interview-synthesis

name	interview-synthesis
description	Use when turning raw interview notes into themes - produces themes, verbatim quotes, and decisions or hypotheses
kind	workflow
audience	anyone
ai-tools	any
complexity	guided
time	30-90 min per round of interviews
version	1.0.0
source	bundled

Interview Synthesis

What this does

Turns raw interview notes (1-N interviews) into a structured synthesis: themes that recurred, verbatim quotes that anchor each theme, and the decisions or hypotheses the synthesis supports.

When to use

• After user research interviews
• After expert interviews for journalism / writing / product work
• After internal stakeholder interviews
• After any structured set of conversations where the next step is "what do we do with this"

What you bring (Inputs)

• Raw notes from N interviews (transcripts or your shorthand)
• The original research question (what were you trying to learn)
• Who the synthesis is for (you / a team / a client) — drives format

What you get (Output)

A document with: research question, methods, themes (3-7), verbatim supporting quotes per theme, surprises, and the decisions or hypotheses the data supports.

How it works (Steps)

This is a workflow.

Stages

Stage 1: Re-read with fresh eyes

Read all the notes through once without highlighting anything. Just take in the totality. This guards against latching onto the first idea.

Stage 2: Tag each note

Go through each interview and tag observations. Tags are short — 2-4 words. Examples: "Pricing confusion," "Wants integration," "Onboarding too slow."

Don't merge tags yet. Keep them granular.

Stage 3: Cluster tags into themes

Group similar tags. Aim for 3-7 themes. If you have 1-2, you're under-clustering. If you have 15+, you're over-tagging.

A theme is interesting if it appears in more than one interview AND is non-obvious.

Stage 4: Find the verbatim quote per theme

For each theme, pull the strongest 1-2 verbatim quotes. Use the speaker's actual words. These anchor the theme and prevent paraphrasing drift.

Stage 5: Note surprises

What did you NOT expect to hear? What contradicted your hypothesis going in? Surprises are often the most valuable output.

Stage 6: State the decisions or hypotheses

The synthesis should support concrete next steps. Examples:

• Decision: "We will rebuild the onboarding flow."
• Hypothesis: "Pricing confusion drives churn in months 2-3 — we should test a clearer pricing page."

If the synthesis doesn't support any decision or hypothesis, you under-interviewed or under-tagged.

Stage 7: Write the document

Standard structure:

• Question: what we were trying to learn
• Method: N interviews, who, when, format
• Themes (3-7) with verbatim quotes
• Surprises
• Decisions and hypotheses
• Next steps (more interviews? a test? a decision?)

Checkpoints

• After Stage 2: every interview has been tagged
• After Stage 3: 3-7 themes, each appearing in 2+ interviews
• After Stage 4: every theme has a verbatim quote
• After Stage 6: at least 1 decision or hypothesis is named

Loop-back conditions

• If a theme has no verbatim quote: cut it (it's your projection, not data)
• If the synthesis supports no decision: you may need more interviews — name the gap explicitly

Quality bar

• Verbatim quotes are quoted exactly, with speaker attribution if appropriate
• Themes are non-obvious (if the theme is "users want it to be easier," cut it — too generic)
• Surprises section is present even if short
• Decisions/hypotheses are concrete

Variations

• Single interview: skip clustering; instead, find the 3-5 most interesting points
• Quantitative + qualitative: pair counts (how many interviewees said X) with the quote
• Long-running research: each round of interviews updates the prior synthesis; mark what changed

Example

Input: 8 interviews with new SaaS users about onboarding.

Output:

• Question: Why are users dropping off in week 1?
• Themes:
  1. Setup felt longer than promised (6/8) — "I thought it'd take 5 minutes; I spent 40."
  2. Sample data made things worse, not better (5/8) — "I couldn't tell what was demo and what was mine."
  3. Email reminders felt naggy (4/8) — "Three emails in two days; I unsubscribed."
• Surprises: nobody mentioned pricing.
• Decisions: cut sample data from default; reduce onboarding emails to 1.
• Hypothesis: clearer setup-time estimate would reduce week-1 drop by 20%.
• Next: A/B test the setup-time estimate language.