A survey comes back with 30% of respondents saying they're "dissatisfied with the price." So we should just lower it, right? It's never that simple. Behind "dissatisfied with the price" sit completely different reasons all tangled together: "the value never got through," "they compared us to a competitor," "they never really learned to use the product." Numbers tell you what happened. They don't tell you why.
The method for digging into that why is the in-depth interview (IDI). One-on-one, 60 to 90 minutes, you descend through the person's words and silences toward motivations they themselves haven't put into language. Do it wrong and it becomes "a ritual for confirming your own hypotheses." Do it right and you surface discoveries you never saw coming. This guide walks through designing the interview guide, the deep-dive technique called laddering, how to ask without leading, how to judge when you've talked to enough people, and how to analyze the recordings — all with a practitioner's feel for the work.
1. What in-depth interviews are actually for
An in-depth interview is qualitative research that goes deep with a single participant. Its purpose differs fundamentally from a survey (quantitative).
- Survey (quantitative): measures "what and how much" across many people. Its strength is representativeness and generalization.
- In-depth interview (qualitative): digs deep into "why and how" with a few people. Its strength is understanding context and motivation.
This division of labor is the same one laid out in quantitative vs. qualitative research methods: neither is superior — they make different things visible. When a survey shows "30% price dissatisfaction," you interview to dig into the why. Conversely, a hypothesis surfaced in an interview gets validated with a survey to learn "how widely it spreads across the whole." That back-and-forth is the logic of mixed methods.
How it differs from focus groups (FGI)
Qualitative work has another format: the focus group (FGI), a 6-to-10-person group discussion. The guiding principle for choosing between them:
- In-depth interview (IDI): sensitive topics, individual deep psychology, themes where people clam up when others are watching.
- FGI: when you want to see group chemistry, the spread of ideas, and the diversity of opinions all at once.
If you're asking about "things people find hard to say in front of others," IDI is the only choice.
2. Designing the interview guide — a map, not a script
In-depth interviews are run semi-structured by default. Not a fully scripted (structured) interview, and not a no-plan (unstructured) one. The style is to fix only the rough flow and the points you must cover, then dig deeper as the conversation moves.
Kallio et al. (2016) systematically reviewed how semi-structured interview guides are developed and concluded that a guide should be built from "main themes grounded in prior knowledge plus follow-up questions."
The basic structure of a guide
A good interview guide roughly follows this arc:
- Opening (5 min): introduce yourself, get consent to record, reassure them that "there are no right answers" and to "speak frankly." Small talk to ease the tension.
- Warm-up (10 min): start with easy factual questions ("How do you usually use it?"). Don't go for the core right away.
- Main body (30 to 50 min): three to five main themes. For each theme, draw out a "concrete episode."
- Deep dive (as needed): "Why is that?" "What were you feeling in that moment?" — move from the surface down to motivation.
- Closing (5 min): "Is there anything else you wanted to add?", then thank them.
Design pointers
- Hold it to 10 to 15 questions: even in 60 to 90 minutes, once you start digging there aren't many themes you can fully work through. Cram more in and everything gets shallow.
- Ask through chronology and concrete episodes: "Tell me, from the very start, about the last time you did X." Memories of concrete behavior surface more truth than abstract opinions.
- Don't make it a place to confirm hypotheses: a guide is "what you want to ask," not "what you want them to say."
3. Laddering — stacking "why?" to descend toward values
The single most powerful technique in in-depth interviewing is laddering. Starting from a surface-level "attribute," you repeatedly ask "why does that matter?" and ultimately descend to that person's values.
Reynolds & Gutman (1988) systematized this as "means-end chain" theory. You dig through three tiers — "Attribute → Consequence/benefit → Value" — climbing them like the rungs of a ladder.
A laddering example
"I go with a dark roast for this coffee." (Attribute) —— Why a dark roast? "Because it has a really solid bitterness." (Attribute) —— And what's good about a solid bitterness? "It shakes off the drowsiness so I can focus on work." (Consequence) —— And when you can focus, what's good about that for you? "I get results in the morning, and that becomes confidence that I'm good at my job." (Value)
From the surface — "I like dark roast" — we descended to the value of "self-efficacy." Only when you've dug this far can you see the role the product plays in the customer's life. Both your messaging and the direction of product improvements hit the mark only when they grasp the value tier.
Cautions on laddering
- A barrage of "why?" turns into an interrogation: nothing but "why?" makes people feel cross-examined. Mix in rephrasings like "What does that look like, concretely?" or "Tell me a bit more about that."
- When they stall, step back down: people often haven't put the value tier into words themselves. Don't fear silence — wait. If they stall, step down a rung and ask again.
4. Asking without leading — silence and neutrality
The quality of an interview is decided by how you ask. With the same guide, the information you get can differ completely depending on the interviewer. The biggest enemy is unconscious leading.
Ways of asking you must avoid
- Leading questions: "This feature is convenient, isn't it?" → the person can only say "yes." Stay neutral: "How did you feel about this feature?"
- Forcing a binary: "Which do you prefer, A or B?" → this crushes "neither" and "both." Let them speak freely first.
- Jargon and internal terms: words the person doesn't know breed intimidation and misunderstanding.
- Talking too much yourself: aim for the interviewer to do 20 to 30% of the talking. Let the other person do 70 to 80%.
These are the same principles as the leading-question and double-barreled avoidance in the complete guide to survey question wording, but face-to-face they're harder to hold to on the fly.
Techniques that work
- Don't fear silence: when the other person falls silent thinking, don't rush to fill it. The truth often comes right after a silence.
- Mirroring: repeating their words back — "…you can't focus, then" — prompts them to keep going.
- Prompt for episodes: "Concretely?" "When did you last feel that way?" pulls abstraction back to the concrete.
- More than a nod, less than an opinion: prompt with "I see" or "and then?", but never inject your own judgment ("nice" or "that's wrong").
Social desirability bias (answering to look good) shows up more strongly face-to-face. For that mechanism, see also designing around social desirability bias.
5. How many people are enough — the idea of saturation
"How many in-depth interviews should I run?" There's no sample-size calculation like in quantitative work. Instead, you judge by the concept of saturation. Saturation is the state where talking to a new participant no longer surfaces new themes or discoveries.
Guest et al. (2006) analyzed 60 interviews and demonstrated that the bulk of the main themes were exhausted within the first 12 participants (around 80% of the codes appeared by the sixth). That's for "a relatively homogeneous group with clear themes," but it's often cited as a starting point for thinking about qualitative sample sizes.
Practical rules of thumb
- Homogeneous target, clear themes: you approach saturation at 6 to 12 people.
- Spanning diverse segments: a few people per segment × (number of segments). The total can reach 15 to 30.
- Judge by "saturation," not by "a number": don't lock in on 12. Decide whether to continue or stop based on "are new discoveries still coming out?"
This is a different animal from quantitative representativeness. "I interviewed 30 people, so this represents the whole" does not hold. Saturation is "coverage of themes," not "representation of a population" — keeping these from blurring together is critical. If you need to generalize to the whole, validate the hypotheses you found with a survey. For narrowing down participants, see the screening question design and operation guide.
6. Analysis — how to turn recordings into "findings"
An interview isn't "done once you've done it." It only becomes data once you transcribe the recording and structure it. Skip this step and only "the remarks that stuck with you" survive in memory, opening the door to convenient interpretation.
The basic analysis steps
- Transcription: turn the recording into verbatim text. AI transcription has recently cut this effort dramatically.
- Coding: attach "tags (codes)" to remarks — "price dissatisfaction," "considering switching," "feature discovery," and so on. A technique rooted in Glaser & Strauss's grounded theory.
- Theme extraction: group the codes and find recurring themes (patterns).
- Interpretation: read the relationships among themes and describe the "why" as a structure.
Using AI, and its limits
The first-pass processing of transcription and coding can be made dramatically more efficient with LLMs in recent years. That practice is covered in analyzing open-ended responses with AI. But the final interpretation is human work. AI can pick up "what was said," but it can't read the meaning of "what went unsaid, the silences, the hesitations."
7. The editorial take — 5 things you must not do in in-depth interviews
From the vantage point of someone who continuously tracks industry cases and the voices of practitioners, here are five accidents that recur in in-depth interviews.
1. Turning it into a hypothesis-confirmation exercise
The most frequent and most damning accident. An interview where you go around getting people to agree with your own hypothesis — "Right, that's how it is, isn't it?" This isn't a place for discovery; it's a ritual of self-satisfaction. The value of an interview lies in "it wasn't what I thought." Go in with a posture that welcomes disconfirmation. If all you get is agreement, suspect that you're leading.
2. The interviewer talks too much
Out of nervousness or fear of silence, you end up explaining, leading, and supplying the answer yourself. The one who should talk is the participant; the interviewer does 20 to 30%. Silence is precious time for the other person to think. Don't fill it.
3. Asking for "opinions" instead of "behavior"
Answers to "What do you think?" are usually polite fictions or idealized talk. Ask "What did you actually do recently?" — for concrete behavior and episodes — and the truth comes out. People can lie in their opinions, but they rarely lie in the memory of their behavior.
4. Skipping transcription and coding
You jot down "the remarks that stuck with you" and feel like you've analyzed. This is a breeding ground for confirmation bias — only the remarks you wanted to hear survive in memory. However tedious, transcribe verbatim, attach codes, and view the whole as a structure. AI transcription can lower the burden substantially.
5. Confusing saturation with representativeness
Generalizing — "I asked 20 people, so this is the voice of all customers." Qualitative work is a method of depth; it does not carry quantitative representativeness. If you want to say "what percentage of the whole has this problem," validate the hypothesis you found in interviews with a survey. Don't confuse the roles of qualitative and quantitative right to the end.
8. How the survey tool Kicue relates to in-depth interviews
To be honest, running the in-depth interviews themselves is outside Kicue's scope. Kicue is a survey (quantitative) tool, and it has no features for recording, transcribing, or coding interviews.
That said, where Kicue can contribute to qualitative research is before and after the interview.
- Recruiting participants (before): build a screening survey in Kicue to gather interview candidates. Filter for people who meet your criteria and shortlist interview candidates (screening question guide).
- Quantitative validation of hypotheses (after): take the hypotheses you found in interviews and validate "how widely they spread across the whole" with a Kicue survey. This is how you implement mixed methods.
- Linking via respondent ID: identify, by ID, the survey respondents you'd "like to hear more from," and turn that into interview invitations.
⚠️ What Kicue can't do
- No interview hosting, recording, or video-call features: run interviews on Zoom / Google Meet / in person, and record with a dedicated tool.
- No transcription or auto-coding: use a dedicated AI transcription service for transcripts, and external tools or manual work for coding — the kind covered in analyzing open-ended responses with AI.
- No incentive payment processing: handle interview incentives through an external payment or gift-delivery service.
As related reading, pairing quantitative vs. qualitative research methods, the mixed-methods research design guide, analyzing open-ended responses with AI, the screening question design and operation guide, and the complete guide to survey question wording reveals the full picture of the research rhythm: "discover hypotheses qualitatively, validate them quantitatively."
Summary — 6 points for making in-depth interviews a place of discovery
- A method for digging into "why" — if quantitative is "what," qualitative is "why and how." Different roles.
- The guide is a map, not a script — go semi-structured. Fix only the points; dig deeper through dialogue.
- Ladder down to values — attribute → consequence → value. Stack the "whys," but don't make it an interrogation.
- Don't lead, don't over-talk — the interviewer does 20 to 30%. Don't fear silence; ask about behavior, not opinions.
- Judge by saturation, not by a number — the main themes tend to be exhausted at 6 to 12 people. Saturation is not representativeness.
- Structure the recording and analyze it — transcription plus coding. Impression notes are a breeding ground for confirmation bias.
In-depth interviews look like the easy method of "just listening," but they're a discipline of self-restraint: don't lead, don't over-talk, dig into behavior. You unearth the "why" a survey can't capture, then validate that hypothesis with a survey — this back-and-forth between qualitative and quantitative is what raises the precision of decision-making by a notch.
If you want to recruit participants for in-depth interviews or run a quantitative validation of the hypotheses you got from them, why not try the free survey tool Kicue? From screening surveys that narrow down participants, to a main survey for validating hypotheses, to shortlisting follow-up interview candidates by respondent ID, you can get started on the quantitative side that supports qualitative research — before and after — with a single account. (Running, recording, transcribing, and coding the interviews themselves happens in combination with Zoom / a dedicated transcription service / an analysis tool.)
References (4)
- Guest, G., Bunce, A., & Johnson, L. (2006). How Many Interviews Are Enough? An Experiment with Data Saturation and Variability. Field Methods, 18(1), 59-82.
- Kallio, H., Pietilä, A.-M., Johnson, M., & Kangasniemi, M. (2016). Systematic methodological review: developing a framework for a qualitative semi-structured interview guide. Journal of Advanced Nursing, 72(12), 2954-2965.
- Reynolds, T. J., & Gutman, J. (1988). Laddering Theory, Method, Analysis, and Interpretation. Journal of Advertising Research, 28(1), 11-31.
- Glaser, B. G., & Strauss, A. L. (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Aldine.