Cohort Analysis

A client needs to understand retention patterns, compare user groups over time, diagnose churn, or forecast future behavior based on historical cohort data. Works for subscription, usage-based, e-commerce, and marketplace businesses.

---

How it works

1. You provide the product/service, available data, retention definition, time period, and key events or milestones
2. The skill designs retention curves, defines behavioral segments for cohort comparison, diagnoses churn patterns with root cause hypotheses, and recommends intervention timing
3. It returns a cohort analysis framework with interpretation guide, churn diagnosis, and cohort-based forecast Kate can use for retention strategy conversations

---

Prompt

You are building a cohort analysis framework for a Kate Makrigiannis consulting engagement. Kate uses this to help clients see past vanity metrics and understand how real groups of users behave over time. Cohort analysis reveals whether the product is actually getting better at retaining people, or just growing fast enough to hide churn. Before writing, read knowledge/voice-tone-guide.md -- use the client-facing voice.

Inputs I will provide:

• Product/Service: {{PRODUCT}} (what the product is, business model, stage)
• Available data: {{DATA}} (what data exists -- e.g., "signup dates, login activity, subscription status," "purchase history by customer," "event-level analytics in Mixpanel," or "we have limited data, mostly spreadsheets")
• Retention definition: {{RETENTION}} (what "retained" means for this business -- e.g., "logged in at least once in a 7-day window," "made a purchase in the calendar month," "active subscription," or "not sure, help me define it")
• Time period: {{TIME_PERIOD}} (range of data available and analysis window -- e.g., "12 months of data, want to look at monthly cohorts" or "6 weeks of data, daily cohorts")
• Key events/milestones: {{EVENTS}} (product changes, pricing changes, marketing campaigns, seasonal factors that might affect cohort behavior)
• Context (optional): {{CONTEXT}} (specific questions, known retention problems, growth targets, segments of interest)

Step 1: Define the cohort framework

Retention Definition

Before building any analysis, lock down precisely what "retained" means:

Parameter	Definition	Rationale
Cohort grouping	[How users are grouped -- e.g., by signup week, first purchase month, activation date]	[Why this grouping makes sense for the business]
Retention event	[The specific action that counts as "retained" -- e.g., "completed a session," "made a purchase," "logged in"]	[Why this event is the right signal]
Retention window	[Time bucket for measuring -- daily, weekly, monthly]	[Matches natural usage frequency of the product]
Measurement method	[Bounded vs. unbounded retention -- "did the event happen in week N" vs. "did the event happen on or after day N"]	[Which is more appropriate for this product]

Alternative Retention Definitions to Consider

Alternative	When to Use	Trade-off
[e.g., "Any login" vs. "Core action completed"]	[Broad engagement vs. meaningful usage]	[Broad definition flatters retention; narrow definition is more honest]
[e.g., "Weekly" vs. "Monthly" windows]	[High-frequency vs. low-frequency products]	[Wider windows smooth noise but hide early drop-off]

Step 2: Retention curve design

Standard Retention Table Template

Design the cohort retention table structure:

Cohort	Size	Period 0	Period 1	Period 2	Period 3	Period 4	Period 5	Period 6
[Month/Week 1]	[N users]	100%	[X%]	[X%]	[X%]	[X%]	[X%]	[X%]
[Month/Week 2]	[N users]	100%	[X%]	[X%]	[X%]	[X%]	[X%]	--
[Month/Week 3]	[N users]	100%	[X%]	[X%]	[X%]	[X%]	--	--

If the client provides actual data, populate the table. If designing the framework, explain what goes in each cell and how to compute it.

Show the math: "Period 1 retention = Users who performed [retention event] in Period 1 / Total users in cohort x 100"

Retention Curve Interpretation Guide

Curve Shape	What It Means	Typical Cause	Action
Steep early drop, then flattens	Product has a core retained audience but loses most users quickly	Onboarding friction, wrong audience, unclear value prop	Focus on activation and time-to-value
Gradual steady decline	Users slowly disengage over time, no stable floor	Product lacks habit loops or ongoing value	Build engagement hooks, recurring value triggers
Flat high retention	Most users stick around	Strong product-market fit for this segment	Focus on acquisition, the product retains well
Flat low retention	Almost everyone leaves quickly	Fundamental product or audience problem	Revisit product-market fit before optimizing retention
Improving over time (newer cohorts retain better)	Product is getting better at retaining	Product improvements, better onboarding, better targeting	Keep iterating, quantify what changed
Declining over time (newer cohorts retain worse)	Product or audience quality is degrading	Channel mix shift, market saturation, product neglect	Diagnose urgently -- this compounds fast

Step 3: Behavioral segmentation for cohort comparison

Define segments that are worth comparing as separate cohorts:

Segmentation Criteria

Segment Dimension	Segments to Compare	Hypothesis
Acquisition channel	[e.g., Organic vs. Paid vs. Referral]	[Organic users may retain better because higher intent]
Activation behavior	[e.g., Completed onboarding vs. Did not]	[Users who hit the aha moment retain at higher rates]
Plan/Tier	[e.g., Free vs. Paid vs. Enterprise]	[Paid users have sunk cost, likely higher retention]
Geography	[e.g., US vs. International]	[Product-market fit may vary by region]
Use case	[e.g., Primary use case A vs. B]	[One use case may have stronger retention loops]
Time-based	[e.g., Pre-launch vs. Post-launch of feature X]	[Feature X was supposed to improve retention -- did it?]

Priority segments to analyze first:

1. [Segment comparison] -- because [reason this is the highest-value comparison]
2. [Segment comparison] -- because [reason]
3. [Segment comparison] -- because [reason]

Step 4: Churn pattern identification

Churn Timing Analysis

Churn Window	% of Total Churn	Cumulative	Pattern
Period 0-1	[X%]	[X%]	[Early churn -- activation problem]
Period 1-3	[X%]	[X%]	[Short-term churn -- value realization gap]
Period 3-6	[X%]	[X%]	[Medium-term churn -- engagement decay]
Period 6-12	[X%]	[X%]	[Long-term churn -- competitive displacement or needs change]
Period 12+	[X%]	[X%]	[Mature churn -- natural lifecycle]

If actual data is provided, compute these. If designing the framework, explain how to calculate each and what to look for.

Root Cause Hypothesis Matrix

For each significant churn window, generate testable hypotheses:

Churn Window	Hypothesis	Supporting Signal	How to Validate	Confidence
Period 0-1	[e.g., Users do not understand value in first session]	[e.g., <30% complete onboarding]	[Onboarding completion funnel analysis]	[High / Medium / Low]
Period 0-1	[e.g., Wrong audience from paid acquisition]	[e.g., Paid cohorts churn 2x vs. organic]	[Compare cohorts by acquisition channel]	[Confidence]
Period 1-3	[e.g., No habit loop after initial use]	[e.g., Usage drops 80% after week 1]	[Session frequency analysis by cohort week]	[Confidence]
Period 3-6	[e.g., Users hit a capability ceiling]	[e.g., Power users upgrade, others leave]	[Feature usage correlation with retention]	[Confidence]

Step 4b: Statistical rigor for cohort comparisons

When comparing retention between segments or testing whether a cohort difference is real vs. noise:

Statistical Validation

Comparison	Method	When to use
Two cohort retention rates at a single time point	Z-test for proportions or chi-square	"Is the January cohort's Month 3 retention different from February's?"
Full retention curves between two groups	Log-rank test	"Does the entire retention trajectory differ between organic and paid users?"
Retention with covariates	Cox proportional hazards regression	"After controlling for plan type and geography, does acquisition channel affect retention?"
Time-to-event (time to churn)	Kaplan-Meier estimator	"What's the median time to churn, accounting for users who haven't churned yet?"

Censoring: the most common cohort analysis mistake. Users who signed up recently haven't had the opportunity to churn at later periods. This isn't "100% retention at Month 6" -- it's missing data. Kaplan-Meier curves handle this correctly by adjusting the denominator as users are "censored" (their observation window hasn't reached that period yet). Standard retention tables handle this by only showing cells where the cohort has had enough time.

Confidence intervals on retention rates: Report retention rates with confidence intervals, especially for small cohorts. A cohort of 50 users showing 60% Month 1 retention has a 95% CI of roughly [45%, 74%] -- that's a wide range. A cohort of 5,000 at 60% has a CI of [58.6%, 61.4%]. The sample size determines whether the difference you see is signal or noise.

Related skills: For choosing the right statistical test, use /statistical-test-selector. For understanding whether a retention intervention caused the improvement, use /causal-inference-guide.

Step 5: Intervention timing recommendations

Intervention Map

Churn Risk Window	Intervention	Trigger	Channel	Expected Impact
Day 0-3	[Onboarding email sequence]	[Signup without completing core action]	[Email + in-app]	[Increase activation by X%]
Day 7-14	[Re-engagement nudge]	[No activity for 5+ days]	[Push / Email]	[Recover X% of dormant users]
Day 30	[Value check-in]	[End of first month]	[Email / in-app survey]	[Identify at-risk users early]
Day 60-90	[Feature education]	[Users not using key features]	[In-app walkthrough]	[Expand usage depth]
Day 180+	[Win-back campaign]	[Churned for 30+ days]	[Email with incentive]	[Recover X% at lower LTV]

Step 6: Cohort-based forecasting

Retention Forecast Model

If the client wants to project future revenue or user counts:

Input	Value	Source
Monthly new users	[X]	[Current acquisition rate or target]
Retention curve (mature cohort)	[X% at Month 1, X% at Month 3, X% at Month 6, X% at Month 12]	[Historical cohort data or benchmark]
Revenue per retained user	[$X/month]	[ARPU or subscription price]

Forward Projection

Month	New Users	Retained from Prior Cohorts	Total Active Users	Monthly Revenue
Month 1	[X]	[0]	[X]	[$X]
Month 2	[X]	[X from M1 x M1 retention %]	[X]	[$X]
Month 3	[X]	[Sum of retained from all prior cohorts]	[X]	[$X]
...	...	...	...	...
Month 12	[X]	[X]	[X]	[$X]

Show the math for at least Month 3: "Month 3 active = New M3 users + (M1 cohort x M3 retention %) + (M2 cohort x M2 retention %) = X + X + X = X users"

Scenario Comparison

Scenario	Retention Change	Impact on Month 12 Active Users	Revenue Impact
Baseline	Current retention curve	[X users]	[$X/month]
+5% retention improvement	[Adjusted curve]	[X users (+Y%)]	[$X/month (+$Z)]
+10% retention improvement	[Adjusted curve]	[X users (+Y%)]	[$X/month (+$Z)]

"A 5 percentage-point improvement in Month 1 retention compounds to [X] additional active users by Month 12, worth approximately $[X] in additional monthly revenue."

Kate's Talking Points

• "Your retention curve shows [shape]. This tells us [interpretation]. The biggest opportunity is [specific window and action]."
• "Newer cohorts are retaining [better/worse/the same] as older cohorts. This means [the product is improving / something is degrading / retention is stable]."
• "If we improve [specific retention window] by [X] percentage points, that compounds to [X additional users and $X revenue] over 12 months."

Related skills: Feeds into /funnel-analysis for understanding where in the funnel retention breaks down. Pairs with /growth-model-builder for growth projections using cohort retention inputs. Uses /audience-segmentation for defining behavioral segments to compare.

---