pattern-recognition

name	pattern-recognition
description	Master pattern recognition - identify recurring themes, outliers, trends, inconsistencies across documents and data. Use for extracting meaningful patterns from interviews, feedback, and data sets.
allowed-tools	Read, Grep
metadata	{"version":"2.0","claude-code":{"context":"fork","agent":"Explore","user-invocable":true}}

Pattern Recognition for BA

What do the numbers show? What's really happening?

What is Pattern Recognition?

Definition: Identifying recurring themes, anomalies, and trends across data sets, interviews, or documents - finding the signal in noise.

Why it matters:

• Hidden patterns: One-off observation vs actual trend
• Outlier detection: Which observations are unusual/important?
• Data-driven insights: "Many users said it" vs "A few users said it"
• Contradiction detection: "Users love feature X" but "Nobody uses feature X"
• Trend identification: Direction of change (improving/declining)

When to Use:

• ✅ Analyzing 5-8 user interviews (find common themes)
• ✅ Reviewing customer feedback (what's most mentioned?)
• ✅ Analyzing data sets (what's the trend?)
• ✅ Finding inconsistencies (why do findings conflict?)
• ✅ Identifying root causes (what's the common factor?)

---

3 Pattern Types

Type 1: Frequency Patterns (The 60% Rule)

What appeared in 60%+ of observations?

Interview Finding: "Users find setup confusing"

Count appearances:
Interview 1: User said "setup confusing" ✓
Interview 2: User said "takes too long" ✓
Interview 3: User said "confusing" ✓
Interview 4: User completed easily ✗
Interview 5: User said "confusing" ✓
Interview 6: User said "unclear" ✓
Interview 7: User completed in 5 min ✗
Interview 8: User said "complex" ✓

Count: 6 out of 8 mentioned confusion = 75%
→ PATTERN FOUND (60%+ threshold)

Conclusion: Setup confusion is widespread issue,
            not just one user's problem

---

Type 2: Contradiction Patterns

What conflicts with other observations?

Observation 1: "Users love the dashboard feature"
- Quote 1: "Dashboard gives me visibility I need"
- Quote 2: "Dashboard is my favorite feature"

But Data Shows: Dashboard usage is only 15% of users
                80% never open dashboard

CONTRADICTION FOUND!
→ Investigation: Maybe users love it when they use it,
                 but most don't discover it or use it
                 (awareness/discoverability problem, not feature problem)

---

Type 3: Outlier Patterns

What's unusual or stands out?

Sales Data:
- Sales rep A: $500K revenue (typical)
- Sales rep B: $520K revenue (typical)
- Sales rep C: $850K revenue (outlier - 60% higher!)
- Sales rep D: $480K revenue (typical)

Question: Why is Sales rep C so much higher?

Investigation:
→ Rep C focuses on Enterprise segment (higher deal size)
→ Other reps focus on SMB (smaller deals)

Insight: Enterprise segment is 60% higher value
→ Strategy implication: Shift towards Enterprise focus?

---

5 Pattern Recognition Techniques

1. Frequency Analysis

Count how many times something appears.

Process:
1. Review all interviews/data
2. Mark each instance of observation
3. Count total instances
4. Divide by total observations
5. Calculate percentage

Rules:
- Pattern threshold: 60%+ = significant pattern
- 30-60% = interesting trend, but not consensus
- <30% = minority view, document but don't prioritize

Example:
Pain point: "No real-time collaboration"
User 1: Mentioned ✓
User 2: Mentioned ✓
User 3: Mentioned ✓
User 4: Not mentioned ✗
User 5: Mentioned ✓
User 6: Mentioned ✓

5 of 6 mentioned = 83% → STRONG PATTERN

---

2. Segmentation Analysis

Does pattern appear in specific segments?

Finding: "Users frustrated with performance"

Segment analysis:
- Enterprise users: 80% frustrated with performance
- Mid-market users: 40% frustrated with performance
- SMB users: 20% frustrated with performance

Insight: Performance is enterprise problem primarily
         Not universal issue, segment-specific
→ Design: Enterprise scalability first, SMB later

---

3. Timeline Analysis

Does pattern change over time?

User churn by cohort:

Cohort A (Jan 2026): 5% monthly churn
Cohort B (Feb 2026): 6% monthly churn
Cohort C (Mar 2026): 12% monthly churn
Cohort D (Apr 2026): 15% monthly churn

Trend: Churn increasing for newer cohorts
→ Insight: Something is getting worse (product, onboarding, market fit)
→ Action: Investigate what changed

---

4. Correlation Analysis

Do two things happen together?

Observation: Users with low onboarding completion have high churn

Data:
Users completing <3 steps: 70% churn
Users completing 3-5 steps: 40% churn
Users completing all steps: 5% churn

Correlation Found: Onboarding completion → Churn reduction
→ Insight: Completing onboarding indicates product fit
→ Action: Make onboarding better to improve retention

---

5. Root Cause Analysis via Patterns

Use patterns to trace to cause.

Pattern: "Users abandon at step 3"

Investigation:
Interview review: 6 of 8 users mentioned "password step"
→ Deeper: When asked why, got:
   - "Requirements unclear"
   - "Password validation error"
   - "Confusing message"

Pattern: All mention password step
         All have confusion/clarity issues
→ ROOT CAUSE: Password UX is broken
→ Solution: Redesign password step for clarity

---

Common Pattern Analysis Mistakes

❌ Small Sample Size

Problem: "All 2 users said feature X is broken"
Issue: 2 users is too small, could be coincidence
Fix: Need 5-8 observations minimum to establish pattern

❌ Confirmation Bias

Problem: "I expected users to want feature A, so I count evidence
          supporting A and ignore evidence for B"
Fix: Count ALL observations objectively, not just ones you expected

❌ Selection Bias

Problem: "Interviewed only early adopters, they love feature"
Issue: Early adopters ≠ typical users, they have different needs
Fix: Interview diverse user segments

❌ Mixing Unrelated Observations

Problem: "Users mentioned: slow performance, confusing UI, needs
         integration, wants mobile. These are 4 separate issues."
Issue: Looking like one big pattern when they're separate
Fix: Categorize observations into distinct patterns

---

Pattern Recognition Workflow

Step 1: Prepare Data

• Gather all observations (interviews, feedback, metrics)
• Organize in consistent format
• Ensure sample size adequate (5-8+ minimum)

Step 2: Code Observations

• Read each observation
• Identify key themes/topics
• Label/tag by theme
• Note exact quote/evidence

Step 3: Tabulate

• Count frequency by theme
• Calculate percentages
• Identify patterns (60%+ threshold)
• Spot contradictions

Step 4: Investigate

• For each pattern, ask "Why?"
• Look for sub-patterns (is it segment-specific?)
• Compare with other data (does it correlate?)
• Identify root causes

Step 5: Document

• List patterns found
• Evidence for each (quotes, percentages)
• Implications/recommendations
• Remaining questions

---

Tools & Templates

• 📋 Pattern Analysis Template: assets/pattern-analysis-template.md
• 📋 Frequency Counting Worksheet: assets/frequency-template.md
• 📋 Coding Framework: assets/coding-framework-template.md
• 📋 Trend Analysis Template: assets/trend-analysis-template.md
• 📋 Contradiction Assessment: assets/contradiction-template.md

---

Usage Examples

/pattern-recognition "interview-transcripts" "Common pain points?"
/pattern-recognition "customer-feedback-data" "Recurring complaints?"
/pattern-recognition "usage-metrics" "What features get used most?"

---

BA Standards & References

Based on:

• Qualitative Analysis: Coding and theme identification
• Quantitative Analysis: Frequency and correlation
• Data Science: Pattern detection, outlier analysis
• User Research: Thematic analysis methodology

Detailed Frameworks:

• 📚 references/frequency-analysis.md - Counting and the 60% rule
• 📚 references/coding-framework.md - How to code observations
• 📚 references/theme-identification.md - Finding main themes
• 📚 references/outlier-detection.md - Finding and investigating anomalies
• 📚 references/correlation-causation.md - When patterns are related vs coincidence