In-Depth Interviews: getting past the first answer

Most IDIs are semi-structured: an interview guide fixes the territories to cover while leaving order, wording, and depth to the conversation. This is the right default — comparable coverage across participants, freedom to follow the unexpected.

Unstructured interviews (a single opening invitation, then following the participant) suit narrative and exploratory work where imposing topics would destroy the data. Structured qualitative interviews (fixed wording, fixed order) buy comparability for multi-interviewer teams at the cost of depth, and usually signal that a survey was the honest instrument.

The standard methodological orientation for health and social research is DiCicco-Bloom and Crabtree’s The qualitative research interview; the fullest treatment of the craft and its epistemology is Brinkmann and Kvale’s InterViews (SAGE) — source of the useful framing that the interviewer is a traveler constructing knowledge with the participant, not a miner extracting it.

• Territories, not scripts: five to eight topic areas with a natural arc — opening ease, core depth, sensitive material late, future/wrap-up last.
• Episodic anchors: for each territory, a question that asks for a specific occasion (“tell me about the last time you…”) — concrete episodes carry detail and defeat generality.
• Probe banks, prepared: silence, echo (“punishing?”), elaboration (“walk me through it”), contrast (“how did that compare with…”), and the devil’s-advocate probe for testing strength of view.
• No leading constructions: “how satisfied were you…” presumes; “what was that like?” does not.
• Pilot and cut: a guide that cannot be covered in the time with two probes per answer is a questionnaire in disguise; pilots exist to force the cuts.

IDI samples are purposive: chosen for relevant experience and for the variation the question needs (roles, sites, positions in the process), not for representativeness. Deviant and boundary cases — the churned customer, the exception patient — earn their slots by testing emerging patterns.

On numbers, the honest answer is that adequacy depends on the question’s scope, the sample’s heterogeneity, and the analysis’s ambition. Well-known empirical work on saturation found most codes present within the first dozen interviews of a homogeneous sample; comparative designs across segments multiply that. The defensible practice is to state a planned range, monitor what late interviews add, and report the decision — “n=20” justified by budget honesty beats “saturation” asserted as ritual.

A retention team ran 24 IDIs with churned subscribers, sampled across tenure bands and plan types, with boundary cases deliberately included (four who churned and returned). The guide’s core was episodic: the moment of decision, the weeks before it, the cancellation itself.

The first answers were price — uniformly, immediately, and (as the probing showed) mostly secondarily. Episode reconstruction found the actual sequence in seventeen of twenty-four accounts: a failed moment of need (“the one week I really needed it, it let me down”), followed by weeks of non-use, with price becoming salient only when the renewal notice arrived to a customer already emotionally gone. The contradiction probe did its work on participants who both praised the service and had cancelled it: what surfaced was ledger-keeping — a running private tally of “times it earned its fee”.

The analysis, coded across transcripts, reframed the retention problem from pricing to failure-recovery: the churn window opened at the failed moment, weeks before any dashboard flagged risk. The four boomerang customers confirmed it — each had returned after an unprompted service improvement addressed their specific failure. Interviewing past the first answer was the entire finding.

• Interviewing the guide, not the person. Marching through questions while the participant keeps offering doors nobody opens.
• Accepting the rehearsed layer. Studies built on first answers reproduce what any survey already knew.
• Leading and confirming. Questions carrying the hoped-for answer; nods and “exactly” training the participant toward it.
• Opinion harvesting. Collecting attitudes without episodes; analysis then has nothing concrete to work with.
• Over-recruiting the articulate. Samples of the willing and fluent, systematically missing the experiences hardest to talk about.
• Notes-only capture. Memory-based records lose the verbatim texture analysis needs; record and transcribe, with consent.

Interviews collect accounts, not behaviour: what people say they do, filtered through memory, self-presentation, and the interview relationship itself. Where the say–do gap matters, pair with observation or records.

The instrument is the interviewer, and quality varies with skill, energy, and rapport in ways that are hard to standardise across a team — training, shared guides, and early transcript review are the practical controls.

And depth costs coverage: two dozen hours of conversation is a small, purposive window. IDIs answer how and why superbly and how many not at all; the write-up should never let readers forget which.

Focus groups trade individual depth for interaction data — the right choice when the social negotiation of views is itself the object.

Contextual inquiry moves the conversation to where the behaviour happens, adding observation to account.

Diary studies capture experience longitudinally, escaping the single-sitting memory limits of the IDI.

Jobs-to-be-done interviews are a specialised IDI form built around switching decisions.

The interview’s afterlife is where teams drown: recording, transcription, and coding across dozens of hours. Evidano transcribes interviews with speaker separation in 100+ languages and codes the corpus — inductively or against your framework — with every theme linked to clickable quotes, so the move from “24 conversations” to “evidence-backed pattern” takes days rather than the weeks that quietly ration how many interviews teams dare to run.

The room itself stays human. Probing, silence, and the earned trust that produces second-layer answers are the method — and no tool attends a conversation for you.