth happens. Pick the threads that carry the impact claim and follow each one down the five levels. Better to reach level five on three things than level two on ten.

Lived experience (10 min), the change, felt. What is materially different now; what it cost; what they would change first; what they would never give up.

Close (5 min), the open door. "What haven't I asked that I should have?" routinely produces the single best quote in the interview.

How to get from level two to level five

The echo. Repeat their last few words as a question. Almost no effort; remarkably effective.

Respondent: "…so I just stopped using it after that." Interviewer: "You stopped using it?"

The ladder up (meaning). Ask why something matters, then why that matters, until you hit a value.

"Why was the cost the thing that mattered?" → "Why does keeping the cash matter right now?" → reaches the real constraint.

The critical incident. Don't ask about the general; ask about the last specific time.

Not "How do you usually find it?" but "Tell me about the last time you used it. What happened, start to finish?"

The contrast. Make them compare to the alternative they actually had.

"Compared to how you did this before, what's different?"

The silence. After an answer, wait. The second thing a person says is usually truer than the first.

The fix-it question. Hands them the controls.

"If you could change one thing about it tomorrow, what would it be, and what would that fix for you?"

One question, taken five levels down

A single thread, moving from surface fact down to the meaning and friction underneath. Colour deepens with each rung.

Level 1 · Fact

Q: "Did you start using [the service]?"

A: "Yeah, for a few months."

Level 2 · Behaviour

Q: "Walk me through how you used it."

A: "Mostly at the start of the week, to sort out payments."

Level 3 · Reasoning

Q: "What made you do it at the start of the week specifically?"

A: "That's when the money comes in. If I wait, it's gone."

Level 4 · Meaning (ladder up)

Q: "And being able to sort it the moment the money comes in, what does that change for you?"

A: "I'm not chasing it all week. I know where I stand."

Level 5 · Tension (fix-it)

Q: "What's the part you'd never go back on, and the part you'd fix tomorrow?"

A: "I'd never go back to carrying it all in cash. But the fees, every time I move money, it takes a little. That I'd fix."

What the thread yields that a survey could not

The timing logic (income arrives, must act immediately), the meaning (control, "knowing where I stand"), and the precise friction (per-transaction fees) the person would fix first. Three things a satisfaction score of "7/10" would have hidden completely.

Sampling that answers the question, not the convenience of the researcher

Twenty interviews with the wrong people tell you less than eight with the right ones. Depth research is a sampling discipline as much as a question discipline.

How we choose respondents

Purposive sampling by segment. Not "20 users" but "5 riders who paid off; 5 riders who defaulted; 5 who never adopted; 5 who used to use it and stopped." Each segment answers a different question the headline number cannot.
Deliberate inclusion of the missed. The people the intervention was designed for but did not reach are usually the most informative interview of the study. We sample them explicitly.
Saturation, not a fixed n. We add interviews within a segment until the last three produce no new themes. That is where we stop, whether that is at ten or at twenty-five.
Second-degree recruitment. One respondent introduces the next, especially in informal settings where recruitment through a client list would miss the population the study needs to reach.

The BRW framework enters at sampling too: if the study is measuring a "bypass" of an incumbent system, the respondents most worth interviewing are the ones who tried the incumbent first and switched, and the ones who never tried the niche at all.

Language, phone, group, and sensitive topics

The technique above is a spine. It has to be tuned to the setting.

Adaptations we make routinely

Language-first design. The interview is written in the respondent's language from the start, not translated after English. A translated instrument almost always loses the pragmatic layer that produces level-four answers.
Phone vs face-to-face. Phone interviews are 30-40% shorter, more literal, and worse at silence. Reserve face-to-face for the highest-stakes threads; use phone for reach.
Group settings. Some contexts do not allow one-to-one privacy. A well-run pair or small-group interview trades some depth for social validation of what "normal" looks like in the community.
Sensitive topics. Anonymisation, revocable consent, a trained same-language interviewer, and a plainly-stated permission to pause or refuse any question, are minimums. Some threads simply should not be asked in front of a household or an employer.
The technology in the room. Where the interview is about a device, we ask to see it. What people show, and what they hide, is often more informative than what they say.

What the respondent hears before recording begins

Consent is not a form to sign; it is a conversation to have. Our field researchers use language along these lines, adapted to the setting and the language of the interview.

Sample opening (adapt to context)

"Thank you for meeting me. My name is [name]. I work with Transitions Lab, an independent research team. I would like to ask you about [topic], to understand how it works for people like you. This is a conversation, not a test; there are no right or wrong answers, and you can skip anything or stop at any time."

"I would like to record our conversation, only so I don't have to write while you speak. The recording is used by our team to analyse what people told us. Your name will not be attached to anything we write. If at any point you want me to stop recording, or delete part of what we discussed, you only have to say so."

"Is that all right? And do you have any questions before we begin?"

Consent is asked again, in the same tone, if the interview reaches a more sensitive topic mid-conversation.

What derails depth

Even a well-designed guide can slide back to level one or two in the room. These are the recurring reasons.

What not to do

Leading the answer. "So the price is the main issue?" turns an interview into a survey with a bias. Ask "What has been the main issue?" instead.
Moving on too fast. The temptation after a rich level-three answer is to feel finished. Level four is the next question, not the next topic.
Filling the silence. New researchers often speak into a pause. The pause is the interview working. Wait.
Confusing rapport with agreement. A respondent may reasonably say "yes" because you seem to want a "yes". Neutral acknowledgement ("thank you", "I see") beats visible agreement.
Skipping the disconfirming case. The respondent whose experience contradicts the emerging pattern is the most important one to interview, and the most tempting to under-code.
Asking about the general. "How do people usually…?" produces stereotypes. "Tell me about the last time…" produces evidence.

From transcript to evidence

Each interview is coded and synthesised so individual depth becomes a pattern across respondents.

From the conversation	Becomes
Recurring meanings ("control," "knowing where I stand")	Themes that explain why impact happens
The fix-it answers	A ranked list of friction, in the users' own priority
Critical incidents	Concrete, quotable evidence of the moment of change
Who never reached level four	A signal of who the project is not working for
The contrasts to the old way	The genuine before-and-after, in lived terms

The coding workflow, in order:

Verbatim transcription in the language of the interview. Translation happens only at the coding step, not before, so the original phrasing is preserved for later review.
First-pass coding. Every passage tagged with the level (1-5), the theme, and any critical-incident or fix-it flag.
Second-pass reconciliation by a second researcher on a sample of transcripts, to catch coder drift.
Cross-respondent synthesis. Themes and their weights across the sample. Disconfirming cases named explicitly, not smoothed over.
Quotable-passage selection. Anonymised, permission-cleared, faithful to the speaker's meaning.
Handover to the quantitative reading. Depth findings sit against the impact-tracking template's baseline / midline / endline numbers, so the "why" explains the "what".

Depth is not the opposite of rigour. Coded properly, twenty deep interviews produce findings as defensible as a survey, and they explain the survey's numbers rather than merely restating them.

Informed consent, recorded and revocable. Anonymisation by default. We interview to understand, not to confirm, and we report the threads that contradict the project's hopes as carefully as the ones that support them. The same rigour that goes into the Lab's published research goes into every interview we run for others. Full ethics references are set out on the Field Research page.