Interview Synthesis - AI Agent Skill

Use this when processing raw interview notes or transcripts into structured insights after conducting user interviews.

How it works

You provide interview notes or transcript, research objective, and client name
The skill extracts structured insights using JTBD framing, surfaces patterns, and flags assumptions
It returns a synthesis document with key findings, pain points, quotes, and recommended next steps

Prompt

You are synthesizing user interview data for Kate's ux-research-synthesis engagement. Before writing, read knowledge/voice-tone-guide.md -- use the internal voice (direct, working-doc tone -- this is an internal analysis artifact, not a client deliverable).

Inputs I will provide:

Interview Notes: {{INTERVIEW_NOTES}} (raw transcript, notes, or audio transcription from one or more interviews)
Research Objective: {{RESEARCH_OBJECTIVE}} (what questions this research is answering, what decisions it informs)
Client: {{CLIENT}} (company name and engagement context)

Step 1: Read and orient

Read the full interview content before extracting anything. Identify:

Who was interviewed (name, role, context)
Date and duration of the interview
Background and relationship to the product area

If any of these are missing from the notes, note them as [CLARIFY WITH KATE: ...].

Step 2: Pull relevant frameworks

Read knowledge/pm-discovery-frameworks.md, specifically:

Jobs to Be Done framing to structure what the participant is trying to accomplish
The Mom Test principles to evaluate signal quality (past behavior vs. future intent, specifics vs. generalities)
Assumption identification (four-risk model) to tag findings against value, usability, viability, and feasibility risks

Search knowledge/engagement-history.md for past engagements with {{CLIENT}} or similar research contexts. Look for:

Prior interview findings that this data should build on
Patterns or themes from earlier rounds
Known assumptions that these interviews were meant to test

Step 3: Produce the synthesis

Interview Summary

Field	Detail
Date	[date of interview]
Participant	[name, role, relevant context]
Interview Length	[duration]
Interviewer	Kate Makrigiannis

Current State

What they do today: [current solution, workflow, or workaround]
What they like about it: [functional jobs being satisfied, with importance and satisfaction signals]
What is broken: [unmet needs, friction, pain -- use their words]

Jobs to Be Done

For each job identified:

Job: [what the participant is trying to accomplish]
Desired outcome: [what success looks like in their words]
Importance: [how much this matters to them -- high/medium/low with evidence]
Current satisfaction: [how well their current approach works -- with evidence]

Key Quotes

Verbatim quotes that carry real signal. For each:

"[exact quote]" -- re: [topic or theme]

Include 3-8 quotes. Prioritize quotes that reveal pain, motivation, or surprising behavior over polite agreement.

Pain Points (ranked)

Rank by energy and emphasis the participant put behind each problem:

[Pain point] -- [evidence: what they said, how they said it]
[Pain point] -- [evidence]

Surprise Findings

Anything unexpected that challenges existing assumptions or reveals something the team did not anticipate.

Assumption Validation

Assumption	Supported?	Evidence
[assumption from research objective]	Yes / No / Unclear	[what they said or did]

Action Items

[specific follow-up, tagged with urgency and owner where possible]
[e.g., "Follow up with participant about their data migration workflow -- they mentioned a workaround worth exploring"]

Cross-Interview Patterns

If multiple interviews are provided, add:

Themes appearing across participants
Contradictions or divergences worth investigating
Confidence level in emerging patterns (how many participants, how consistent)

Research Nuggets

After completing the synthesis, extract 3-8 atomic research nuggets. Each nugget is a self-contained insight that can be tagged, searched, and reused across studies without needing the full synthesis context.

Format each nugget as:

#### Nugget: (Insight title — 5-10 words)
- **Quote:** "(Verbatim quote that grounds this insight)"
- **Interpretation:** (One sentence — what this means for the product)
- **Tags:** (2-4 tags from: pain-point, workaround, unmet-need, mental-model, decision-factor, trigger, workflow, delight, assumption-validated, assumption-challenged)
- **Confidence:** High / Medium / Low (based on signal quality and corroboration)
- **Source:** (Participant name/role, date)

Nugget rules:

Each nugget stands alone — someone reading it without the full synthesis should understand the insight
One nugget per distinct insight. Don't combine multiple findings into one nugget
Prefer nuggets that challenge assumptions or reveal surprises over nuggets that confirm expectations
Low-signal nuggets (vague answers, hypothetical responses, polite agreement) should be tagged Low confidence or omitted entirely

Step 4: Add synthesis pointers

Include this note at the bottom:

For interview prep, use the interview-script skill. For discovery call processing, use discovery-debrief.

Example Output

Input

Client: Fieldvine (Series B agri-tech company; Kate is embedded with their product team for a 10-week discovery sprint on a new crop monitoring mobile app)
Research Objective: Understand how agronomists currently track field observations and share findings with farm owners — to inform MVP feature prioritization and validate the assumption that real-time photo logging is the highest-value feature
Interview Notes:

Talked to Marcus Delgado, Senior Agronomist at Hensley Family Farms (client of Fieldvine's existing soil-testing service). Interview was about 45 minutes, done over Zoom on June 3. Marcus has been an agronomist for 11 years, manages ~14,000 acres across three farm sites. He's not a tech-avoider — uses an iPad in the field — but says most ag apps "feel like they were built by people who've never worn boots."

Currently uses a combination of Google Sheets, WhatsApp, and paper scouting forms. Takes photos on his iPhone and texts them to the farm owner, Dale Hensley, directly. Says Dale likes that — feels personal. "Dale doesn't want a dashboard, he wants to hear from me."

Big frustration: when he's back at the office he has to reconstruct everything from memory and photos into a formal scouting report. "I'm basically doing the same work twice. I write it down in the field, then I type it up at the end of the day, and half the time I can't read my own handwriting." Spends 45–60 min per day on this.

Photo logging specifically — I asked him about it. He said photos are already part of his workflow, but the problem isn't taking them, it's connecting them to a specific field location and a specific crop issue. "I've got 400 photos on my phone and no idea which field half of them are from." He seemed genuinely frustrated here, almost laughed about it.

Asked what would make his job easier. He said: "If I could just talk into my phone and have it show up in the report, that would save me more time than anything." Unprompted, he mentioned this twice. Also mentioned wanting to flag severity levels — something like a red/yellow/green system — so Dale knows what actually needs attention vs. what Marcus is just noting for the record.

He does not currently use any Fieldvine products directly — he interacts with their soil reports through Dale. Doesn't know much about what Fieldvine is building.

He asked at the end: "Will this app work offline? I lose signal in the south fields constantly." Didn't ask him a follow-up — we were out of time. [Note from Kate: this came up in the Sorenson interview too, flag this]

Overall vibe: Marcus is a credible, experienced user. Very specific about his pain, not speculating. Strong Mom Test signal throughout.

Output (abbreviated)

Interview Synthesis — Fieldvine / Crop Monitoring MVP

Engagement: Fieldvine discovery sprint, Week 3 of 10 Synthesized by: Kate Makrigiannis