User Interview Synthesis

A Claude Code skill that transforms raw, messy user interview notes into structured discovery output: themes, pain points, opportunity areas, supporting quotes, and frequency counts.

Built for product managers doing continuous discovery — works at any research stage from early problem exploration to solution validation.

---

What it does

• Gathers context: raw notes, research question, user segments, participant count, discovery stage, and hypotheses to test
• Reads all notes fully before drawing conclusions — tags observations to participants, separates direct quotes from paraphrases
• Synthesizes themes with real frequency counts ("8 of 12, 6 unprompted") not vague descriptors
• Surfaces behavioral observations vs. stated preferences — where what users did differed from what they said
• Frames opportunity areas as user needs, not solutions
• Calls out contradictions and counter-signals rather than smoothing them over
• Offers optional JTBD framing, OST branches, or a stakeholder summary on request

---

Who it's for

Product managers turning raw research into something shareable and actionable. Especially useful after a sprint of discovery interviews when you have 6–12 sets of notes and need to find the signal quickly.

---

Output structure

1. Overview — who was interviewed, what was explored, honest signal quality read
2. Key themes — 3–6 themes with frequency counts and representative quotes
3. Pain points — specific friction ordered by frequency × severity
4. Behavioral observations vs. stated preferences
5. Opportunity areas — framed as user needs
6. Surprising or contradictory findings
7. Signal quality and gaps — what's well-supported vs. what needs more signal
8. Optional: JTBD statements, OST branches, stakeholder summary

---

How to use it

Install via Claude Code. Clone this repo and copy the folder into your skills directory:

macOS / Linux:

cp -r ProductManagerTranscriptAnalysis ~/.claude/skills/user-interview-synthesis

Windows:

Copy-Item -Recurse ProductManagerTranscriptAnalysis "$env:USERPROFILE\.claude\skills\user-interview-synthesis"

Restart Claude Code. Then trigger with natural language:

• "Synthesize these 8 user interviews — we were exploring why users abandon onboarding"
• "What patterns are in my research notes?"
• "Turn these interview notes into a shareable insight doc"
• "Help me make sense of this research" + paste notes

Paste notes in whatever format you have them — fragments, timestamps, stream-of-consciousness. Don't clean them up first.

---

Optional output formats

Add these to your request:

• "Include JTBD framing" — rewrites top opportunity areas as Jobs-to-be-Done statements
• "Map to an OST" — structures findings as Opportunity Solutions Tree branches
• "Give me a stakeholder summary" — 5–7 bullet exec-ready version for Slack or a leadership update

---

What makes it rigorous

• Frequency counts are real ("5 of 8"), not rounded ("most users")
• Contradictions between participants are surfaced, not smoothed over
• Thin or low-quality notes are called out — no confident synthesis built on weak data
• Behavioral evidence is separated from stated preferences

---

Evals

Three test cases in evals/:

1. 8 interviews on workflow automation abandonment — mixed user types, clear outlier (the one user who succeeded)
2. 10 interviews on a solution concept (enterprise vs. SMB split) — conflicting signals across segments, JTBD framing requested
3. 6 rough interviews on a potential new product — tests honesty about thin data quality, OST framing requested