Analysis Planner

Structure investigations before diving into data to prevent wasted time, ensure thoroughness, and deliver actionable insights.

Core Planning Framework

Phase 1: Define the Question

Start with: "What decision needs to be made?"

Quality checklist:

• [ ] Specific (not vague)
• [ ] Answerable with available data
• [ ] Tied to a decision or action
• [ ] Has clear success criteria
• [ ] Time-bounded

Good vs Bad:

• ✓ "Should we prioritize mobile app performance or new features next quarter?"
• ✓ "Which customer segment should we focus retention efforts on?"
• ✗ "Tell me about our users" (too broad)
• ✗ "Find something interesting" (no direction)

Phase 2: Define Success Criteria

Ask: "What would a good answer look like?"

Template:

A successful analysis will:
1. [Specific outcome]
2. [Specific outcome]

We'll know we're done when:
- [Criterion]
- [Criterion]

The answer will enable us to:
- [Decision/action that will be taken]

Phase 3: Generate Hypotheses

Process:

1. List potential drivers from KPI Tree
2. Add business context (recent changes, events)
3. Combine into testable hypotheses

Hypothesis quality criteria:

• [ ] Testable with available data
• [ ] Specific (not "something changed")
• [ ] Has clear validation method
• [ ] Mutually exclusive from others
• [ ] Collectively exhaustive

Prioritize by:

1. Probability: How likely?
2. Impact: How much does it explain?
3. Actionability: Can we do something about it?

Phase 4: Design Analysis Approach

For each hypothesis, provide detailed analysis steps in descriptive language (NOT SQL):

For each hypothesis:

1. Data to Collect - Describe in words what data points are needed (e.g., "number of sign-ups by day, with their marketing channel, time from sign-up to first action")
2. How to Analyze - Detail the visualization approach:

• Type of chart/table (line chart, bar chart, histogram, table, etc.)
• What data goes on which axis
• What lines/bars/segments to plot
• Any comparisons to show (before/after, segment A vs B)

1. What to Look For - Describe the patterns that would validate or invalidate the hypothesis:

• Changes in trend lines
• Differences in values between segments
• Timing of changes
• Magnitude of differences

1. Proposed Deep-Dives - Suggest follow-up analyses if hypothesis is validated

Critical: Never include SQL queries or code. Always describe analysis in plain language that any analyst can translate to their own tools.

Phase 5: Structure Output

Organize the analysis plan with:

• Executive Summary (problem, question, decision, timeline)
• Success Criteria (what done looks like)
• Context (metric definition, KPI tree, recent changes)
• Hypotheses (prioritized tiers with scores)
• Analysis Approach for each hypothesis

Do NOT include:

• Analysis Roadmap & Timeline section
• Data Requirements section with SQL
• Roles & Responsibilities section
• Technical implementation details

Analysis Types

Type 1: Root Cause Investigation

When: KPI changed, need to know why Timeline: 1-3 days Key techniques: KPI tree navigation, segmentation, timeline correlation Agents needed: 🔍 Root Cause Investigator, 👥 Segmentation Expert (to identify which customer groups drove the change)

When to use Segmentation Expert:

• KPI changed but don't know which customers drove it
• Need to understand "who" behind the "what"
• Investigating whether it's a mix effect or performance effect

Type 2: Opportunity Sizing

When: Evaluating potential initiative Timeline: 2-5 days Key techniques: Market sizing, segment analysis, conversion math Agents needed: 👥 Segmentation Expert (to identify target segments and their potential), 📊 Chart Advisor

When to use Segmentation Expert:

• Sizing opportunity by customer segment
• Understanding which segments would benefit most
• Estimating addressable market by segment

Type 3: Segment Deep Dive

When: Understanding specific customer group Timeline: 3-7 days Key techniques: Behavioral analysis, cohort tracking, feedback synthesis Agents needed: 👥 Segmentation Expert (primary), 📅 Cohort Analyst (to see how segment evolves), 💬 Feedback Synthesizer

When to use both Segmentation and Cohort Analysis:

• Understanding how different segments perform over time
• Comparing retention across customer segments
• Identifying which segments have best/worst long-term value

Type 4: Feature Performance Review

When: Evaluate feature success Timeline: 2-4 days Key techniques: Adoption funnel, user feedback, cohort comparison Agents needed: 📅 Cohort Analyst (to compare cohorts pre/post launch), 🔬 Influential Factors Detective

When to use Cohort Analysis Specialist:

• Comparing user cohorts before and after feature launch
• Tracking feature adoption over time by cohort
• Understanding if newer users adopt feature faster

Type 5: Experiment Analysis

When: Evaluate A/B test Timeline: 1-2 days Key techniques: Statistical testing, segment analysis, guardrail checks Agents needed: 🧪 A/B Testing Advisor, 👥 Segmentation Expert (to understand if effect varies by segment)

When to use Segmentation Expert:

• Checking if experiment effect consistent across segments
• Identifying which segments benefit most from change
• Understanding heterogeneous treatment effects

Type 6: Retention Analysis

When: Understanding why customers stay or leave Timeline: 3-5 days Key techniques: Cohort retention tables, segment comparison, leading indicators Agents needed: 📅 Cohort Analysis Specialist (primary), 👥 Segmentation Expert (to compare segments)

When to use Cohort Analysis Specialist:

• Measuring retention rates over time
• Comparing new vs old user behavior
• Identifying early warning signals in recent cohorts
• Calculating lifetime value by cohort
• Understanding maturation patterns

Analysis Approach Format

For each hypothesis, provide a detailed, step-by-step analysis plan using this structure:

Example: H1 - Setup Completion Dropped

Data to Collect:

• Number of sign-ups per week for the last 8 weeks
• Number of users who completed each step of setup (name entry, company entry, job entry, goal selection)
• Time spent at each setup step
• Drop-off points (which step users abandoned)
• Segmented by: acquisition channel, device type (desktop/mobile), geography

How to Analyze:

Analysis 1: Setup Funnel Trends

• Chart type: Multi-line chart
• X-axis: Week (last 8 weeks)
• Y-axis: Completion rate (%)
• Lines to plot:
• Overall setup completion rate
• Completion rate for each individual step (name, company, job, goal)
• Comparison: Overlay a reference line showing the average from 2 months ago

Analysis 2: Step-by-Step Micro-Funnel

• Chart type: Table with color-coded cells
• Rows: Each week
• Columns: Each setup step with completion percentage
• Color coding: Green for improvements, red for declines >5pp

Analysis 3: Segment Comparison

• Chart type: Grouped bar chart
• X-axis: Segments (acquisition channels, device types)
• Y-axis: Setup completion rate (%)
• Bars: Two bars per segment (previous month average vs last month)
• Highlight: Segments with largest drops

What to Look For:

To VALIDATE this hypothesis:

• Setup completion rate decreased by 5+ percentage points in last month
• The decrease is visible in one specific step (not distributed equally)
• The timing of the decrease correlates with a product change
• Multiple segments show the decrease (not just one channel or device)

To INVALIDATE this hypothesis:

• Setup completion rate remained stable or increased
• The decrease is <2 percentage points (within normal variance)
• Only one specific segment affected (suggests mix effect, not product issue)
• The decline started before any product changes

Proposed Deep-Dives:

If hypothesis is validated:

1. Specific Step Analysis: Deep-dive into the exact step showing drop-off

• Time distribution at that step (did users spend more time?)
• Error logs or failed attempts at that step
• Session recordings of users abandoning at that step

1. A/B Test Analysis: If there was a recent change

• Compare users who saw old version vs new version
• Measure the exact impact of the change

1. Device-Specific Investigation: If mobile shows bigger drop

• Mobile UI issues or bugs
• Screen size rendering problems
• Touch interaction issues

Type 6: Retention Analysis

When: Understanding why customers stay or leave Timeline: 3-5 days Key techniques: Cohort retention tables, segment comparison, leading indicators Agents needed: 📅 Cohort Analysis Specialist (primary), 👥 Segmentation Expert (to compare segments)

When to use Cohort Analysis Specialist:

• Measuring retention rates over time
• Comparing new vs old user behavior
• Identifying early warning signals in recent cohorts
• Calculating lifetime value by cohort
• Understanding maturation patterns

Combining Segmentation & Cohort Analysis

Many of the most powerful analyses use BOTH segmentation and cohort analysis together:

Pattern 1: Segment-Specific Cohort Analysis

Question: "Do enterprise customers retain better than SMBs over time?"

Approach:

1. Use 👥 Segmentation Expert to define segment rules (Enterprise: >500 employees, SMB: <500)
2. Use 📅 Cohort Analysis Specialist to build retention table for each segment separately
3. Compare retention curves between segments

Data to Collect:

• Cohort (signup month) for all users
• Company size for each user
• Activity data for retention calculation

How to Analyze:

• Create two retention tables: one for Enterprise, one for SMB
• Plot retention curves on same chart (different colored lines)
• Compare M1, M3, M6 retention rates between segments

What to Look For:

• Does one segment consistently retain better?
• Do gaps widen or narrow over time?
• Are recent cohorts improving in both segments?

Pattern 2: Cohort-Specific Segmentation

Question: "Has the mix of customer segments changed over time?"

Approach:

1. Use 📅 Cohort Analysis Specialist to define cohorts (monthly)
2. Use 👥 Segmentation Expert to analyze segment distribution within each cohort
3. Track how segment mix evolves across cohorts

Data to Collect:

• Cohort for all users
• Segment classification for all users
• Size of each segment within each cohort

How to Analyze:

• Create stacked bar chart with cohorts on X-axis
• Stack shows % of each segment within cohort
• Track how segment composition changes

What to Look For:

• Is one segment growing as % of new signups?
• Does segment mix correlate with overall metric changes?
• Are certain channels bringing different segment mixes?

Pattern 3: Explaining Overall KPI Changes

Question: "Our retention dropped 10pp - is it because of segment mix or performance?"

Approach:

1. Use 📅 Cohort Analysis Specialist to confirm retention drop in recent cohorts
2. Use 👥 Segmentation Expert to check if segment mix changed
3. Use 🎯 Mix Effect Analyzer to quantify mix vs performance contribution

Analysis Steps:

1. Build retention table to confirm drop
2. Check segment distribution: Did low-retention segments grow?
3. Check within-segment retention: Did retention drop within segments too?
4. Calculate: Mix effect + Performance effect = Total effect

Possible Outcomes:

• Pure mix effect: Segment performance stable, but more low-retention segments
• Pure performance effect: Segment mix stable, but all segments performing worse
• Combined effect: Both mix shifted AND performance declined

Pattern 4: Predictive Segmentation from Cohort Behavior

Question: "Which early behaviors predict long-term retention?"

Approach:

1. Use 📅 Cohort Analysis Specialist to identify mature cohorts with good retention
2. Look back at their Week 1 behavior patterns
3. Use 👥 Segmentation Expert to create behavioral segments (e.g., "High Engagers")
4. Apply to new cohorts for early prediction

Data to Collect:

• Mature cohorts (6+ months old) with retention outcomes
• Week 1 behavior for all users in those cohorts
• Same behavior data for recent cohorts

How to Analyze:

• For retained users in mature cohorts, identify common Week 1 behaviors
• Create behavioral segments based on these patterns
• Track % of new cohorts falling into "high retention" behavioral segments
• Predict their likely long-term retention

What to Look For:

• Strong correlations between early behavior and retention (e.g., 3+ sessions in Week 1 → 75% M6 retention)
• Behavioral segments that are predictive and actionable
• Early warning when new cohorts show worse early behavior patterns

Common Mistakes & Fixes

Mistake 1: No Hypotheses Upfront

Problem: Jump straight to data, get lost Fix: List 3-5 hypotheses before touching data. Use KPI Tree + recent events.

Mistake 2: Trying to Answer Everything

Problem: Analysis paralysis, never finishes Fix: Prioritize top 3 hypotheses. Focus on 80/20. Set time limits per phase.

Mistake 3: No Decision Criteria Upfront

Problem: Analysis complete but "Is this good enough?" Fix: Define minimum improvement threshold upfront. Define decision criteria. Identify decision maker.

Mistake 4: Not Involving Right People

Problem: Wasted effort, no trust Fix: Involve engineer (validate data), PM (business context), stakeholders (alignment) early.

Mistake 5: Over-Scoping Timeline

Problem: Silence for weeks, stakeholder frustration Fix: Break into phases with checkpoints. Share interim findings. Set realistic timelines (2x initial estimate).

Time-Boxing Strategies

The 80/20 Rule

Total 10 hours budget:

• Phase 1 (2h): Quick exploration → 80% of answer
• Phase 2 (4h): Deep dive → 95% of answer
• Phase 3 (2h): Validation → 100% confidence
• Phase 4 (2h): Presentation prep

Progressive Elaboration

• Hour 1: Look at everything (dashboard scan)
• Hour 2: Focus on what's interesting
• Hour 3: Go deep on root cause
• Hour 4: Document findings

Check-in Cadence

Daily: 15-min standup (learned, analyzing today, blockers) Every 2-3 days: Stakeholder checkpoint (interim findings, feedback, adjust)

Quality Checklist

Before starting:

• [ ] Clear question defined
• [ ] Success criteria established
• [ ] Hypotheses prioritized (top 3-5)
• [ ] Data sources identified and accessible
• [ ] Timeline estimated and communicated
• [ ] Stakeholders aligned on plan
• [ ] Resources allocated

During analysis:

• [ ] Checking hypotheses in priority order
• [ ] Documenting findings as you go
• [ ] Checkpointing with stakeholders
• [ ] Staying on timeline (or updating it)
• [ ] Avoiding scope creep

Before sharing results:

• [ ] Question fully answered
• [ ] Recommendations clear and actionable
• [ ] Supporting data is solid
• [ ] Alternative explanations considered
• [ ] "So What?" is obvious
• [ ] Ready for tough questions

Related Agents & Skills

Use before starting:

• 🌳 KPI Tree Architect - Framework for hypotheses
• 📍 North Star Metric Advisor - Ensure analysis ties to strategy

Use during:

• 🔍 Root Cause Investigator - Execute investigation
• 👥 Segmentation Expert - Analyze performance differences across customer segments, identify which segments drove changes
• 📅 Cohort Analysis Specialist - Track behavior over time, compare user vintages, measure retention
• 🎯 Mix Effect Analyzer - Separate mix vs performance
• 🔬 Influential Factors Detective - Qualitative factors
• 💬 Customer Feedback Synthesizer - Customer voice

Use after:

• 📊 Chart & Visualization Advisor - Create compelling visuals
• 📝 Executive Summary Writer - Document findings
• 🎬 Presentation Architect - Share results

Key Principles

From "The Power of Analytics" book:

• "Running an analysis by jumping into the data is not the right approach and is doomed for failure"
• "Take the time to list down your hypotheses before collecting data"
• "Use your KPI tree to zero in on the most probable area where to dig"
• "Come up with a list of 3 to 5 main hypotheses about the drivers worth exploring"

Success Metrics

Planning is effective when:

• ✅ Analyses finish on time
• ✅ Findings are actionable
• ✅ Stakeholders don't say "but did you check..."
• ✅ No analysis paralysis
• ✅ High confidence in results
• ✅ Team can plan analyses independently
• ✅ Minimal do-overs or scope changes

Additional Resources

See references/roadmap_template.md for complete analysis roadmap template See references/analysis_templates.md for templates by analysis type