code-ai-installer 4.0.0 → 4.0.1-b

package/domains/product/locales/en/.agents/skills/user-interview-script/SKILL.md

@@ -1,297 +1,314 @@
----
-name: user-interview-script
-description: Structured script for B2B interviews — separation of buyer and end-user, pain-discovery without leading questions
----
-# User Interview Script (B2B)
-
-> **Category:** Discovery  ·  **Slug:** `user-interview-script`
-
-## When to Use
-
-- Before a major product initiative, to capture real pain points and JTBD.
-- When preparing `/ship-right-thing` or `/shape-prioritize` — as the foundation of the Discovery Brief.
-- When validating new ideas / hypotheses from `$opportunity-solution-tree`.
-- When customer support / CRM data is insufficient to understand root cause.
-
-## Input
-
-| Field | Required | Description |
-|-------|:--------:|-------------|
-| Segment | ✅ | SMB / mid-market / enterprise |
-| Interviewee role(s) | ✅ | Buyer / end-user / admin / champion — separately |
-| Product context | ✅ | Current product, category, alternatives |
-| Hypotheses / areas of interest | ✅ | What we want to learn |
-| Duration | ⬚ | Default 30-45 min |
-| Format | ⬚ | 1-on-1 / 2-on-1 (usually 2: one leads, one takes notes) |
-
-> If segment and roles are not defined — **request from the user**.
-
-## Data Sources
-
-1. Existing interviews (mempalace, support tickets, NPS comments) — avoid duplicating questions.
-2. CRM data — deal status, cohort, usage patterns.
-3. Product analytics — understand how the product is really used (don't rely on «how they say» they use it).
-
-### Related Skills
-
-| Skill | What we take | When to invoke |
-|-------|-------------|----------------|
-| `jtbd-canvas` | Jobs structure → questions for discovery | After formulating job hypotheses |
-| `problem-statement` | Problem draft → focused questions | If problem is already formulated |
-| `assumption-mapping` | Top assumptions → validation questions | For high-risk assumptions |
-| `opportunity-solution-tree` | Opportunities → exploration questions | When discovering new opportunities |
-
-## Protocol
-
-### Step 0 — Segmentation
-
-Split interviews by role, **do not mix buyer and end-user in one script**:
-
-- **Buyer** (VP Engineering, Director of Ops, CIO) — ROI, compliance, fit with current stack, purchasing concerns.
-- **End-user** (developer, HR manager, sales rep) — daily experience, friction points, workarounds.
-- **Champion** — who promotes the product internally, their motivations.
-- **Admin** — who configures, integrates, monitors compliance.
-
-### Step 1 — Opening (5 min)
-
-Goal — establish context, not sell.
-
-1. Introduce yourself: «I work on [...] at [company]. Today I want to learn how you [context], not to sell anything.»
-2. Permission to record (if applicable).
-3. Warm-up: «Tell me in 2-3 sentences — what does your day look like? What's your role on the team?»
-4. **Rule:** never mention the product first.
-
-### Step 2 — Context Building (10 min)
-
-Understand the interviewee's workflow.
-
-Framework questions:
-- «Tell me about the last time you [situation]. What happened? Step by step.»
-- «How do you currently solve [proximate problem]?»
-- «Who else is involved? Who makes the decision?»
-- «What tools do you use?»
-
-**Rule:** ask about a specific recent case, not abstract generalizations.
-
-### Step 3 — Pain Discovery (15 min)
-
-Don't ask directly «what hurts». Look for evidence:
-
-- «What was difficult about that case?»
-- «What would you like to change?»
-- «When was the last time you complained about this?»
-- «What takes up the most time that you don't like?»
-- «If you had a magic wand — what would you fix?»
-
-**Red flags:**
-- Leading question: «You find [X] difficult, right?»
-- Hypothetical: «If we built [feature] — would you use it?»
-- Direct feature request: «Do you need [feature]?» → dig down to why.
-
-### Step 4 — Buyer-specific (for buyer, 10 min)
-
-- «Who is involved in the decision? (procurement, finance, legal, IT)»
-- «How do you evaluate ROI?»
-- «What made you look at [category] now?»
-- «What compliance / security requirements do you have?»
-- «What is your budget cycle?»
-- «What didn't work with similar tools in the past?»
-
-### Step 5 — End-user-specific (for end-user, 10 min)
-
-- «How often do you use [category]? When was the last time?»
-- «What annoys you? What do you complain about to colleagues?»
-- «What workarounds do you use?»
-- «What would make your work 30% easier?»
-
-### Step 6 — Wrap-up (5 min)
-
-- «What should I have asked but didn't?»
-- «Who else would you recommend talking to?»
-- Thank you + follow-up offer.
-
-### Step 7 — Post-interview (15 min, immediately after)
-
-1. Record top 3 quotes (verbatim).
-2. Mark moments of surprise / contradiction.
-3. Add to Discovery log: date, segment, role, top pains, evidence quotes, open questions.
-
-## Validation (Quality Gate)
-
-- [ ] Buyer vs end-user interviews are separated
-- [ ] Minimum 5 interviews per role (or explicit rationale)
-- [ ] Context through recent specific case, not generic
-- [ ] Pain discovery — without leading questions
-- [ ] Verbatim quotes recorded
-- [ ] Surprises / contradictions noted
-- [ ] Wrap-up done with «what else?»
-
-## Handoff
-
-The result is input for:
-- **`jtbd-canvas`** — jobs hypothesis → validation through quotes
-- **`problem-statement`** — evidence for actor/context/pain
-- **`assumption-mapping`** — new assumptions from surprises
-- **Discovery agent** — compilation into Discovery Brief
-
-Transfer format: interview log (markdown) with segmentation, top quotes, open questions. Via `$handoff`.
-
-## Anti-patterns
-
-| Error | Why it's bad | How to do it right |
-|-------|-------------|-------------------|
-| Leading questions | Distort data | Open-ended, about specific case |
-| Mixing buyer and end-user in one script | Different JTBD | Separate scripts |
-| Pitching instead of listening | Interview → demo | No product talk in opening |
-| Hypothetical questions | Don't predict behavior | Ask about past behavior |
-| Agreeing to vague requests | «I'd like automation» — empty insight | Dig to specific workflow |
-| Once or twice and done | Saturation ≠ 2 interviews | 5+ per role, until repeat patterns |
-
-## Interview Log Template
-
-```markdown
-# Interview: [Name or anonymized ID]
-- **Date:** YYYY-MM-DD
-- **Segment:** SMB / Mid-market / Enterprise
-- **Role:** Buyer / End-user / Champion / Admin
-- **Product context:** [current stack, alternatives]
-
-## Top Pains (ranked)
-1. [Pain] — evidence: «[verbatim quote]»
-
-## Top Quotes (verbatim)
-> «[Quote]» — on topic X
-
-## Surprises / contradictions
-- [What was unexpected]
-
-## Open questions (for next session)
-- [Question]
-```
-
-## Worked Example — TeamFlow (B2B SaaS, HR-tech)
-
-**Context:** TeamFlow — B2B SaaS for performance management (OKR + 1:1 notes + feedback + reviews). $8M ARR, 200 customers (120 SMB, 70 mid-market, 10 enterprise). NRR 105%, gross churn 8%. Considering initiative: **AI-powered 1:1 note summarization**. Discovery cycle.
-
-**Interview plan (initial):**
-- **Buyers (4):** VP People at 2 mid-market customers + CPO at 1 enterprise + VP People at 1 prospect
-- **End-users (4):** Manager with 8 reports (mid-market), manager with 5 reports (SMB), manager with 12 reports (enterprise), IC on the receiving end
-- Duration: 45 min each
-- Format: 2-on-1 (lead + conotetaker)
-
-### Interview log — Interview #3 (End-user, Manager, Mid-market)
-
-```markdown
-# Interview: E3 (anonymized: Marcus, Engineering Manager)
-- **Date:** 2026-04-08
-- **Segment:** Mid-market (250 employees, B2B SaaS)
-- **Role:** End-user (Engineering Manager, 8 direct reports)
-- **Product context:** Active TeamFlow user 18 months, also uses Notion for personal notes, Asana for team tasks
-- **Interviewer:** Alex K. (PM), with notetaker Priya S.
-
-## Opening (5 min)
-«Marcus, today I want to understand your 1:1 workflow — not about TeamFlow specifically, but about how your day with 1:1s is structured.» [user relaxed, confirmed permission to record]
-
-## Context Building (10 min) — recent specific case
-
-**Q:** «Tell me about the last time you prepared for a 1:1. What happened? Step by step.»
-
-**Marcus:** «Yesterday, with Sarah [report]. Usually I... look in TeamFlow at previous 1:1 notes, then in Slack for the past week to remember what we discussed, then in Asana to look at blockers on her tasks, then I write 3-5 bullet points in TeamFlow as the agenda. That takes about 20-25 minutes. Before I didn't do this — just came and asked "how's it going" — but that was a disaster, we didn't discuss anything concrete.»
-
-**Q:** «And what about action items from previous 1:1s?»
-
-**Marcus:** *[long pause]* «Honestly? I sometimes forget to check. Sometimes TeamFlow shows them in the summary, but if I don't mark them as completed — they just accumulate. I probably have 20-30 open action items, half of which are no longer relevant... [laughs] and generally already forgotten.»
-
-*[Interviewer note: surprise — user acknowledges that action items tracking is weak, but wants automatic reminders]*
-
-## Pain Discovery (15 min)
-
-**Q:** «What was difficult about that case with Sarah? What would you like to change?»
-
-**Marcus:** «The main thing is 25 minutes of prep every week for each of my 8 reports. That's ~3 hours a week just on preparation. Plus 8 hours on the 1:1s themselves. Plus about an hour on follow-up notes. 12 hours a week goes to this. I have a 40-hour week — 30% of my time on 1:1s. Management thinks this is an investment, but honestly I sometimes regret the overhead.»
-
-**Q:** «What takes up the most time that you don't like?»
-
-**Marcus:** «Notes. I type slowly. During the meeting — it's impossible to keep the conversation going + type. After the meeting — it's hard to remember exactly what was said. I often write only headlines and lose context.»
-
-**Q:** «When was the last time you complained about this?»
-
-**Marcus:** *[laughs]* «Honestly? Two days ago. I skipped prep for a 1:1 with a Junior engineer because I was too loaded with code review. The meeting was messy, he was clearly disappointed. I felt guilty — parent-of-the-year syndrome.»
-
-*[Key verbatim quote]*
-
-**Q:** «If you had a magic wand — what would you fix?»
-
-**Marcus:** «Mmm... if I could just speak aloud during the meeting, and the system would write notes itself, highlight action items, and the next week remind me of them — I would pay for that personally. Right now I do something similar through Otter.ai transcription, but that's just a transcript, not a summary. And it's not connected to TeamFlow.»
-
-*[GOLD — this is exactly what we're prototyping]*
-
-**Q:** «Have you tried anything similar?»
-
-**Marcus:** «Otter.ai — transcribes but doesn't summarize in an actionable way. Copy-paste into ChatGPT to summarize — tried it, but it's risky with confidential conversations. I'm not sure where ChatGPT stores data.»
-
-*[Emergent: privacy concern — critical for HR use case]*
-
-## End-user-specific (10 min)
-
-**Q:** «How often do you use TeamFlow? When was the last time?»
-
-**Marcus:** «Daily. Before each 1:1, after each 1:1, weekly for the review section.»
-
-**Q:** «What annoys you? What do you complain about to colleagues?»
-
-**Marcus:** «Action items. Notes — everyone writes them differently. Between managers there's no consistency. My team gets a different experience if they're moved to a different pod.»
-
-*[Organizational pain — great quote for buyer interview]*
-
-**Q:** «What would make your work 30% easier?»
-
-**Marcus:** «Automated note-taking + action item tracking + reminder in a prep email before the next 1:1. This reliably, with privacy guarantees.»
-
-## Wrap-up (5 min)
-
-**Q:** «What should I have asked but didn't?»
-
-**Marcus:** «About cross-manager visibility. How does the skip-level know that my 1:1s are good quality? Right now they're in a vacuum. An aggregate metric would be useful.»
-
-*[Another emergent topic — org-level visibility for skip-levels / VP HR]*
-
-**Q:** «Who else would you recommend talking to?»
-
-**Marcus:** «My skip-level, Jennifer — she would talk about what she does now to understand team health. And probably Sarah (my report) — to hear from her side how she perceives our 1:1s.»
-
----
-
-## Top Pains (ranked)
-
-1. **45-60 min/week on prep + notes + follow-up per report.** Overall 3-4 hours a week with 8 reports just on prep and note-taking. — evidence: «25 minutes of prep every week for each of my 8 reports»
-2. **Notes quality inconsistency** — typing during the meeting → lose conversation, after the meeting → forget context. — evidence: «I type slowly. During the meeting — it's impossible to keep the conversation going + type»
-3. **Action items drop rate ~40%** — forgotten, accumulated, never marked complete. — evidence: «I probably have 20-30 open action items, half of which are no longer relevant»
-4. **Privacy concern with generic AI tools** (Otter.ai + ChatGPT) — confidential HR conversations in untrusted destinations. — evidence: «I'm not sure where ChatGPT stores data»
-5. **Cross-manager inconsistency** — moving between managers = different 1:1 experience. — evidence: «Between managers there's no consistency»
-
-## Top Quotes (verbatim)
-
-> «If I could just speak aloud during the meeting, and the system would write notes itself, highlight action items, and the next week remind me of them — I would pay for that personally.» — on willingness to pay
-
-> «25 minutes of prep every week for each of my 8 reports. That's ~3 hours a week.» — on admin overhead
-
-> «I probably have 20-30 open action items, half of which are no longer relevant.» — on action items drop rate
-
-> «Copy-paste into ChatGPT to summarize — tried it, but it's risky with confidential conversations.» — on privacy
-
-## Surprises / contradictions
-
-- **Surprise:** Marcus already using Otter.ai + ChatGPT workaround (not something initial hypothesis considered). Suggests market education is not needed — workflows are already evolving, just without a safe tool.
-- **Contradiction:** says action items tracking is weak, but wants automatic reminders — means he values structure but doesn't have the will to maintain it manually. AI helper can close this gap.
-- **Emergent topic:** org-level visibility (skip-level, VP HR view of team 1:1 health) — this is buyer pain, but end-user surfaced it.
-
-## Open questions (for next session)
-
-1. Is willingness-to-pay ($X premium) validated? Marcus said «paid personally» — but his company pays. Check buyer interview for enterprise pricing model.
-2. Privacy / compliance: does TeamFlow AI feature need SOC 2 type II + HIPAA for healthcare customers? Check with enterprise customers.
-3. Scope: just 1:1s or all meetings? 5 other interviews showed the same pain for team meetings. Risk scope creep.
-4. Build vs Buy: OpenAI API works, but data residency issues for EU. Explore Azure OpenAI, Anthropic via AWS Bedrock, or fine-tuned OSS (Llama 3, Mistral).
-```
-
-> **Lesson from this interview:** open-ended questions about a specific case ("last time you did X") yield gold. Magic wand question revealed exact feature hypothesis + willingness to pay. Emergent topics (org-level visibility, privacy concern) appeared because the interviewer didn't ask about them directly — gave room for them to surface.
+---
+name: user-interview-script
+description: Structured script for B2B interviews — separation of buyer and end-user, pain-discovery without leading questions
+type: triggered
+domain: product
+owners:
+  - discovery
+gates:
+  - DISCOVERY
+tech: []
+topic: []
+triggers:
+  - "user-interview-script"
+  - "user interview"
+  - "пользовательское интервью"
+  - "B2B interview"
+  - "customer discovery"
+related: []
+budget_lines: 314
+schema_version: 1
+---
+# User Interview Script (B2B)
+
+> **Category:** Discovery  ·  **Slug:** `user-interview-script`
+
+## When to Use
+
+- Before a major product initiative, to capture real pain points and JTBD.
+- When preparing `/ship-right-thing` or `/shape-prioritize` — as the foundation of the Discovery Brief.
+- When validating new ideas / hypotheses from `$opportunity-solution-tree`.
+- When customer support / CRM data is insufficient to understand root cause.
+
+## Input
+
+| Field | Required | Description |
+|-------|:--------:|-------------|
+| Segment | ✅ | SMB / mid-market / enterprise |
+| Interviewee role(s) | ✅ | Buyer / end-user / admin / champion — separately |
+| Product context | ✅ | Current product, category, alternatives |
+| Hypotheses / areas of interest | ✅ | What we want to learn |
+| Duration | ⬚ | Default 30-45 min |
+| Format | ⬚ | 1-on-1 / 2-on-1 (usually 2: one leads, one takes notes) |
+
+> If segment and roles are not defined — **request from the user**.
+
+## Data Sources
+
+1. Existing interviews (mempalace, support tickets, NPS comments) — avoid duplicating questions.
+2. CRM data — deal status, cohort, usage patterns.
+3. Product analytics — understand how the product is really used (don't rely on «how they say» they use it).
+
+### Related Skills
+
+| Skill | What we take | When to invoke |
+|-------|-------------|----------------|
+| `jtbd-canvas` | Jobs structure → questions for discovery | After formulating job hypotheses |
+| `problem-statement` | Problem draft → focused questions | If problem is already formulated |
+| `assumption-mapping` | Top assumptions → validation questions | For high-risk assumptions |
+| `opportunity-solution-tree` | Opportunities → exploration questions | When discovering new opportunities |
+
+## Protocol
+
+### Step 0 — Segmentation
+
+Split interviews by role, **do not mix buyer and end-user in one script**:
+
+- **Buyer** (VP Engineering, Director of Ops, CIO) — ROI, compliance, fit with current stack, purchasing concerns.
+- **End-user** (developer, HR manager, sales rep) — daily experience, friction points, workarounds.
+- **Champion** — who promotes the product internally, their motivations.
+- **Admin** — who configures, integrates, monitors compliance.
+
+### Step 1 — Opening (5 min)
+
+Goal — establish context, not sell.
+
+1. Introduce yourself: «I work on [...] at [company]. Today I want to learn how you [context], not to sell anything.»
+2. Permission to record (if applicable).
+3. Warm-up: «Tell me in 2-3 sentences — what does your day look like? What's your role on the team?»
+4. **Rule:** never mention the product first.
+
+### Step 2 — Context Building (10 min)
+
+Understand the interviewee's workflow.
+
+Framework questions:
+- «Tell me about the last time you [situation]. What happened? Step by step.»
+- «How do you currently solve [proximate problem]?»
+- «Who else is involved? Who makes the decision?»
+- «What tools do you use?»
+
+**Rule:** ask about a specific recent case, not abstract generalizations.
+
+### Step 3 — Pain Discovery (15 min)
+
+Don't ask directly «what hurts». Look for evidence:
+
+- «What was difficult about that case?»
+- «What would you like to change?»
+- «When was the last time you complained about this?»
+- «What takes up the most time that you don't like?»
+- «If you had a magic wand — what would you fix?»
+
+**Red flags:**
+- Leading question: «You find [X] difficult, right?»
+- Hypothetical: «If we built [feature] — would you use it?»
+- Direct feature request: «Do you need [feature]?» → dig down to why.
+
+### Step 4 — Buyer-specific (for buyer, 10 min)
+
+- «Who is involved in the decision? (procurement, finance, legal, IT)»
+- «How do you evaluate ROI?»
+- «What made you look at [category] now?»
+- «What compliance / security requirements do you have?»
+- «What is your budget cycle?»
+- «What didn't work with similar tools in the past?»
+
+### Step 5 — End-user-specific (for end-user, 10 min)
+
+- «How often do you use [category]? When was the last time?»
+- «What annoys you? What do you complain about to colleagues?»
+- «What workarounds do you use?»
+- «What would make your work 30% easier?»
+
+### Step 6 — Wrap-up (5 min)
+
+- «What should I have asked but didn't?»
+- «Who else would you recommend talking to?»
+- Thank you + follow-up offer.
+
+### Step 7 — Post-interview (15 min, immediately after)
+
+1. Record top 3 quotes (verbatim).
+2. Mark moments of surprise / contradiction.
+3. Add to Discovery log: date, segment, role, top pains, evidence quotes, open questions.
+
+## Validation (Quality Gate)
+
+- [ ] Buyer vs end-user interviews are separated
+- [ ] Minimum 5 interviews per role (or explicit rationale)
+- [ ] Context through recent specific case, not generic
+- [ ] Pain discovery — without leading questions
+- [ ] Verbatim quotes recorded
+- [ ] Surprises / contradictions noted
+- [ ] Wrap-up done with «what else?»
+
+## Handoff
+
+The result is input for:
+- **`jtbd-canvas`** — jobs hypothesis → validation through quotes
+- **`problem-statement`** — evidence for actor/context/pain
+- **`assumption-mapping`** — new assumptions from surprises
+- **Discovery agent** — compilation into Discovery Brief
+
+Transfer format: interview log (markdown) with segmentation, top quotes, open questions. Via `$handoff`.
+
+## Anti-patterns
+
+| Error | Why it's bad | How to do it right |
+|-------|-------------|-------------------|
+| Leading questions | Distort data | Open-ended, about specific case |
+| Mixing buyer and end-user in one script | Different JTBD | Separate scripts |
+| Pitching instead of listening | Interview → demo | No product talk in opening |
+| Hypothetical questions | Don't predict behavior | Ask about past behavior |
+| Agreeing to vague requests | «I'd like automation» — empty insight | Dig to specific workflow |
+| Once or twice and done | Saturation ≠ 2 interviews | 5+ per role, until repeat patterns |
+
+## Interview Log Template
+
+```markdown
+# Interview: [Name or anonymized ID]
+- **Date:** YYYY-MM-DD
+- **Segment:** SMB / Mid-market / Enterprise
+- **Role:** Buyer / End-user / Champion / Admin
+- **Product context:** [current stack, alternatives]
+
+## Top Pains (ranked)
+1. [Pain] — evidence: «[verbatim quote]»
+
+## Top Quotes (verbatim)
+> «[Quote]» — on topic X
+
+## Surprises / contradictions
+- [What was unexpected]
+
+## Open questions (for next session)
+- [Question]
+```
+
+## Worked Example — TeamFlow (B2B SaaS, HR-tech)
+
+**Context:** TeamFlow — B2B SaaS for performance management (OKR + 1:1 notes + feedback + reviews). $8M ARR, 200 customers (120 SMB, 70 mid-market, 10 enterprise). NRR 105%, gross churn 8%. Considering initiative: **AI-powered 1:1 note summarization**. Discovery cycle.
+
+**Interview plan (initial):**
+- **Buyers (4):** VP People at 2 mid-market customers + CPO at 1 enterprise + VP People at 1 prospect
+- **End-users (4):** Manager with 8 reports (mid-market), manager with 5 reports (SMB), manager with 12 reports (enterprise), IC on the receiving end
+- Duration: 45 min each
+- Format: 2-on-1 (lead + conotetaker)
+
+### Interview log — Interview #3 (End-user, Manager, Mid-market)
+
+```markdown
+# Interview: E3 (anonymized: Marcus, Engineering Manager)
+- **Date:** 2026-04-08
+- **Segment:** Mid-market (250 employees, B2B SaaS)
+- **Role:** End-user (Engineering Manager, 8 direct reports)
+- **Product context:** Active TeamFlow user 18 months, also uses Notion for personal notes, Asana for team tasks
+- **Interviewer:** Alex K. (PM), with notetaker Priya S.
+
+## Opening (5 min)
+«Marcus, today I want to understand your 1:1 workflow — not about TeamFlow specifically, but about how your day with 1:1s is structured.» [user relaxed, confirmed permission to record]
+
+## Context Building (10 min) — recent specific case
+
+**Q:** «Tell me about the last time you prepared for a 1:1. What happened? Step by step.»
+
+**Marcus:** «Yesterday, with Sarah [report]. Usually I... look in TeamFlow at previous 1:1 notes, then in Slack for the past week to remember what we discussed, then in Asana to look at blockers on her tasks, then I write 3-5 bullet points in TeamFlow as the agenda. That takes about 20-25 minutes. Before I didn't do this — just came and asked "how's it going" — but that was a disaster, we didn't discuss anything concrete.»
+
+**Q:** «And what about action items from previous 1:1s?»
+
+**Marcus:** *[long pause]* «Honestly? I sometimes forget to check. Sometimes TeamFlow shows them in the summary, but if I don't mark them as completed — they just accumulate. I probably have 20-30 open action items, half of which are no longer relevant... [laughs] and generally already forgotten.»
+
+*[Interviewer note: surprise — user acknowledges that action items tracking is weak, but wants automatic reminders]*
+
+## Pain Discovery (15 min)
+
+**Q:** «What was difficult about that case with Sarah? What would you like to change?»
+
+**Marcus:** «The main thing is 25 minutes of prep every week for each of my 8 reports. That's ~3 hours a week just on preparation. Plus 8 hours on the 1:1s themselves. Plus about an hour on follow-up notes. 12 hours a week goes to this. I have a 40-hour week — 30% of my time on 1:1s. Management thinks this is an investment, but honestly I sometimes regret the overhead.»
+
+**Q:** «What takes up the most time that you don't like?»
+
+**Marcus:** «Notes. I type slowly. During the meeting — it's impossible to keep the conversation going + type. After the meeting — it's hard to remember exactly what was said. I often write only headlines and lose context.»
+
+**Q:** «When was the last time you complained about this?»
+
+**Marcus:** *[laughs]* «Honestly? Two days ago. I skipped prep for a 1:1 with a Junior engineer because I was too loaded with code review. The meeting was messy, he was clearly disappointed. I felt guilty — parent-of-the-year syndrome.»
+
+*[Key verbatim quote]*
+
+**Q:** «If you had a magic wand — what would you fix?»
+
+**Marcus:** «Mmm... if I could just speak aloud during the meeting, and the system would write notes itself, highlight action items, and the next week remind me of them — I would pay for that personally. Right now I do something similar through Otter.ai transcription, but that's just a transcript, not a summary. And it's not connected to TeamFlow.»
+
+*[GOLD — this is exactly what we're prototyping]*
+
+**Q:** «Have you tried anything similar?»
+
+**Marcus:** «Otter.ai — transcribes but doesn't summarize in an actionable way. Copy-paste into ChatGPT to summarize — tried it, but it's risky with confidential conversations. I'm not sure where ChatGPT stores data.»
+
+*[Emergent: privacy concern — critical for HR use case]*
+
+## End-user-specific (10 min)
+
+**Q:** «How often do you use TeamFlow? When was the last time?»
+
+**Marcus:** «Daily. Before each 1:1, after each 1:1, weekly for the review section.»
+
+**Q:** «What annoys you? What do you complain about to colleagues?»
+
+**Marcus:** «Action items. Notes — everyone writes them differently. Between managers there's no consistency. My team gets a different experience if they're moved to a different pod.»
+
+*[Organizational pain — great quote for buyer interview]*
+
+**Q:** «What would make your work 30% easier?»
+
+**Marcus:** «Automated note-taking + action item tracking + reminder in a prep email before the next 1:1. This reliably, with privacy guarantees.»
+
+## Wrap-up (5 min)
+
+**Q:** «What should I have asked but didn't?»
+
+**Marcus:** «About cross-manager visibility. How does the skip-level know that my 1:1s are good quality? Right now they're in a vacuum. An aggregate metric would be useful.»
+
+*[Another emergent topic — org-level visibility for skip-levels / VP HR]*
+
+**Q:** «Who else would you recommend talking to?»
+
+**Marcus:** «My skip-level, Jennifer — she would talk about what she does now to understand team health. And probably Sarah (my report) — to hear from her side how she perceives our 1:1s.»
+
+---
+
+## Top Pains (ranked)
+
+1. **45-60 min/week on prep + notes + follow-up per report.** Overall 3-4 hours a week with 8 reports just on prep and note-taking. — evidence: «25 minutes of prep every week for each of my 8 reports»
+2. **Notes quality inconsistency** — typing during the meeting → lose conversation, after the meeting → forget context. — evidence: «I type slowly. During the meeting — it's impossible to keep the conversation going + type»
+3. **Action items drop rate ~40%** — forgotten, accumulated, never marked complete. — evidence: «I probably have 20-30 open action items, half of which are no longer relevant»
+4. **Privacy concern with generic AI tools** (Otter.ai + ChatGPT) — confidential HR conversations in untrusted destinations. — evidence: «I'm not sure where ChatGPT stores data»
+5. **Cross-manager inconsistency** — moving between managers = different 1:1 experience. — evidence: «Between managers there's no consistency»
+
+## Top Quotes (verbatim)
+
+> «If I could just speak aloud during the meeting, and the system would write notes itself, highlight action items, and the next week remind me of them — I would pay for that personally.» — on willingness to pay
+
+> «25 minutes of prep every week for each of my 8 reports. That's ~3 hours a week.» — on admin overhead
+
+> «I probably have 20-30 open action items, half of which are no longer relevant.» — on action items drop rate
+
+> «Copy-paste into ChatGPT to summarize — tried it, but it's risky with confidential conversations.» — on privacy
+
+## Surprises / contradictions
+
+- **Surprise:** Marcus already using Otter.ai + ChatGPT workaround (not something initial hypothesis considered). Suggests market education is not needed — workflows are already evolving, just without a safe tool.
+- **Contradiction:** says action items tracking is weak, but wants automatic reminders — means he values structure but doesn't have the will to maintain it manually. AI helper can close this gap.
+- **Emergent topic:** org-level visibility (skip-level, VP HR view of team 1:1 health) — this is buyer pain, but end-user surfaced it.
+
+## Open questions (for next session)
+
+1. Is willingness-to-pay ($X premium) validated? Marcus said «paid personally» — but his company pays. Check buyer interview for enterprise pricing model.
+2. Privacy / compliance: does TeamFlow AI feature need SOC 2 type II + HIPAA for healthcare customers? Check with enterprise customers.
+3. Scope: just 1:1s or all meetings? 5 other interviews showed the same pain for team meetings. Risk scope creep.
+4. Build vs Buy: OpenAI API works, but data residency issues for EU. Explore Azure OpenAI, Anthropic via AWS Bedrock, or fine-tuned OSS (Llama 3, Mistral).
+```
+
+> **Lesson from this interview:** open-ended questions about a specific case ("last time you did X") yield gold. Magic wand question revealed exact feature hypothesis + willingness to pay. Emergent topics (org-level visibility, privacy concern) appeared because the interviewer didn't ask about them directly — gave room for them to surface.