CONTEXT(CLONE)
Sign in

Customer interview synthesis with every theme quoted

For: User researcher or product manager synthesising a round of discovery interviews

Pattern: Map, verify, reduceNeeds live modelsDesigned for 6 to 80 agents
Get early accessSign in to use

The pain today

Synthesis is sticky notes and memory. The loudest interview shapes the themes, quotes get paraphrased into what the team hoped to hear, and nobody can trace a finding back to who said it.

The ask

“I attached the transcripts from our discovery interviews and the research questions. What themes come up, who said what, and where do participants contradict each other? Quote the participant for every theme and tell me which themes rest on only one or two people.”

Plain words, as you would say it to a colleague. Edit it to fit your case before you send it.

What you attach or connect

  • Interview transcripts with speaker labels
  • Research questions and discussion guide
  • Participant segments

The unit of work

One worker task per one interview transcript.

Why a swarm fits

Each transcript is coded alone against the same questions, so no interview colours the reading of another. Themes are then built from quoted extracts only.

Not for

A handful of short interviews you sat in on: your notes and one strong model are enough.

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

  1. Before workers, before a task runs

    Small worker or strong modelYes or no, with a probability

    Before a task runs

    Are speaker labels present and consistent enough for a small worker to separate participant from interviewer?

    Sees only: A sample of turns from one transcript

    Why: Everything downstream depends on knowing who said it.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenEscalate to a strong model

  2. After workers, the judge checks

    Evidence checkA choice among options

    After a worker answers

    Who says the quoted words, and were they prompted?

    Sees only: The quoted turn with the turns just before it

    Why: Keeps the team's own words, and answers they led people to, out of the findings.

    • Participant, unpromptedthenAccept
    • Participant, answering a leading questionthenMark unresolved
    • Interviewer speakingthenReject and retry
    • Speaker unclearthenEscalate to a strong model

  3. Evidence checkYes or no, with a probability

    After a worker answers

    Does the participant's quoted sentence express the coded theme itself, such as distrust of automatic changes, rather than merely mention the topic?

    Sees only: The quoted sentence and the code's definition

    Why: Stops paraphrase from turning what was said into what the team hoped to hear.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry

  4. Reconciler, while merging

    Conflict checkYes or no, with a probability

    While reconciling

    Do these two participants say opposite things about the same experience, such as one relying on a feature the other avoids?

    Sees only: Two verified quotes under one theme with participant segments

    Why: Contradicting participants are kept under the same theme, where the insight usually is.

    • Yes: 0.80 or higherthenAccept
    • Unsure: 0.45 up to 0.80thenEscalate to a strong model
    • No: below 0.45thenContinue

  5. Run control, between rounds

    Another round?Yes or no, with a probability

    Between rounds

    Did revising the coding form after the first pass change how any transcript was coded?

    Sees only: The codes per transcript before and after the revision

    Why: Another coding pass is only worth it while the form is still moving.

    • Yes: 0.60 or higherthenContinue
    • Unsure: 0.30 up to 0.60thenStop
    • No: below 0.30thenStop

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Is this theme supported by very few participants, or does one of its quotes contain something that could identify a participant?

    Sees only: One theme with its participant list and quotes

    Why: The researcher decides which themes matter and protects participants' consent.

    Accountable: The researcher decides which themes matter and what they mean. Transcripts need participant consent and masking.

    • Yes: 0.30 or higherthenAsk a person
    • Unsure: 0.10 up to 0.30thenAsk a person
    • No: below 0.10thenAccept

The fleet: who does what

Model tiers by role, not brands: you choose the models. Strong reasoning models plan and reconcile, small fast models do the wide work, and the judge is a decision model from a different family, so it does not share the workers' blind spots.

  1. Planner

    A strong reasoning model turns the research questions into a coding form and revises it after a first pass.

  2. Workers

    Small fast workers from an open-weight family each code one transcript with verbatim participant quotes.

    Designed for 6 to 80 agents, one worker task per one interview transcript. Each worker receives only its own unit.

    Decisions here:1. Small worker or strong model

  3. Judge, from a different model family

    A decision model from a different family checks each quote is the participant speaking and supports the code.

    Decisions here:2. Evidence check3. Evidence check

  4. Reconciler

    A strong reasoning model builds themes by segment and keeps contradicting participants under the same theme.

    Decisions here:4. Conflict check5. Another round?

  5. Accountable person

    The researcher decides which themes matter and what they mean. Transcripts need participant consent and masking.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every theme lists participants and a verbatim quote from each
  • Interviewer statements are never counted as participant views
  • Themes supported by very few participants are labelled thin
  • Answers to leading questions are marked as prompted

What comes back

  • Themes by research question with participant quotes
  • Contradictions between participants or segments
  • Thin themes that need more interviews
  • Questions the round did not answer

What to measure

  • Quotes that support their theme, from a sampled check
  • Themes the researcher adds that the swarm missed
  • Hours from last interview to readout
  • Cost per transcript coded

Names of measures only. No result is claimed for this template.

Templates open in the workspace chat with the ask filled in. Nothing runs until you send it.

Get early accessSign in to use

More in Research and intelligence

Literature review where every claim cites a passage

For: Research lead, analyst or doctoral researcher scoping a field before committing to a direction

A review means a very long reading list.

Pattern: Map, verify, reduceNeeds scale5 decisionsDesigned for 30 to 800 agents

Competitor teardown from public sources with citations

For: Strategy analyst, product leader or founder preparing a market entry or board discussion

A teardown is stitched from a few web pages and hearsay.

Pattern: Hierarchical decompositionNeeds a connector6 decisionsDesigned for 20 to 500 agents