Cohort analysis groups people who share a starting point, such as the month they signed up or made a first purchase, and tracks each group over time. It shows whether newer customers stay, buy and grow better or worse than earlier ones, which a single total cannot.
This guide covers the building blocks of cohort data, the main types of cohort, which metrics to track, how to read a retention table, how to build one in a spreadsheet, and the mistakes that make cohort reports misleading. It pairs with actionable metrics and A/B testing.
Key takeaways
A cohort is a group with a shared starting event. Each cohort is measured at the same ages: month 1, month 2 and so on.
Totals can rise while every cohort gets worse, because new sign-ups hide the people leaving.
Read across a row to see one cohort age, and down a column to compare cohorts at the same age.
Check cohort size first. Small cohorts produce dramatic percentages that mean little.
Every cohort report should start from one decision you need to make.
What is cohort analysis?
A cohort is a group defined by a shared event or trait: everyone who signed up in January, every customer from paid search, every account on a given plan. Cohort analysis measures each group separately over time, so patterns hidden in an overall average become visible.
Here is why that matters. Imagine active users grew 8% last month. A cohort view might show that new sign-ups drove all of the growth while each earlier cohort came back less often. The business has a retention problem hiding behind a growth number, and the right next step is very different.
Cohorts are also central to lean startup measurement. In The Lean Startup, Eric Ries recommends cohort-based reports over cumulative totals, because each new group of customers shows whether the latest product changes are working.
Questions a cohort comparison answers
Most useful cohort reports answer one of these:
Are newer cohorts doing better?
That suggests better acquisition, onboarding or product fit.
Where does engagement drop?
A sharp fall in week two or month three points to a specific gap.
Which channels bring customers who stay?
A channel with many sign-ups can still churn quickly.
Which segments need different treatment?
Self-serve users and large accounts often behave differently.
Did a change work?
Compare cohorts before and after a new onboarding flow or pricing change.
The four building blocks
Agree these definitions before building anything:
Cohort definition
Who belongs to the group, for example customers whose first order fell in a given month.
Starting event
When the clock starts, such as first purchase or subscription start. Pick the event tied to your question. Measuring retention from a first website visit counts many people who never became users.
Time interval
Days, weeks, months or billing periods. Match the rhythm of use: weekly for an app used daily, monthly or quarterly for business-to-business sales.
Outcome metric
What you measure for each cohort at each age, such as retention, repeat purchase or revenue.
Types of cohort
The method stays the same. The grouping and the question change:
| Type | Grouped by | Answers |
|---|---|---|
| Acquisition | When or where people were acquired | Which channels and campaigns bring customers who last |
| Retention and churn | Start date, tracking who stays or leaves | When people stop using, buying or renewing |
| Behavioral | An action taken, such as finishing onboarding | Which early actions go with better outcomes later |
| Revenue | Start date, tracking money over time | Whether value grows, holds or fades after discounts end |
| Sales pipeline | Lead or opportunity creation date | Whether a new qualification process brings better deals |
Acquisition
When or where people were acquired
Answers: Which channels and campaigns bring customers who last
Retention and churn
Start date, tracking who stays or leaves
Answers: When people stop using, buying or renewing
Behavioral
An action taken, such as finishing onboarding
Answers: Which early actions go with better outcomes later
Revenue
Start date, tracking money over time
Answers: Whether value grows, holds or fades after discounts end
Sales pipeline
Lead or opportunity creation date
Answers: Whether a new qualification process brings better deals
Cohort metrics that matter
Pick one primary metric tied to your decision:
| Metric | What it measures | Watch out for |
|---|---|---|
| Retention rate | Share of the cohort still active at each age | 'Active' must match how you create value, such as a monthly login or a renewal |
| Churn rate | Share that leaves in each period | Customer churn and revenue churn can tell different stories |
| Activation rate | Share that reaches an early sign of value | A link to later retention needs checking, not assuming |
| Conversion rate | Share that moves from one stage to the next | Blended rates hide differences by source and plan |
| Revenue per cohort | Money the group produces over time | High early revenue means little if the cohort churns fast |
Retention rate
Share of the cohort still active at each age
Watch out for: 'Active' must match how you create value, such as a monthly login or a renewal
Churn rate
Share that leaves in each period
Watch out for: Customer churn and revenue churn can tell different stories
Activation rate
Share that reaches an early sign of value
Watch out for: A link to later retention needs checking, not assuming
Conversion rate
Share that moves from one stage to the next
Watch out for: Blended rates hide differences by source and plan
Revenue per cohort
Money the group produces over time
Watch out for: High early revenue means little if the cohort churns fast
Reading a retention table
A retention table puts each cohort in a row and each age in a column. The first column is the cohort size, and each later cell is the share still active. Here is an illustrative table for a subscription app, by sign-up month:
| Cohort | Month 0 | Month 1 | Month 2 | Month 3 | Month 4 |
|---|---|---|---|---|---|
| January (412) | 100% | 58% | 44% | 39% | 36% |
| February (388) | 100% | 61% | 47% | 42% | |
| March (455) | 100% | 66% | 53% | ||
| April (431) | 100% | 70% |
Reading down the month 1 column, retention rises from 58% to 70%: each new cohort keeps more users after the first month. A new onboarding checklist launched in March would be a candidate explanation, worth checking against what else changed. Reading across the January row shows the curve flattening around 36% to 39%, which suggests a core group that keeps using the app.
A reading checklist
Before drawing conclusions from a table:
- Check cohort size first: A cohort of 30 can swing 10 points on three people.
- Compare the same ages: Month 2 against month 2, never against a newer cohort's month 0.
- Look for repeated patterns: One odd cohort may be a campaign, a holiday or a tracking bug.
- Separate size from quality: A big cohort is not a good cohort if it leaves quickly.
- Ask what changed: Tie shifts to known changes in product, pricing, marketing or support.
How to build a cohort analysis
Follow the same order every time:
- 1
Start with one decision. For example: did the new onboarding improve week-four retention?
- 2
Define the cohort. Group people by the starting event, source or behavior that fits the question.
- 3
Pick the metric. One primary metric, plus secondary ones only if they clarify the decision.
- 4
Clean the data. Consistent IDs, dates and event names, with duplicates and test accounts removed.
- 5
Assign periods. Give each record a cohort period and an age: month 0, month 1, month 2.
- 6
Build the table. Cohorts in rows, ages in columns, or a line per cohort on a chart.
- 7
Read the pattern. Look for drop-offs, improvements and plateaus, and leave small cohorts aside.
- 8
Decide what to change. Turn the finding into an action and watch the next cohorts.
Cohort analysis in a spreadsheet
A spreadsheet is enough for a first analysis. Start from one row per customer or event, then add helper columns:
Customer ID
A unique identifier, so each person counts once per period.
Start date
The date of the starting event, such as first purchase.
Activity date
The date of each later purchase, login or renewal.
Cohort month
The month of the start date.
Age in months
The number of months between the start date and the activity date.
A pivot table with cohort month in rows, age in columns and a distinct count of customer IDs in the values gives the raw table. Divide each cell by the month 0 count and add color scales to make a heatmap. Once data gets large or needs weekly updates, a BI tool or a product analytics tool is easier to maintain.
Cohort and funnel analysis together
The two answer connected questions:
| Point | Funnel analysis | Cohort analysis |
|---|---|---|
| What it shows | How people move through a sequence of steps | How a group behaves over time after a start |
| Typical question | Where do people get stuck? | Do the people who get through stay? |
| Example | More users finish onboarding after a change | Those users are still active in month three |
What it shows
How people move through a sequence of steps
Cohort analysis: How a group behaves over time after a start
Typical question
Where do people get stuck?
Cohort analysis: Do the people who get through stay?
Example
More users finish onboarding after a change
Cohort analysis: Those users are still active in month three
A funnel can show that more users finish setup. Only a cohort view shows whether they are still around later. When a cohort report shows a retention problem, a funnel helps find the step where people get stuck.
Segmenting cohorts
Segment by factors that could explain differences and would change what you do:
Acquisition source
Organic search, paid ads, referrals, partners, events or outbound.
Customer profile
Company size, industry, region or use case.
Product experience
Plan, features adopted, onboarding completed, integrations connected.
Sales path
Self-serve, sales-assisted or partner-led.
Value level
First order value, contract size or seat count.
Lifecycle moment
Trial start, first purchase, renewal, upgrade or reactivation.
A segment needs to be large enough to read and close enough to an action to matter.
Cohort analysis by team
The same method supports different decisions:
Product teams
Group users by sign-up week and track activation and retention. If users who connect an integration in week one retain better, improve the prompts around that step, then test whether it causes the difference.
Marketing teams
Compare channels by the retention and revenue of their cohorts, not just by sign-ups. A channel with cheap leads and fast churn costs more than it looks.
Sales teams
Compare lead cohorts before and after a new qualification rule on win rate, deal size and time to close.
Customer success
Group accounts by when they finished onboarding and track renewal and expansion. Early milestones that predict renewal become leading indicators.
Common cohort analysis mistakes
Cohort reports mislead when definitions are loose or data is thin:
- 1
Vague definitions. 'active user' means nothing until it is defined precisely.
- 2
Definitions changed midway. Retention measured two ways cannot be compared.
- 3
Ignoring cohort size. Small groups make later percentages look extreme.
- 4
Comparing incomplete periods. The newest cohort has not reached month six yet.
- 5
Mixing customer types. Self-serve and enterprise journeys may need separate tables.
- 6
Chasing one metric. Higher conversion can bring poor-fit customers who leave.
- 7
Correlation taken as cause. A behavior linked to retention may be a signal, not the cause. Test it, for example with an A/B test.
- 8
Dashboards without decisions. A table nobody acts on is decoration.
Before you share a cohort dashboard
Check both the data and the story:
- The question: The dashboard answers a specific decision.
- The cohort rule: The grouping logic is written down.
- The start event: Meaningful and captured the same way for everyone.
- The metric: Defined in plain language next to the table.
- The intervals: The same across all cohorts.
- Data quality: Duplicates, missing dates and test records handled.
- Segments: Actionable and large enough to read.
- Incomplete periods: Labeled or left out.
- The visual: A reader can tell what rows, columns, colors and values mean.
- The action: It points to a decision, an experiment or a follow-up analysis.
Frequently asked questions
What is cohort analysis?
How do you read a cohort retention table?
What is the difference between cohort analysis and segmentation?
Can you do cohort analysis in Excel?
Choosing A/B Testing Tools?
Read the buying guide: what to look for, mistakes to avoid and questions to ask vendors.
Related guides
- A/B testing: how to plan, run and read an experimentA/B testing explained: hypotheses, primary and guardrail metrics, sample size, test types, reading results, a QA checklist and common mistakes.Read the guide
- Actionable metrics vs vanity metrics, with examplesActionable metrics explained: how they differ from vanity metrics, the three A's, examples by team, flow metrics for predictability and a checklist.Read the guide
- Customer discovery: process, questions and decision treeCustomer discovery for startups: map assumptions, interview without leading, score needs, build evidence-based personas, read buying signals and decide what next.Read the guide
- 8D problem solving: the eight disciplines, step by step8D problem solving explained: D0 to D8, is/is not, containment, occurrence and escape causes, verified corrective actions, an example and the report.Read the guide
Sources
Books and papers show their publication year. Web pages show the month we last checked them.
- 1.Eric Ries, The Lean Startup (Crown Business, 2011)
- 2.Alistair Croll and Benjamin Yoskovitz, Lean Analytics: Use Data to Build a Better Startup Faster (O'Reilly Media, 2013)
- 3.Ron Kohavi, Diane Tang and Ya Xu, Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing (Cambridge University Press, 2020)