New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
Interview as a data collection method — structured, semi-structured, and unstructured types compared, plus a four-stage pipeline from question to decision.
This guide explains the interview method of data collection in plain terms. The four stages of a working pipeline (question, transcript, structured extraction, report), the structured / semi-structured / unstructured types compared, and three visual examples from impact evaluation, longitudinal coaching, and grant-accelerator intake. No prior background in research methods needed.
"When I was running a quality workshop for our team, three weeks in I noticed the standards were not sticking. So I scrapped the format and rebuilt it as a peer-review protocol. Took two more weeks to see the team start catching defects in each other's work."
Source quote stored alongside the cell. Verifiable, sortable, and ready for the decision report.
Every working interview-data pipeline runs through four stages. The same example walked through each stage below: a single semi-structured prompt about adaptability, asked to fifty candidates in a hiring round. What the data looks like at each stage, and what each stage decides about what comes next.
"Tell me about a time you had to change your approach mid-project. What did you change, and what made you change it?"
How long did the change take to land?
What did you learn from the change?
Would you do it the same way again?
What lives here: a designed prompt plus optional probes. Same prompt asked to every respondent. The probe layer adapts to the response.
"When I was running a quality workshop, three weeks in I noticed the standards were not sticking. So I scrapped the format and rebuilt it as a peer-review protocol. Took two more weeks to see results. The team started catching defects in each other's work."
What lives here: the respondent's words, verbatim. Audio recording plus auto-transcription. Timestamped, attributed, ready to code.
What lives here: structured fields extracted against a fixed scheme. Source quote stored alongside. Verifiable against the transcript.
What lives here: aggregated codes across all interviews, with quotes still attached. The decision report ranks, segments, or summarizes.
Stages 1, 2, and 4 are well understood. Most teams know how to design a prompt, run an interview, and make a chart. Stage 3 is where interview-data work fails. Without a designed coding scheme and an extraction step that turns transcripts into structured cells, stage 4 collapses into anecdote and the team ends up debating impressions instead of routing decisions to the data.
The four stages above are the same fifty respondents at four levels of structure. The form decides the data shape, and the data shape decides what you can do with it. Words at stage 2; structured rows at stage 3; ranked cohorts at stage 4.
Five questions worth answering before designing an interview-data project. Plain-language definitions for the head terms students, researchers, and program teams arrive on this page asking, with the typology distinctions where they matter.
The interview method of data collection is a research technique in which an interviewer asks a respondent a designed set of questions and records the responses for analysis. The output is a transcript: the respondent's words, captured verbatim. The transcript becomes data when it gets coded against a scheme, which lets the team count themes, compare across respondents, and route findings to a decision.
The four stages of a working pipeline are: design the question, capture the transcript, extract structured signal from the transcript, and route the signal into a report. The page above walks through the four stages with the same example at each stage.
Three main types: structured, semi-structured, and unstructured. A structured interview asks every respondent the same questions in the same order, with response wording fixed in advance. A semi-structured interview uses a fixed list of designed prompts that the interviewer can probe and reorder as the conversation moves. An unstructured interview has a topic but no fixed prompt list and follows the respondent's framing.
Most program evaluation, qualitative research, and applicant-intake work uses semi-structured interviews because the structure makes responses comparable while the probing keeps depth available. The phrase personal interview is sometimes used in older research methods literature to describe any face-to-face interview, and is largely independent of the structured / semi-structured / unstructured distinction.
Interviews capture reasoning, context, and unanticipated detail that closed surveys cannot. The interviewer can probe a vague answer, follow an unexpected thread, and confirm understanding in real time. Interviews work well for sensitive or complex topics where respondents need space to think.
They produce data in the respondent's own words, which gives reports their direct-quote material. They reach respondents who would skip a written survey, including respondents with low literacy or limited time. And the depth-per-respondent ratio is higher than any other collection method, which is why interviews are the right tool for grant intake, program evaluation, and longitudinal research where each individual response carries weight.
Interviews cost more per response than surveys. Each one takes 20 to 60 minutes of interviewer time, plus transcription, plus coding. Sample sizes stay small, which limits prevalence claims. Interviewer effects matter: phrasing and probe choices vary between interviewers, which makes responses harder to compare cleanly.
The analysis step is the bottleneck. A transcript is not data until someone or something codes it against a scheme. Most teams that struggle with interview data are not stuck on collection; they are stuck on the gap between transcript and report. The four-stage pipeline is the framework for closing that gap.
Use interviews when the response options are not knowable in advance, when reasoning matters more than prevalence, or when the topic needs probing to surface a usable answer.
Interviews are the right tool for grant or accelerator intake (where the question is whether to fund this applicant), program evaluation (where you want to hear what changed in participants' words), longitudinal coaching or research (where the same person is followed over time), and qualitative research where unanticipated themes drive the finding. Use a survey instead when you need prevalence, when the response options are known, or when sample sizes need to be in the hundreds.
Structured reads from a fixed script: every respondent gets identical wording in identical order. Highly comparable, low depth. Semi-structured uses a fixed prompt list as the spine and lets the interviewer probe, reorder, or skip prompts based on the response. Comparable on the spine, deep where probing kicks in.
Semi-structured has a designed prompt list. Unstructured has a topic, no fixed prompts, and follows the respondent's framing. Semi-structured produces transcripts that compare across respondents on common questions; unstructured produces deeply individualized transcripts that resist clean comparison but surface the most unanticipated themes.
Older research methods literature called any face-to-face interview a "personal interview" to distinguish from telephone or mailed forms. "In-depth interview" describes a longer (60-90 minute), open-ended conversation focused on one individual's experience. The two terms overlap heavily and are often used interchangeably in current practice.
Interviews capture one respondent's experience without group dynamics shaping the response. Focus groups capture how a topic plays in a room (agreement, disagreement, social pressure). Different methods, different questions answered. Most projects use interviews for individual-experience data and focus groups for group-dynamics data.
The principles that decide whether interview data ends up powering a decision report or sitting in a folder of unread PDFs. Six rules covering the design, capture, extraction, and reporting stages.
The scheme is what turns transcripts into data.
A coding scheme names the fields the team will extract from every transcript. Adaptability, learning loop, founder-market fit, barrier types, theory-of-change layers, whatever the research or decision question requires. Designed before collection so each interview is asked the questions that produce the codes the report needs.
Why it matters: schemes designed after collection always miss the question that should have been asked.
Comparable on the spine, deep where probing earns its keep.
A fixed list of designed prompts every respondent answers (the spine) plus optional probes the interviewer can deploy when a response opens an unanticipated thread. The spine produces comparable transcripts; the probes capture depth on the points that matter most. Most program evaluation, intake, and qualitative work runs on this shape.
Why it matters: pure structured loses depth, pure unstructured loses comparability; the spine plus probes keeps both.
Notes are summaries; transcripts are data.
Audio recording captures the response in the respondent's words. Auto-transcription turns audio into searchable, codable text. Interview notes taken during the conversation are summaries, not data: they reflect the interviewer's filter, not the respondent's framing. Verbatim transcripts are the only form of capture that supports downstream coding and direct-quote reporting.
Why it matters: only verbatim transcripts can be re-coded as the scheme evolves; notes cannot.
A code without a quote is an opinion.
The intelligent cell pattern: each extracted field carries the transcript passage it was extracted from. Reviewers can verify any cell against the original words. Borderline extractions get human review. Strong extractions earn ranking weight. The pattern is what turns interview-data extraction from a black box into a verifiable pipeline.
Why it matters: verifiability is what makes structured extraction trustworthy for high-stakes decisions like grant funding.
Saturation is observable, not pre-committed.
Track new codes per interview as collection progresses. When several interviews in a row produce no new codes, you have reached saturation and additional interviews stop adding signal. For a homogeneous group on one research question, saturation typically arrives between 10 and 15 interviews. For heterogeneous populations, 20 to 30. Pre-committing to a number without tracking new-codes-per-interview wastes researcher time or stops short.
Why it matters: saturation is the only honest answer to "how many interviews"; the rest is guesswork.
Prevalence sets the scale; quotes give it texture.
Coded counts at the top of the report set the scale: how many respondents showed which signal, ranked by strength. Selected direct quotes from the transcripts sit alongside, giving the counts texture and the reader something to ground the prevalence claim in. Counts without quotes read as black box; quotes without counts read as anecdote. Both together is the report.
Why it matters: the report decides what gets done with the data, and both halves are needed for the report to land.
Six decisions a research team faces before the first interview. The broken pattern is what most teams default to. The working pattern is what produces transcripts that flow into a decision report.
| The choice | Broken way | Working way | What this decides |
|---|---|---|---|
|
Type of interview
Structured, semi-structured, or unstructured. |
Broken
Pure unstructured "let the conversation flow" interviews. Each transcript covers different ground because each interviewer chased different threads. Transcripts cannot be compared across respondents on common questions, so coded counts at stage 4 are unreliable. |
Working
Semi-structured with a fixed spine of designed prompts plus optional probes. Every respondent answers the spine prompts; probing depth varies based on the response. Comparable on the spine, deep where probing earns its keep. |
Whether responses are comparable across the cohort, which decides whether stage 4 reports are valid. |
|
When to design the coding scheme
Before, during, or after collection. |
Broken
Scheme drafted after the interviews are complete. The team realizes mid-coding that several spine prompts should have been asked. Now they need follow-up interviews or have to extract weak signal from off-topic transcript passages. |
Working
Scheme drafted before the first interview, then refined after a 3-5 interview pilot. The fields the report will need are decided up front, so the prompts that produce those fields make it onto the spine. |
Whether the team has to re-interview or can move directly from collection to extraction. |
|
How to capture the response
Notes, audio, or audio plus auto-transcription. |
Broken
Interviewer notes only. The notes are a summary of what the interviewer noticed, filtered through their attention during a conversation. The respondent's actual phrasing is gone. Direct-quote reporting is impossible; re-coding under a refined scheme is impossible. |
Working
Audio recording with auto-transcription, every interview. Verbatim transcripts. The interviewer is free to listen rather than write. Any passage can be re-coded as the scheme evolves. |
Whether transcripts can power a report at all, or only an interviewer summary. |
|
How to extract structured signal
Manual coding, multi-coder reliability, AI extraction. |
Broken
Single researcher reads every transcript and writes a narrative summary. No structured fields, no source quotes, no ranking surface. The summary reads as the researcher's interpretation, not as the respondents' words. Volumes above 20 transcripts become unmanageable on this method. |
Working
AI extraction against a fixed scheme, with every cell storing its source quote. Borderline extractions reviewed by a human; strong extractions earn ranking weight. Volumes of 50, 200, or 1,000 transcripts stay tractable. |
Whether the project scales beyond 20 transcripts or has to commission a research team to read each one. |
|
Sample size
How many interviews is enough. |
Broken
Pre-committed number based on a budget or grant proposal. Either too few (saturation not reached, claims do not hold) or too many (researcher time wasted on redundant interviews). The number was guessed at before the team had any signal on what saturation looked like. |
Working
Track new codes per interview as you go; stop when several interviews in a row produce no new codes. Pre-commit to a range (10-15 for homogeneous, 20-30 for heterogeneous) and end at whichever bound observable saturation hits. |
Whether claims drawn from the data hold against a saturation challenge from a reviewer. |
|
Report shape
What gets routed to the decision team. |
Broken
A 30-page narrative summary or a folder of PDF transcripts emailed to the decision team. Reviewers read what they have time for, debate impressions, and disagree on what the data showed. The report does not produce a ranked output the team can act on. |
Working
Coded counts at the top, ranked respondents by signal, with the strongest source quote for each ranking attached. Reviewers see prevalence and texture together. The decision team works from a structured artifact, not from impressions. |
Whether the data drives a decision or sits next to it as supporting documentation. |
Rows 1 and 2 control everything downstream. If the interview type and coding scheme are decided before the first interview, the rest of the pipeline can deliver. If they are decided after collection, every later stage spends time patching what the spine should have asked. The choice that costs nothing on day one (designing a scheme) is the choice that decides whether the project ships a decision report or a folder of transcripts.
A foundation accelerator running 200 applicants per cycle. The decision question: which applicants get funded. The interview-data pipeline replaces panel debate with ranked, verifiable structured signal across the full applicant pool.
"We were getting 200 applications a cycle. The panel would meet for two days, debate impressions, and walk out with a list. We had no way to see whether the strongest applicants on Monday were as strong as the strongest applicants on Friday. The transcripts existed but we read them like book reports. The shift was treating the interview as data, not a conversation. Five semi-structured prompts, 35 minutes per applicant, AI extraction against nine signals, and a ranked report. We still read the transcripts. We stopped relying on memory to compare them."
Program director, foundation accelerator (year 2 with structured pipeline)
Each applicant gets the same five spine prompts in a 35-minute video interview. Transcripts go through AI extraction against a nine-signal scheme. The selection panel reviews ranked signal distributions with source quotes attached, alongside the full transcripts.
Q1What's the most ambitious thing you've shipped, and what did you learn from it?
Q2Walk me through a time the market told you something you didn't expect.
Q3Tell me about a time you had to change your approach mid-project.
Q4Who is the person closest to your customer, and what would they say about you?
Q5What does the next twelve months look like, and what would slow you down?
Same prompts every applicant. Probes adapt to the response. Designed before collection so the scheme can extract what the panel needs.
"The most ambitious thing was rebuilding our onboarding flow when activation dropped from 41% to 23% in six weeks. I rewrote the first-run experience three times before we found a version that held activation above 40%. What I learned is I had been measuring the wrong thing. We tracked sign-ups and assumed activation followed. The market told us activation was a separate problem and we had to design for it directly."
Verbatim transcript with timestamps. The interviewer can probe in real time; the audio plus auto-transcription preserves the exact phrasing.
Each cell stores its source quote. A reviewer can verify any extraction against the actual transcript passage. Borderline cells (below 0.7 confidence) get human review.
Ranked across the full pool. The 36 "strong" applicants get prioritized for panel deep-review. Source quotes for each ranking attached.
Two full days of panel meetings, plus a week of pre-reading transcripts. Panelists arrived with different impressions of different applicants.
No way to compare the strongest Monday applicant against the strongest Friday applicant. Memory and notes filled the gap, which meant impressions filled the gap.
Decisions hard to explain to declined applicants and to the board. The team could point to the panel discussion, but not to a structured signal.
200 applicants was the breaking point. Beyond that, panel members could not hold the cohort in mind well enough to compare.
Four-hour panel reviewing the top tier of ranked applicants on each signal, with source quotes attached. Pre-reading replaced by ranked artifact.
Every applicant scored against the same nine signals from the same five prompts. Cross-cohort ranking is direct, not inferred from memory.
Every cell carries a source quote. Declined applicants can be told which signals were weak with the actual transcript passage attached. The board sees the same artifact the panel saw.
Capacity now sits in the interviewer schedule, not the panel. The same pipeline scales to 500 or 1,000 applicants without changing the report shape.
Most accelerators that try to fix the panel-debate problem buy better transcription software. Better transcripts do not produce better decisions. The constraint is upstream: the interview itself was designed as a conversation, not as a structured data collection event. No coding scheme, no spine prompts that map to the scheme, no extraction step that turns transcripts into ranked signals. Transcription is stage 2; the broken stage is stage 3.
The structural shift is to design the scheme before the first interview, build the spine prompts to produce the data the scheme needs, and treat extraction as a real engineering surface, not a downstream task. The interview becomes a data collection event, and the report becomes the decision artifact. That is the difference between a folder of transcripts and a ranked decision report, and it costs nothing on day one to choose.
The pipeline pattern repeats across domains. Three short walked examples below: program evaluation (mapping a coordinator's interview to a Theory of Change framework), longitudinal coaching (one client tracked across six sessions), and grant accelerator intake (compact echo of the worked example above).
A program evaluator interviewing a workforce-development coordinator. The transcript gets mapped, line by line, onto the program's Theory of Change layers (activities, outputs, outcomes, impact). The report shows the framework filled in with the coordinator's own words.
"What outcomes did your participants experience because of this program in the last twelve months?"
"Out of twenty-three graduates, four secured roles in their target field within six months. Two more got informational interviews that became referrals. The rest are still active in the network we built."
What the pipeline produces: a Theory of Change framework filled in with the coordinator's own evidence, time-stamped, with the source quote attached to each layer. The evaluator can compare across coordinators, surface where the program does and does not produce its claimed outcomes, and route findings to the funder report.
A career coach running monthly sessions with the same client, asking the same goal-progress prompt every session. Each transcript gets coded against a fixed scheme (goal status, momentum, barriers). The report shows the client's trajectory across six sessions in one view.
"What progress have you made on your career goal since our last session, and what is in the way right now?"
"This month I sent three applications and had two networking conversations. Both calls went better than I expected. The hard part was the imposter feeling before each one."
What the pipeline produces: a six-session momentum trend for one client, with the actual session quote behind each datapoint. The coach can see when momentum shifted, what barriers persisted, which interventions correlated with progress, all without re-reading six transcripts.
A compact echo of the worked example above. Same five spine prompts asked to every applicant, same nine-signal extraction scheme, but a different output view: the top-of-stack ranked list the panel reviews first, with the strongest source quote for each ranking attached.
"What's the most ambitious thing you've shipped, and what did you learn from it?"
"Built a logistics tool for warehouses with 50-200 SKUs that nobody else served. Got it to twelve paying customers in six months. The lesson was that the segment we ignored had the cleanest economics."
What the pipeline produces: a ranked decision artifact the panel reads first. Every ranking row links to the source quote and the full transcript. Decisions are explainable, ranking is comparable across the cohort, and the panel spends time on judgment instead of memory.
Each tool above handles part of the four-stage pipeline well. Otter and Rev produce verbatim transcripts. Dovetail, NVivo, and Atlas.ti support qualitative coding inside their own analysis interfaces. The architectural gap most teams hit is between stage 3 and stage 4: turning extracted codes into a decision report that ranks, segments, and routes signal back to the team that needs it, with the source quote attached to every cell.
Sopact Sense closes that gap by keeping the extraction step (intelligent cell with source quote) inside the same workflow that produces the ranked decision report. The pipeline runs end to end without the team having to copy structured fields out of a coding tool and into a separate reporting system, which is the seam where most interview-data projects lose their verifiability.
The interview method of data collection is a research technique in which an interviewer asks a respondent a designed set of questions and records the responses for analysis. The output is a transcript: the respondent's words, captured verbatim. The transcript becomes data when it gets coded against a scheme, which lets the team count themes, compare across respondents, and route findings to a decision. The four stages of the working pipeline are: design the question, capture the transcript, extract structured signal from the transcript, and route the signal into a report.
Three main types: structured, semi-structured, and unstructured. A structured interview asks every respondent the same questions in the same order, with response wording fixed in advance. A semi-structured interview uses a fixed list of designed prompts that the interviewer can probe and reorder as the conversation moves. An unstructured interview has a topic but no fixed prompt list and follows the respondent's framing. Most program evaluation, qualitative research, and applicant-intake work uses semi-structured interviews because the structure makes responses comparable while the probing keeps depth available.
Interviews capture reasoning, context, and unanticipated detail that closed surveys cannot. The interviewer can probe a vague answer, follow an unexpected thread, and confirm understanding in real time. Interviews work well for sensitive or complex topics where respondents need space to think. They produce data in the respondent's own words, which gives reports their direct-quote material. And they reach respondents who would skip a written survey, including respondents with low literacy or limited time.
Interviews cost more per response than surveys. Each one takes 20 to 60 minutes of interviewer time, plus transcription, plus coding. Sample sizes stay small, which limits prevalence claims. Interviewer effects matter: phrasing and probe choices vary between interviewers, which makes responses harder to compare. And the analysis step is the bottleneck: a transcript is not data until someone or something codes it against a scheme. Most teams that struggle with interview data are not stuck on collection; they are stuck on the gap between transcript and report.
A structured interview reads from a fixed script: every respondent gets the same questions in the same order, often with fixed response options. The data is highly comparable but shallow. A semi-structured interview uses a fixed prompt list as the spine and lets the interviewer probe, reorder, or skip prompts based on the conversation. The data stays comparable across respondents on the spine prompts, while the probing layer captures depth on the points that matter most. Most program evaluation, applicant intake, and qualitative research uses semi-structured because it balances comparability with depth.
Interview data is analyzed by coding the transcript against a scheme. A coding scheme is a fixed set of categories, themes, or extracted fields the team has decided in advance to look for. Each transcript passage gets tagged with one or more codes, and the codes get aggregated to produce counts, themes, and selected quotes for the report. Coding can be done by hand, by multiple coders with reliability checks, or by AI extraction against a scheme with human review on borderline passages. The unit that most teams call an intelligent cell is one extracted field plus its source quote, ready to slot into a structured table.
Use interviews when the response options are not knowable in advance, when reasoning matters more than prevalence, or when the topic needs probing to surface a usable answer. Interviews are the right tool for grant or accelerator intake, program evaluation, longitudinal coaching or research, and qualitative research where unanticipated themes drive the finding. Use a survey instead when you need prevalence, when the response options are known, or when sample sizes need to be in the hundreds.
A personal interview is an interview conducted face-to-face, typically one-on-one between an interviewer and a respondent. The phrase is most common in older research methods literature, where it contrasted with telephone interviews and mailed questionnaires. In current practice, the same form is conducted by video call as often as in person, and the methodology is identical regardless of medium. Personal interviews allow the interviewer to read non-verbal cues, build rapport, and probe sensitive areas more carefully than a survey can.
A grant accelerator running applicant interviews. The accelerator asks every applicant the same five semi-structured prompts (founder background, most ambitious project shipped, learning from failure, market understanding, twelve-month plan). Each interview runs about 35 minutes. Transcripts are AI-coded against a fixed scheme of nine signals. The report ranks all applicants on each signal, with the strongest direct quote for each high-scoring applicant attached. The selection committee reviews ranked applicants and full transcripts together, instead of debating impressions.
Most semi-structured interviews run 25 to 45 minutes. Shorter than 20 minutes and you cannot probe enough to add depth a survey could not capture. Longer than 60 minutes and respondent fatigue degrades the later answers, which means coding spends time on lower-quality data. The exception is in-depth qualitative research, where 60 to 90 minute interviews are normal, and longitudinal coaching, where a 30 minute session is typical and the value comes from repeated sessions over time rather than length per session.
Saturation is the point at which additional interviews stop surfacing new codes. The published estimates vary by topic and population. For a homogeneous group on a single research question, saturation typically arrives between 10 and 15 interviews. For a heterogeneous population or multi-topic research, 20 to 30 is common. The honest answer for any specific project is to track new codes per interview as you go: when several interviews in a row produce no new codes, you have reached saturation. Pre-committing to a number without tracking new-codes-per-interview is a common cause of either too few or too many interviews.
An intelligent cell is one structured field extracted from a transcript passage, with the source quote attached for verification. Example: a transcript passage of a grant applicant describing a workshop pivot gets extracted into the cell adaptability_evidence: present, time_horizon: 3_weeks, learning_loop: explicit, with the actual sentence from the transcript stored alongside. Intelligent cells turn transcripts into rows in a structured table, which lets the team count, sort, segment, and rank across many transcripts at once while still being able to verify any cell against the source quote. The bridge stage between transcript and decision is where most interview-data pipelines fall apart, and the cell-and-quote pattern is what fixes it.
Both, depending on the analysis stage. The raw transcript is qualitative data: words in the respondent's own language. After coding against a scheme, the same data becomes quantitative: counts of codes across respondents, distributions, and trend comparisons across cohorts or waves. Interviews are sometimes labeled as a qualitative method only because the collection stage produces qualitative data, but most modern interview-data pipelines produce both qualitative outputs (selected direct quotes) and quantitative outputs (code counts, ranked respondents, theme prevalence) from the same transcripts.
Six related pages from the use-case cluster: question design, scaled analysis, the open-vs-closed comparison, plus parent topics on qualitative research methods and coding.
A guide to writing the spine prompts that produce codable transcripts. Six rules, a template library, and a worked-through redesign.
The same coding-and-extraction pattern applied to surveys with thousands of free-text responses. Volume, scheme drift, and reviewer workflow.
When to use each, when to mix them, and the four design choices that decide whether your data answers the question you asked.
The methods stack interviews live inside: ethnography, case study, grounded theory, phenomenology, and the design choices that distinguish them.
A deep dive on stage 3 of the pipeline: scheme design, deductive vs inductive coding, intercoder reliability, and AI-assisted extraction.
When a focus group answers a question one-on-one interviews cannot, and how the four-stage pipeline applies when group dynamics are the data.
A working session, not a demo. Walk through an interview transcript you have already collected. Identify the cells worth extracting, draft a coding scheme that turns those cells into ranked signal, and see how the pipeline routes the signal into a decision-ready artifact.
60 minutes, video. Working session structured around your actual interview transcripts and the decision question they need to answer.
One or more interview transcripts in any format. The decision question the interviews were meant to inform. Any draft coding scheme if you have one.
A draft coding scheme tied to your decision question, a sample of cells extracted from your transcripts, and a clear path from those cells to the report shape your team needs.
