customer-conversations

Customer Conversations

Reference

• ID: S-PROD-08
• Category: Product
• Inputs: company.config.yaml, existing CONV- artifacts, ICP artifacts, PRDs
• Outputs: conversation log artifacts → artifacts/decision-memos/CONV-{date}-{type}-{company}.md
• Used by: Product Agent (primary), Ops & Risk Agent (churn signals)
• Tool scripts: ./tools/artifact/validate.sh

Purpose

Structures the process of talking to customers. Provides pre-call preparation (what to ask, what to listen for), captures post-call debriefs in a structured format, and routes insights to the right downstream skills (ICP refinement, PRD evidence, churn detection).

When to Use

• Before a customer call, demo, or interview — run prep <type> for a tailored briefing
• After any customer interaction — run debrief to capture structured insights
• When you have 3+ conversation logs — run patterns to surface cross-conversation themes

Conversation Types

Type	Purpose	Key Signals
discovery	Understand user problems, validate hypotheses	Pain intensity, current workarounds, willingness to pay
mom-test	Mom-Test methodology — validate pain without leading	Current behavior, $$ cost of status quo, workarounds tried
demo	Show the product, gauge interest	Feature reactions, objections, comparison mentions
churn-save	Understand why a user is leaving	Root cause, what would change their mind, competitor mentions
upsell	Explore expansion opportunities	Usage patterns, feature gaps, budget signals
check-in	Relationship maintenance, satisfaction pulse	NPS proxy, feature requests, referral willingness
cold-outreach	Reaching out to potential customers who haven't expressed interest yet	Pain points, willingness to pay signals, competitive mentions

Procedure

Cold Outreach Conversation Type

Cold Outreach conversation type:

• When: Reaching out to potential customers who haven't expressed interest yet
• Prep: Research the prospect, prepare problem-led opening, have 3 discovery questions ready
• During: Listen 80%, talk 20%. Goal is learning, not selling.
• Debrief: Capture pain points, willingness to pay signals, competitive mentions

Starting from Scratch Detection: If no ICP artifact exists in artifacts/product/ and no CONV- artifacts exist in artifacts/decision-memos/, present a "Starting from scratch" briefing:

• Guidance on finding first 10 targets (LinkedIn search, community lurking, competitor reviews)
• Starter interview script focused on problem discovery
• Link to icp-positioning skill to formalize learnings after 5+ conversations

Mom-Test Interview Methodology

When conversation type is mom-test, apply The Mom Test rules strictly:

Core Principle: Talk about their life, not your idea. Extract facts about past behavior, not opinions about the future.

The 5 Mom-Test Rules:

1. Never ask "Would you use this?" — People are polite liars. Instead ask: "How do you handle [problem] today?"
2. Never ask "Would you pay for this?" — Instead ask: "What does [current workaround] cost you in time/money?"
3. Ask about specifics in the past, not hypotheticals — "Tell me about the last time [problem] happened" vs. "Do you ever experience [problem]?"
4. Talk less, listen more — Target 80% listening, 20% talking. Follow their energy.
5. Extract commitment or advancement — End every call with a concrete next step (intro to a colleague, LOI review, pilot agreement).

Mom-Test Question Bank (use 5-7 per call, 15 min max):

• "What's the hardest part of [workflow] for you right now?"
• "Can you walk me through the last time that happened?"
• "What did you do about it? What else have you tried?"
• "How much time/money does that cost you per [week/month]?"
• "Who else in your org deals with this problem?"
• "What would a dream solution look like?" (only AFTER understanding the problem)
• "If you had a magic wand, what would you change about [current tool/process]?"
• "What are you using today to solve this? What's working? What's not?"
• "Have you looked for solutions? What happened?"
• "How urgent is solving this? What's your timeline?"

Anti-Patterns to Avoid:

• Pitching your solution before minute 10
• Asking binary yes/no questions
• Fishing for compliments ("Don't you think X would be great?")
• Talking about features before understanding the problem
• Accepting "that would be nice" as validation (it's not)

Debrief Extras for Mom-Test: When debriefing a mom-test conversation, also capture:

• $$ Quantification: estimated cost of the problem per month/year
• Workarounds tried: what they've already done to solve this
• Commitment signal: did they offer an intro, agree to review an LOI, or ask for a follow-up?

Mode: prep <type>

1. Load Context

   • Read company.config.yaml for product name, stage, ICP
   • Check artifacts/product/ for the most recent ICP positioning artifact
   • Check artifacts/prds/ for in-progress features (gives context on what's being built)
   • Check recent CONV- artifacts for patterns from this company/segment

2. Generate Pre-Call Briefing

   Produce a briefing document with:

   Do Ask (5 open-ended questions tailored to conversation type):

   • Discovery: "What's the hardest part of [workflow] for you right now?"
   • Demo: "Which of these capabilities matters most to your day-to-day?"
   • Churn-save: "What changed since you first signed up?"
   • Upsell: "If you could wave a magic wand and add one capability, what would it be?"
   • Check-in: "What's working well? What's frustrating?"

   Don't Ask (anti-patterns to avoid):

   • Leading questions ("Don't you think our feature X is great?")
   • Binary questions when you need depth ("Do you like it?" → "What's your experience with it?")
   • Feature-first questions before understanding the problem

   Listen For (signals specific to conversation type):

   • ICP fit signals: role, company size, tech stack, budget authority
   • Competitor mentions: who else they're evaluating or using
   • Churn risk indicators: frustration level, timeline pressure, alternative mentions
   • Willingness-to-pay signals: asking about pricing, comparing to current spend

   Context from Recent Conversations (if CONV- artifacts exist for this company):

   • Previous pain points mentioned
   • Commitments made
   • Open questions from last call

3. Present Briefing — show the briefing to the user before the call. Do not save as an artifact (ephemeral output).

Mode: debrief

4. Capture Structured Debrief

   Prompt the user to provide (or extract from conversation notes):

   Field	Description
   Date	When the conversation happened
   Participant	Name, role, company, segment
   Conversation type	discovery / demo / churn-save / upsell / check-in
   Pain level	1 (minor inconvenience) to 5 (hair-on-fire)
   Key quotes	2-3 verbatim quotes that capture the essence
   Willingness to pay	Strong signal / Weak signal / No signal / Price sensitive
   Competitor mentions	Any alternatives mentioned (by name)
   Churn signals	None / Low / Medium / High
   Upsell signals	None / Low / Medium / High
   Feature requests	Specific capabilities mentioned
   ICP fit	Strong / Partial / Weak — with reasoning
   Recommended next action	Follow-up call, send proposal, route to support, etc.

5. Classify and Tag

   Based on the debrief, assign tags:

   • prd_evidence: true — if the conversation surfaced a clear product need
   • icp_signal: true — if the conversation provided ICP refinement data
   • churn_signal: true — if churn risk was detected (Medium or High)
   • upsell_signal: true — if expansion opportunity was identified

6. Save Artifact

   Write to artifacts/decision-memos/CONV-{date}-{type}-{company}.md:

   ---
   id: CONV-{date}-{type}-{company}
   type: decision-memo
   title: "{Type} Conversation — {Company}"
   status: review
   created: {date}
   tags: [customer-conversation, {type}]
   prd_evidence: true|false
   icp_signal: true|false
   churn_signal: true|false
   ---
   

   Body contains the structured debrief fields above.

7. Route Signals

   After saving:

   • If churn_signal: true → inform the user: "Churn signal detected. Consider running /support-operations to review escalation paths, or /activation-onboarding to design a re-engagement flow."
   • If icp_signal: true → inform the user: "ICP data captured. Run /icp-positioning to update your ideal customer profile with this evidence."
   • If prd_evidence: true → inform the user: "Product evidence captured. This will be consumed by /feedback-synthesizer in the next synthesis cycle."

8. Validate — run ./tools/artifact/validate.sh on the saved artifact.

Mode: patterns

9. Cross-Conversation Analysis

   Scan all CONV- artifacts in artifacts/decision-memos/:

   • Count by conversation type
   • Extract all Key quotes fields
   • Tally pain levels by segment
   • List all competitor mentions with frequency
   • Identify recurring feature requests (3+ mentions = pattern)
   • Aggregate churn signals by segment

10. Surface Patterns

    Present a summary:

    • Top pain points: ranked by frequency × pain level
    • Competitor landscape: who's mentioned most, in what context
    • ICP signals: which segments show strongest fit
    • Emerging feature requests: requests that appear 3+ times
    • Churn risk segments: segments with concentrated churn signals

    This output is ephemeral (presented to user, not saved as artifact). The user can then feed specific findings into /feedback-synthesizer or /icp-positioning.

Quality Checklist

• [ ] Pre-call briefings include 5 open-ended questions (no leading questions)
• [ ] Debriefs capture all structured fields
• [ ] Artifact frontmatter uses type decision-memo (passes validate.sh)
• [ ] Signal routing recommendations are provided
• [ ] Pattern analysis only runs with 3+ CONV- artifacts
• [ ] Artifact passes validation

Cross-References

• feedback-synthesizer (S-PROD-04) — consumes CONV- artifacts as input during synthesis
• icp-positioning (S-PROD-01) — CONV- artifacts provide ICP evidence
• prd-writer (S-PROD-02) — CONV- artifacts populate PRD evidence sections
• support-operations (S-OPS-07) — churn signals feed support escalation review
• inbound-loop-sop — standards/ops/inbound-loop-sop.md documents the full pipeline