Growth Marketing skill 11 of 13

Cohort Retention Analysis SKILL.md

The Cohort Retention Analysis skill makes Claude, Codex or another AI agent build and read retention properly: users, accounts and revenue grouped by when they started and followed over time. It defines retention by value, chooses N-day, unbounded or bracket retention, adds logo retention, net revenue retention and repeat purchase, checks whether curves flatten, compares cohorts and segments and finds behaviours linked to retention.

cohort-retention-analysis/SKILL.md, version 1.0.0, 1,403 words.

By Afeez Adebayo, Marketing, SEO & Growth Expert with 10+ years of experience.

Created 10 October 2026. Updated 10 October 2026. Free to use under CC BY 4.0.

Follow me onX LinkedIn

Download .md Skill folder (.zip)All 13 in one folder (.zip)

Retention is the foundation of growth. If users stop coming back, every acquisition pound leaks away, and growth stalls however much you spend. Cohort analysis shows retention properly: group users by when they started, follow each group over time, and see whether the curve flattens (people who stay, stay) or keeps falling towards zero. It also shows whether recent changes made newer cohorts better than older ones.

This skill builds and reads cohort retention analyses for users, accounts and revenue, segments them to find who retains and why, identifies behaviours linked to retention and turns findings into retention work.

Operating principles

Use cohorts, not averages. Blended retention mixes old and new users and hides change.
Look for the flattening. A curve that levels off shows product-market fit for that segment; one that keeps falling doesn't.
Define "retained" by value. Returning to do the core action beats simply logging in.
Choose the right frame. Daily for social apps, weekly for work tools, monthly for many SaaS and e-commerce, by purchase cycle for consumables.
Revenue retention matters too. Expansion can make revenue curves rise even when user curves fall.

Types of retention

Type	Definition	Use
N-day / N-week retention	Share active exactly on day or week N	Engagement products
Unbounded (rolling) retention	Share active on day N or later	Lower-frequency products
Bracket retention	Share active in a window (for example days 7 to 13)	Smoother curves for irregular use
Logo retention (B2B)	Share of accounts still customers	Subscription health
Net revenue retention (NRR)	Revenue from a cohort including expansion, minus churn and contraction	SaaS growth efficiency
Repeat purchase rate	Share buying again within a window	E-commerce

When to use, and when not to

Use it in growth audits, before scaling acquisition, after onboarding or product changes, when comparing channels or segments, and to find the activation moment.

Don't use it for short-term conversion steps (funnel-analysis) or to run retention campaigns (the CRM and Lifecycle retention-churn-prevention skill).

Inputs to ask for, in one message

Event or order data with user or account IDs and timestamps.
The core action that represents value.
Segments: acquisition channel, plan, country, device, use case, company size.
Revenue data for revenue retention.
Dates of major product or pricing changes.

Raw SKILL.md file

.md

This is the exact file agents read, including the credit header and licence notice. Save it as cohort-retention-analysis/SKILL.md.

---
name: cohort-retention-analysis
description: "Build and read cohort retention analyses for users, accounts and revenue: value-based activity definitions, N-day, unbounded and bracket retention, logo retention, net revenue retention and repeat purchase rate, cohort tables and curves, flattening, cohort comparisons over time, segments by channel and plan, retention-linked behaviours and actions. Use whenever someone asks about retention analysis, cohort analysis, retention curves, churn by cohort, net revenue retention, repeat purchase rate, product-market fit signals or why users don't come back."
license: CC-BY-4.0
metadata:
  author: Afeez Adebayo
  author_url: https://afeezadebayo.com
  author_linkedin: https://www.linkedin.com/in/afeez-adebayo/
  author_x: https://x.com/za_icon
  source: https://afeezadebayo.com/skills-md/marketing/growth-marketing/cohort-retention-analysis/
  category: marketing
  subcategory: growth-marketing
  version: "1.0.0"
  created: "2026-10-10"
  updated: "2026-10-10"
---

# Cohort Retention Analysis

Retention is the foundation of growth. If users stop coming back, every acquisition pound leaks away, and growth stalls however much you spend. Cohort analysis shows retention properly: group users by when they started, follow each group over time, and see whether the curve flattens (people who stay, stay) or keeps falling towards zero. It also shows whether recent changes made newer cohorts better than older ones.

This skill builds and reads cohort retention analyses for users, accounts and revenue, segments them to find who retains and why, identifies behaviours linked to retention and turns findings into retention work.

## Operating principles
1. **Use cohorts, not averages.** Blended retention mixes old and new users and hides change.
2. **Look for the flattening.** A curve that levels off shows product-market fit for that segment; one that keeps falling doesn't.
3. **Define "retained" by value.** Returning to do the core action beats simply logging in.
4. **Choose the right frame.** Daily for social apps, weekly for work tools, monthly for many SaaS and e-commerce, by purchase cycle for consumables.
5. **Revenue retention matters too.** Expansion can make revenue curves rise even when user curves fall.

## Types of retention
| Type | Definition | Use |
|---|---|---|
| N-day / N-week retention | Share active exactly on day or week N | Engagement products |
| Unbounded (rolling) retention | Share active on day N or later | Lower-frequency products |
| Bracket retention | Share active in a window (for example days 7 to 13) | Smoother curves for irregular use |
| Logo retention (B2B) | Share of accounts still customers | Subscription health |
| Net revenue retention (NRR) | Revenue from a cohort including expansion, minus churn and contraction | SaaS growth efficiency |
| Repeat purchase rate | Share buying again within a window | E-commerce |

## When to use, and when not to
Use it in growth audits, before scaling acquisition, after onboarding or product changes, when comparing channels or segments, and to find the activation moment.

Don't use it for short-term conversion steps (`funnel-analysis`) or to run retention campaigns (the CRM and Lifecycle `retention-churn-prevention` skill).

## Inputs to ask for, in one message
1. Event or order data with user or account IDs and timestamps.
2. The core action that represents value.
3. Segments: acquisition channel, plan, country, device, use case, company size.
4. Revenue data for revenue retention.
5. Dates of major product or pricing changes.

## Workflow

### Step 1. Define the cohort and activity
Cohort by sign-up or first-purchase week or month; activity as the core value action.

### Step 2. Build the cohort table
Rows are cohorts, columns are periods since start, cells are share retained.

### Step 3. Plot curves
One line per cohort or segment; look for flattening and its level.

### Step 4. Compare cohorts over time
Are newer cohorts retaining better after changes? Annotate change dates.

### Step 5. Segment
By channel, plan, use case and company size. Find segments with flat, high curves (your best customers) and those that never flatten.

### Step 6. Find retention-linked behaviours
Compare early behaviours of retained and churned users; candidate activation moments (validate with experiments).

### Step 7. Add revenue retention
NRR or revenue cohort curves for subscription and B2B businesses.

### Step 8. Recommend actions
Onboarding to the value moment, focusing acquisition on high-retention segments, product fixes for drop-off points, and lifecycle interventions.

## Reading curves
| Pattern | Meaning | Action |
|---|---|---|
| Falls then flattens at a healthy level | Core users found value | Scale acquisition of similar users |
| Keeps falling towards zero | No lasting value for this group | Fix product or narrow the target segment |
| Newer cohorts flatten higher | Changes are working | Keep going; measure which change |
| One channel much lower | Channel brings poor-fit users | Change targeting or reduce spend |
| Revenue curve rises above 100% | Expansion outweighs churn | Invest in expansion levers |

## Cohort table example
| Sign-up week | Users | Week 1 | Week 2 | Week 4 | Week 8 | Week 12 |
|---|---|---|---|---|---|---|
| 2026-06-01 | 2,100 | 48% | 36% | 27% | 22% | 21% |
| 2026-06-08 | 2,300 | 51% | 39% | 30% | 25% | 24% |
| 2026-07-06 (new onboarding) | 2,250 | 58% | 46% | 36% | 31% | |
Illustrative; note the flattening after week 8 and the higher curve after the onboarding change.

## Cohort query example (SQL)
```sql
WITH first_seen AS (
  SELECT user_id, DATE_TRUNC('week', MIN(ts)) AS cohort_week
  FROM events WHERE event = 'signed_up' GROUP BY user_id
), activity AS (
  SELECT DISTINCT user_id, DATE_TRUNC('week', ts) AS active_week
  FROM events WHERE event = 'completed_lesson'
)
SELECT f.cohort_week,
       DATE_DIFF(a.active_week, f.cohort_week, WEEK) AS week_n,
       COUNT(DISTINCT a.user_id) AS active_users
FROM first_seen f JOIN activity a USING (user_id)
GROUP BY 1, 2 ORDER BY 1, 2;
```
Divide by cohort size to get retention rates; adapt syntax to your warehouse.

## Retention benchmarks: use with care
Published retention benchmarks vary widely by category, platform and definition. Compare your own cohorts over time and across segments first; use outside benchmarks only as rough context, with their definitions checked.

## Finding retention-linked behaviours
1. Take users from cohorts at least 90 days old.
2. Split into retained and churned by your definition.
3. Compare their first 7 to 14 days: actions taken, features used, invites sent, settings completed.
4. Rank behaviours by the retention difference and by how many users do them.
5. Pick candidates the team can influence and test them with experiments.

## Revenue retention view
| Cohort | Month 0 revenue | Month 6 | Month 12 | NRR at 12 months |
|---|---|---|---|---|
A cohort whose revenue rises above its starting point shows expansion outweighing churn, a strong sign for subscription businesses.

## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Irregular usage | Use bracket or unbounded retention |
| Seasonal products | Compare like-for-like cohorts year on year |
| B2B with seats | Track account retention and seat expansion separately |
| Small cohorts | Group into months or quarters; show confidence |
| Free and paid users mixed | Analyse separately and together |

## Handoffs to related skills
- Growth model inputs: `growth-model`
- North star validation: `north-star-metric-tree`
- Funnel before retention: `funnel-analysis`
- Event definitions: `tracking-plan`
- Retention programmes: the CRM and Lifecycle skills

## Worked example
A language learning app saw monthly active users growing but revenue flat. Weekly cohort curves for the core action (completing a lesson) never flattened for users from a paid social campaign, while organic and referral cohorts flattened around a much higher level. Users who completed three lessons in their first week retained far better. The team shifted paid social targeting, rebuilt the first week around three short lessons and a streak, and newer cohorts began to flatten at a higher level within two months.

## Output format
```
# Cohort retention analysis: <product>, <date>
## Definitions (cohort, activity, frame)
## Cohort table and curves
## Changes over time (annotated)
## Segment comparison
## Retention-linked behaviours (labelled correlation)
## Revenue retention
## Recommendations
```

## Common mistakes
- Blended retention instead of cohorts.
- Defining retention as any login.
- Wrong time frame for the product.
- Ignoring revenue retention and expansion.
- Treating correlated behaviours as proven causes.

## Questions to ask before you analyse
- What action shows a user got value?
- How often should a healthy user do it?
- What changed in the product or acquisition recently?

## Self-check before you answer
- Are cohorts and activity defined by value?
- Did I check for flattening and compare cohorts over time?
- Did I segment by channel and plan?
- Are behaviours labelled as correlation with tests planned?

## Further reading
- Amplitude, retention analysis guide: https://amplitude.com/blog
- Mixpanel, retention reports: https://docs.mixpanel.com/

---
Created by Afeez Adebayo, Marketing, SEO & Growth Expert
Website: https://afeezadebayo.com
LinkedIn: https://www.linkedin.com/in/afeez-adebayo/
X: https://x.com/za_icon
Source: https://afeezadebayo.com/skills-md/marketing/growth-marketing/cohort-retention-analysis/

Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.

Workflow

Step 1. Define the cohort and activity

Cohort by sign-up or first-purchase week or month; activity as the core value action.

Step 2. Build the cohort table

Rows are cohorts, columns are periods since start, cells are share retained.

Step 3. Plot curves

One line per cohort or segment; look for flattening and its level.

Step 4. Compare cohorts over time

Are newer cohorts retaining better after changes? Annotate change dates.

Step 5. Segment

By channel, plan, use case and company size. Find segments with flat, high curves (your best customers) and those that never flatten.

Step 6. Find retention-linked behaviours

Compare early behaviours of retained and churned users; candidate activation moments (validate with experiments).

Step 7. Add revenue retention

NRR or revenue cohort curves for subscription and B2B businesses.

Onboarding to the value moment, focusing acquisition on high-retention segments, product fixes for drop-off points, and lifecycle interventions.

Reading curves

Pattern	Meaning	Action
Falls then flattens at a healthy level	Core users found value	Scale acquisition of similar users
Keeps falling towards zero	No lasting value for this group	Fix product or narrow the target segment
Newer cohorts flatten higher	Changes are working	Keep going; measure which change
One channel much lower	Channel brings poor-fit users	Change targeting or reduce spend
Revenue curve rises above 100%	Expansion outweighs churn	Invest in expansion levers

Cohort table example

Sign-up week	Users	Week 1	Week 2	Week 4	Week 8	Week 12
2026-06-01	2,100	48%	36%	27%	22%	21%
2026-06-08	2,300	51%	39%	30%	25%	24%
2026-07-06 (new onboarding)	2,250	58%	46%	36%	31%

Illustrative; note the flattening after week 8 and the higher curve after the onboarding change.

Cohort query example (SQL)

WITH first_seen AS (
  SELECT user_id, DATE_TRUNC('week', MIN(ts)) AS cohort_week
  FROM events WHERE event = 'signed_up' GROUP BY user_id
), activity AS (
  SELECT DISTINCT user_id, DATE_TRUNC('week', ts) AS active_week
  FROM events WHERE event = 'completed_lesson'
)
SELECT f.cohort_week,
       DATE_DIFF(a.active_week, f.cohort_week, WEEK) AS week_n,
       COUNT(DISTINCT a.user_id) AS active_users
FROM first_seen f JOIN activity a USING (user_id)
GROUP BY 1, 2 ORDER BY 1, 2;

Divide by cohort size to get retention rates; adapt syntax to your warehouse.

Retention benchmarks: use with care

Published retention benchmarks vary widely by category, platform and definition. Compare your own cohorts over time and across segments first; use outside benchmarks only as rough context, with their definitions checked.

Finding retention-linked behaviours

Take users from cohorts at least 90 days old.
Split into retained and churned by your definition.
Compare their first 7 to 14 days: actions taken, features used, invites sent, settings completed.
Rank behaviours by the retention difference and by how many users do them.
Pick candidates the team can influence and test them with experiments.

Revenue retention view

Cohort	Month 0 revenue	Month 6	Month 12	NRR at 12 months

A cohort whose revenue rises above its starting point shows expansion outweighing churn, a strong sign for subscription businesses.

Edge cases and judgement calls

Situation	What to do
Irregular usage	Use bracket or unbounded retention
Seasonal products	Compare like-for-like cohorts year on year
B2B with seats	Track account retention and seat expansion separately
Small cohorts	Group into months or quarters; show confidence
Free and paid users mixed	Analyse separately and together

Growth model inputs: growth-model
North star validation: north-star-metric-tree
Funnel before retention: funnel-analysis
Event definitions: tracking-plan
Retention programmes: the CRM and Lifecycle skills

Worked example

A language learning app saw monthly active users growing but revenue flat. Weekly cohort curves for the core action (completing a lesson) never flattened for users from a paid social campaign, while organic and referral cohorts flattened around a much higher level. Users who completed three lessons in their first week retained far better. The team shifted paid social targeting, rebuilt the first week around three short lessons and a streak, and newer cohorts began to flatten at a higher level within two months.

Output format

# Cohort retention analysis: <product>, <date>
## Definitions (cohort, activity, frame)
## Cohort table and curves
## Changes over time (annotated)
## Segment comparison
## Retention-linked behaviours (labelled correlation)
## Revenue retention
## Recommendations

Common mistakes

Blended retention instead of cohorts.
Defining retention as any login.
Wrong time frame for the product.
Ignoring revenue retention and expansion.
Treating correlated behaviours as proven causes.

Questions to ask before you analyse

What action shows a user got value?
How often should a healthy user do it?
What changed in the product or acquisition recently?

Self-check before you answer

Are cohorts and activity defined by value?
Did I check for flattening and compare cohorts over time?
Did I segment by channel and plan?
Are behaviours labelled as correlation with tests planned?

When to use this skill, and when not to

Use it in growth audits, before scaling acquisition, after onboarding or product changes, when comparing channels or segments and when looking for the activation moment.

Retention is the foundation of growth: if users stop coming back, every acquisition pound leaks away. Cohort curves show whether the people who stay keep staying, a strong product-market fit signal, and whether recent changes made newer cohorts better than older ones.

Don’t use it for short-term conversion (use Funnel Analysis) or to run retention campaigns (use the CRM and Lifecycle Retention and Churn Prevention skill).

Example prompts

“Build cohort curves and tell me if we have product-market fit.”
“Which acquisition channel brings users who stay?”
“Did our new onboarding improve retention?”
“Calculate net revenue retention by cohort.”
“Write the SQL for weekly cohort retention.”

Example output

The skill returns a fixed structure. A shortened example for an imaginary site:

# Cohort retention analysis: example language app, 10 October 2026

## Definition
Retained = completed a lesson in the week; weekly cohorts by sign-up

## Findings
| Segment | Week-8 retention | Curve |
| Organic | 31% | Flattens from week 6 |
| Referral | 34% | Flattens from week 6 |
| Paid social campaign X | 6% | Never flattens |

## Retention-linked behaviour (correlation)
3 lessons in week 1 -> much higher week-8 retention

Behaviours linked to retention are labelled as correlation and paired with an experiment, so the team doesn’t mistake a pattern for a proven cause.

What the agent needs from you

The skill asks for event or order data with user or account IDs and timestamps, the core action that represents value, segments such as channel, plan, country, device and company size, revenue data for revenue retention and the dates of major product or pricing changes.

How this skill hands off to related skills

Retention connects strategy and execution. Model inputs go to Growth Model, north star validation to North Star Metric, the funnel before retention to Funnel Analysis, event definitions to Tracking Plan and retention programmes to the CRM and Lifecycle skills.

How to run Cohort Retention Analysis step by step

Install the skill as cohort-retention-analysis/SKILL.md.
Name the core value action and the right time frame.
Export event or order data with IDs and timestamps.
Add segments and change dates.
Review curves, cohort comparisons and segments.
Act on findings and plan experiments for retention-linked behaviours.

Situations this skill handles

Retention looks different by product. The skill handles these situations:

Situation	What the skill does
Irregular usage	Bracket or unbounded retention
Seasonal products	Year-on-year cohort comparisons
B2B with seats	Account retention and seat expansion separately
Small cohorts	Monthly or quarterly grouping with confidence shown
Free and paid users mixed	Analysed separately and together
Channel brings poor-fit users	Retention by channel guides targeting

What makes this skill different

Most retention reporting is a single blended number. This skill builds cohorts, defines retention by value, checks for flattening, compares cohorts before and after changes and segments by channel and plan to show who your best customers are.

It adds revenue retention for subscription businesses, because expansion can make revenue curves rise even while user curves fall.

Practical tips from real projects

Cohorts, not averages.
Retention by value action.
Match the frame to usage frequency.
Correlation is a hypothesis, not proof.

A worked example: a language learning app saw monthly active users rising but revenue flat. Weekly cohorts for lesson completion never flattened for users from one paid social campaign, while organic and referral cohorts flattened at a much higher level. Users who completed three lessons in week one retained far better. Changing paid social targeting and rebuilding the first week around three short lessons and a streak made newer cohorts flatten higher within two months.

Reading curves follows clear patterns. A curve that falls then flattens at a healthy level shows core users found value, so scale acquisition of similar users. A curve that keeps falling towards zero shows no lasting value for that group, so fix the product or narrow the segment. Newer cohorts flattening higher means changes are working. One channel far lower means it brings poor-fit users. A revenue curve above 100% means expansion outweighs churn.

Cohort tables make change visible: rows are sign-up weeks or months, columns are periods since start and cells are the share retained. Marking the date of an onboarding or pricing change shows at a glance whether later rows sit higher.

The skill includes SQL to build weekly cohort tables from event data and finds retention-linked behaviours by comparing what retained and churned users did in their first days, which often points straight to the activation moment.

Retention types answer different questions. N-day or N-week retention shows the share active exactly on a given day or week, suited to engagement products. Unbounded retention shows the share active on that day or later, suited to lower-frequency products. Bracket retention uses windows such as days 7 to 13, smoothing curves for irregular use. Logo retention tracks the share of B2B accounts still customers. Net revenue retention follows a cohort’s revenue including expansion. Repeat purchase rate shows the share buying again within a window for e-commerce. Choosing the wrong type can make a healthy product look weak or hide a real problem.

The time frame matters as much as the type: daily for social apps, weekly for work tools, monthly for many SaaS and e-commerce businesses and by purchase cycle for consumables. A furniture retailer measured weekly would look broken; a messaging app measured monthly would hide early churn.

Revenue cohorts complete the picture for subscription businesses, showing whether upgrades and seat growth outweigh churn and contraction over time.

Changelog

1.0.0, 10 October 2026: first release.

Use, adapt and share this skill under CC BY 4.0: credit Afeez Adebayo and link to this page.

Previous skillFunnel Analysis Next skillTracking Plan and Event Taxonomy

All Growth Marketing skills