AI customer-interview synthesis (L3)
Run discovery / win-loss / churn interviews, auto-transcribe with Granola or Fireflies, and use Claude (Projects + 200K context) to synthesize themes across 20+ interviews into a single PMM-ready insight doc. Replaces 2 weeks of manual qualitative coding with 2 hours.
The steps
- 01
Standardize interview capture
Tool: Granola
Pick one notetaker (Granola for in-person + remote, Fireflies for bot-joined). Standardize a 30-min interview template: 5 questions, every interview. Without a consistent question stem, synthesis quality drops 60%. Owner: PMM or UX research lead. DoD: every interview lives in one folder, named {date}-{customer}-{segment}.md with the 5-question structure.
- 02
Build a Claude Project per insight cycle
Tool: Claude Projects
Create a Claude for Work Project: "Win-loss synthesis Q2 2026". Knowledge: every interview transcript (paste up to ~200K tokens), your ICP doc, your messaging frame. System prompt: "You are a senior PMM doing qualitative research synthesis. Cite specific interviews when you make claims. Distinguish strong-signal (3+ customers) from weak-signal themes." Owner: PMM. Pitfall: throwing transcripts into a fresh ChatGPT each time, you lose context and the synthesis is shallower. DoD: Project exists, all transcripts loaded, system prompt set.
- 03
Iterate themes with Claude, with citation discipline
Tool: Claude
Workflow: 1) Ask Claude: "What are the top 5 themes across these wins? For each, cite the interviews." 2) Ask: "Which themes appear in wins but NOT in losses?" 3) Ask: "What language do champions use to describe value? Quote verbatim." 4) Ask: "What are the disqualifying patterns in losses?" PMM verifies each cited quote by clicking back into the transcript. If Claude hallucinates a quote, raise it in the chat, it self-corrects. Owner: PMM. DoD: each theme in the final doc has a verified quote and an interview count.
- 04
Ship the insight doc to product + sales
Tool: Notion
Output: a 3-page "Q2 customer truth" Notion doc. Sections: top themes (with verbatims), messaging tweaks, product asks, disqualifying patterns. Walk product through it in a 30-min session. Walk sales through it in their next kickoff. Owner: PMM. DoD: doc is dated, has named owners for each follow-up action, and shows up in the next quarterly roadmap discussion.
Tools in this playbook
Next playbooks
Unfamiliar terms are defined in the AI and Revenue Dictionary. Related frameworks live in the framework library.