PM Analyst Skill

The pragmatic product manager's workflow - from data to shipping.

When to Use

Use this skill when you need to:

• Answer product questions with data ("How did users react to our 1:1 messages?")
• Identify the #1 product opportunity from analysis
• Create a Linear ticket with complete requirements
• Audit that ticket to ship 80% of value at 20% of complexity

The PM Analyst Philosophy

Core Principles

1. Data-driven, not opinion-driven

   • Start with real user behavior
   • Measure what matters
   • Let data reveal opportunities

2. Impact > Effort

   • Identify the 80/20 solution
   • Ship small, learn fast
   • Avoid premature optimization

3. Ship to learn

   • Perfect is the enemy of shipped
   • Prefer MVPs over big designs
   • Measure, then iterate

4. Lean startup thinking

   • Smallest change that tests hypothesis
   • Gradual rollouts
   • Decision gates based on data

The Workflow

Step 1: Analyze (Data → Insights)

Use skills:

• funnel-analysis - For activation/retention questions ("show me the funnel")
• sql-reader + product-analytics - For general product questions
• therapist-data-scientist - For therapeutic metrics

Question format:

• "Show me the funnel" → use funnel-analysis
• "How did users react to our 1:1 messages?" → use sql-reader
• "What % of users are power users?" → use funnel-analysis

Output: Comprehensive analysis markdown file

Quality checks:

• ✅ Queried production database
• ✅ Sample size documented
• ✅ Patterns identified and named
• ✅ Qualitative examples included
• ✅ Failure modes analyzed
• ✅ Statistical context provided

Example: codel_1on1_reactions_analysis.md

Step 2: Identify Opportunity (Insights → Priority)

Framework: Apply PM judgment to analysis

Questions to answer:

1. What's the #1 insight?

   • Which finding has highest impact?
   • What creates most user value?
   • What's unique to our product?

2. Is this worth building?

   • How many users affected?
   • How big is the pain?
   • What's the revenue impact?

3. Can we ship it?

   • What's the technical complexity?
   • Do we have the resources?
   • What's the risk?

Prioritization matrix:

Output: 1-2 paragraph justification of top opportunity

Example:

"Meta-Commentary (🌟) is the #1 opportunity. It's already our highest-performing feature (85.9% loved), but we're only celebrating ~5-10% of positive moments due to a 6-hour cooldown. Reducing to 1-hour cooldown is HIGH IMPACT (2-3x more of what users love) and LOW EFFORT (change one SQL condition). This is our product differentiation—'coach watching over your shoulder' in real-time."

Step 3: Spec the Feature (Opportunity → Requirements)

Use skill: feature-spec-writer

Template: Use "Quick Spec for Linear" format

Requirements:

## Problem
{1-2 sentences: What user pain are we solving?}

## Proposed Solution (MVP)
{1-2 sentences: What's the SMALLEST change that delivers value?}

## Success Metrics
- {Metric 1}: {Current} → {Target}
- {Metric 2}: {Current} → {Target}

## Implementation
{3-5 bullet points: What needs to change?}

## Non-Goals
{What are we explicitly NOT doing in v1?}

Output: Feature spec markdown

Step 4: Create Linear Ticket

Use skill: linear-manager

Format:

export LINEAR_API_KEY={key_from_arsenal/.env}

# Get team ID
.claude/skills/linear-manager/scripts/get_teams.sh

# Create ticket
.claude/skills/linear-manager/scripts/create_issue.sh \
  --title "{Concise title with key metrics/changes}" \
  --team-id "{team-uuid}" \
  --priority "{high/medium/low}" \
  --description "{Feature spec from Step 3}"

Title format: "{Action}: {Key change} ({Metric})"

• ✅ Good: "Reduce PR cooldown from 6h to 1h (2-3x increase)"
• ❌ Bad: "Improve positive reinforcement"

Priority guide:

• High: Affects many users, high impact, low risk
• Medium: Moderate impact or effort
• Low: Nice-to-have, low impact

Output: Linear ticket URL (CODEL-XXX)

Step 5: AUDIT THE TICKET 🔍

This is where PMs fail. They create elaborate tickets without asking:

• Is this the simplest version?
• What can we cut and still deliver 80% of value?
• Are we solving real problems or theoretical ones?

Audit Framework

Read the ticket and ask:

1. Scope Creep Check

• How many phases/sprints?
• Is Phase 1 independently valuable?
• Can we ship Phase 1 and stop there?

If >1 sprint: → Challenge: "Why can't we ship Phase 1 alone?"

2. Premature Optimization Check

• Are we building infrastructure for hypothetical problems?
• Are we solving edge cases before validating core value?
• Do we have user evidence this complexity is needed?

Red flags:

• "Variety tracking for power users" (before increasing volume)
• "A/B testing framework" (before simple change)
• "User preference system" (before knowing preferences vary)

3. Power User Bias Check

• What % of users are power users?
• Are we optimizing for the 1% or the 99%?
• Does the ticket clearly state user distribution?

If optimizing for outliers: → Challenge: "Should we solve for typical users first?"

4. Evidence-Based Decision Check

• Do we have data showing this problem exists?
• Or are we assuming users will have this problem?
• Can we ship small and THEN decide?

Principle: Ship to learn, don't build to guess

5. Complexity Cost Check

• What's the eng effort? (days/weeks/months)
• What's the opportunity cost? (what else could we build?)
• Is the ROI worth it?

Calculate: (Value delivered) / (Eng weeks spent)

Audit Output Format

## 🚨 TICKET AUDIT (PM's Boss Perspective)

### Original Scope
- **Effort:** {X sprints / Y eng weeks}
- **Value:** {Expected impact}
- **Phases:** {Number}

### Critical Questions
1. **{Question 1}** - {Challenge assumption}
2. **{Question 2}** - {Challenge assumption}
3. **{Question 3}** - {Challenge assumption}

### What to Cut
❌ **Phase X:** {Why this is premature optimization}
❌ **Feature Y:** {Why this solves theoretical problem}
❌ **Infrastructure Z:** {Why this is overengineering}

### 80/20 Solution
✅ **Ship ONLY:** {Minimal scope}
✅ **Effort:** {Reduced effort}
✅ **Value:** {80% of original value}

### Revised Scope
{Simplified ticket description}

### Decision Gate
**After {timeframe}, measure:**
- {Metric 1}
- {Metric 2}

**Then decide:**
- ✅ If {condition} → Ship and move on
- ✅ If {condition} → Consider Phase 2
- ✅ If {condition} → Rollback

### Cost/Benefit
- **Original:** {X weeks} for {Y value} = {Z ROI}
- **Simplified:** {A weeks} for {B value} = **{C ROI}** (better)

Complete Workflow Example

User asks: "How did users react to our 1:1 messages?"

Step 1: Analyze

# Use therapist-data-scientist skill
# Query: message_reactions table
# Output: codel_1on1_reactions_analysis.md

Key findings:

• 71 reactions total (0.7% rate)
• 85.9% loved, 14.1% liked, 4.2% disliked
• Meta-Commentary (MC) highest-performing type
• craig (1 user) = 49% of all reactions

Step 2: Identify Opportunity

Analysis:

• MC is already best feature (85.9% loved)
• Currently under-utilized (~5-10% coverage)
• 6-hour cooldown limits frequency
• Opportunity: Celebrate 3-4x MORE growth moments

Priority: 🔥 HIGH IMPACT + LOW EFFORT = SHIP NOW

Step 3: Spec the Feature

## Problem
Positive Reinforcement (🌟) is highest-performing message type,
but 6-hour cooldown limits coverage to ~5-10% of positive moments.

## Proposed Solution (MVP)
Reduce cooldown from 6 hours to 1 hour.

## Success Metrics
- PR volume: 2-3x increase
- Love rate: Maintain >80%
- No spam complaints

## Implementation
1. Change SQL condition: `last_6_hours` → `last_1_hour`
2. Deploy to staging (3 days)
3. Deploy to prod gradual rollout (25%→50%→100%)
4. Measure for 4 weeks

## Non-Goals
- ❌ Expanding to Medium emotional depth (Phase 2)
- ❌ Variety tracking infrastructure (Phase 3)
- ❌ Additional affect categories (Phase 4)

Step 4: Create Linear Ticket

.claude/skills/linear-manager/scripts/create_issue.sh \
  --title "Reduce PR cooldown from 6h to 1h (2-3x increase)" \
  --team-id "cf35cf6c-5d97-4123-a8d3-0f257364a795" \
  --priority "high" \
  --description "{spec from Step 3}"

Output: CODEL-803 created ✅

Step 5: Audit the Ticket

PM initially proposed:

• Phase 1: Reduce cooldown (1 sprint)
• Phase 2: Expand depth (1 sprint + A/B test)
• Phase 3: Variety tracking (2 sprints + database)
• Phase 4: Expand affects (1 sprint)
• Total: 4-6 sprints

Audit findings:

• Phase 1 alone delivers 80% of value
• Phases 2-4 are premature optimization
• No user evidence variety tracking is needed
• Building for craig (1 user) instead of 99%

Revised scope: Ship ONLY Phase 1

• Effort: 1 week (vs 4-6 sprints)
• Value: 80% of original (2-3x increase)
• ROI: 20x better

Decision gate: Measure for 4 weeks, then decide on Phase 2

PM Analyst Checklist

Before considering your work done:

Analysis Quality

• Used production database
• Sample size documented
• Patterns identified with clear names
• Qualitative examples included
• Failure modes analyzed
• Statistical context provided

Opportunity Identification

• #1 insight clearly stated
• Impact vs. Effort assessed
• User segment size known
• Competitive differentiation noted

Feature Spec

• Problem statement (1-2 sentences)
• MVP solution (simplest version)
• Success metrics with targets
• Implementation steps (3-5 bullets)
• Non-goals explicitly stated

Linear Ticket

• Created with linear-manager skill
• Descriptive title with metrics
• Appropriate priority set
• Team ID correct
• URL returned and tested

Audit Completion

• Challenged scope creep
• Identified premature optimization
• Checked for power user bias
• Verified evidence-based decisions
• Calculated complexity cost
• Proposed 80/20 solution
• Reduced effort by >50%
• Maintained >70% of value

Red Flags to Watch For

In Analysis

• ❌ Small sample sizes (<30)
• ❌ No qualitative examples
• ❌ Missing failure modes
• ❌ Power user skew not acknowledged

In Opportunity Selection

• ❌ Optimizing for outliers (1% of users)
• ❌ No clear success metric
• ❌ Theoretical problem (no user evidence)
• ❌ Low impact / high effort

In Specs

• ❌ "Big design up front"
• ❌ Multi-phase plans without decision gates
• ❌ Building infrastructure before validating need
• ❌ No non-goals section

In Tickets

• ❌ >2 sprints for Phase 1
• ❌ A/B testing before simple change
• ❌ Database migrations for unproven features
• ❌ Solving problems that don't exist yet

Integration with Other Skills

Analysis skills:

• sql-reader - Query production data
• funnel-analysis - User activation funnel (use for retention/activation questions)
• therapist-data-scientist - Calculate therapeutic metrics
• product-analytics - General analysis framework

Output skills:

• feature-spec-writer - Structure requirements
• linear-manager - Create tickets

Output Files

A complete pm-analyst run produces:

1. Analysis: {topic}_analysis.md

   • Data tables
   • Pattern analysis
   • Insights and recommendations

2. Audit: {topic}_audit.md OR append to analysis

   • Critical questions
   • What to cut
   • 80/20 solution
   • Revised scope

3. Linear ticket: CODEL-XXX

   • Created with simplified scope
   • Decision gate at 4 weeks
   • Clear success metrics

Notes

Why this skill exists:

Product managers often:

• Create elaborate multi-phase plans
• Build infrastructure for hypothetical problems
• Optimize for edge cases (power users)
• Skip the "ship small and learn" step

This skill enforces:

• Ship Phase 1 alone
• Measure for 4 weeks
• THEN decide on Phase 2
• Based on DATA, not assumptions

The philosophy:

"Perfect is the enemy of shipped. Ship 80% of value at 20% of complexity. Measure. Iterate."

Quick Reference

# Step 1: Analyze
# Use sql-reader or therapist-data-scientist
# Output: {topic}_analysis.md

# Step 2: Identify top opportunity
# Manual: Review analysis, pick #1 insight

# Step 3+4: Spec + Create ticket
export LINEAR_API_KEY={from arsenal/.env}
.claude/skills/linear-manager/scripts/get_teams.sh
.claude/skills/linear-manager/scripts/create_issue.sh \
  --title "{concise with metrics}" \
  --team-id "{uuid}" \
  --priority "{high/med/low}" \
  --description "{feature spec}"

# Step 5: Audit
# Review ticket through "PM's boss" lens
# Cut scope to 80/20 solution
# Add decision gate for Phase 2

Remember: The best PMs ship small and learn fast. Avoid "big design up front."