What is Customer Cohort Analysis?
Also known as: Cohort analysis, User cohort analysis
What is customer cohort analysis?
Customer cohort analysis is a method that groups users by a shared starting event, then follows each group forward in time to compare how they behave. The shared event is usually signup, install, or first purchase. The behavior tracked is usually retention, revenue, or frequency.
A blended retention number lies. Fast new-user growth makes a leaky product look healthy. Cohorts strip the lie out, the same way a clean churn rate does. They show whether the January cohort still uses the product in March, regardless of how many users joined in February.
It is the cleanest way to answer one question. Is the product getting stickier or weaker over time?
How a cohort table is built
A cohort table has three axes. Acquisition cohorts on rows. Time periods on columns. A single metric in each cell.
Rows hold the groups. Each row is one cohort, like "users who signed up in week 14 of 2026." Columns hold the elapsed periods since the cohort started. Week 0, week 1, week 2, and so on. Each cell holds the metric for that cohort at that elapsed period.
The first column always shows 100 percent (or the full starting count). Every column to the right shows the share of that cohort still active, still paying, or still purchasing. Read across a row to see one cohort decay. Read down a column to compare the same elapsed period across cohorts.
[UNIQUE INSIGHT] Most teams stop at one metric per table. Stack two. Build a retention table and a revenue-per-user table side by side. When retention falls but revenue per surviving user climbs, you are not losing a product fight. You are filtering toward power users.
Common cohort metrics
Four metrics dominate cohort tables. Pick the one that matches the question.
| Metric | What the cell holds | Best question it answers |
|---|---|---|
| Retention | % of cohort active in period N | Is the product sticky? |
| Revenue per cohort | Total $ from cohort by period N | Is the cohort paying back acquisition cost? |
| Purchase frequency | Avg orders per surviving user | Are repeat habits forming? |
| Cumulative LTV | Running gross margin per cohort | When do cohorts cross CAC? |
Retention is the default starting point. Revenue cohorts are the truth-teller for ecommerce and subscription. Frequency cohorts catch habit decay before retention does. Cumulative LTV cohorts are how finance models payback.
Where cohort analysis lives
Cohort analysis runs in five common places. Each one trades flexibility for setup speed.
GA4
GA4 has a built-in Cohort Exploration template under the Explore tab. It is free, fast to set up, and pulls data straight from web and app streams. Per Google's GA4 cohort documentation, you can pivot cohorts by acquisition date, audience, or first-touch source. The trade-off is rigidity. Custom cohorts beyond the template require BigQuery export.
Mixpanel and Amplitude
Both tools were built around cohorts. Behavioral cohorts, retention curves, and stickiness reports are first-class features. Per Mixpanel's cohort analysis guide, product teams use them to compare onboarding flows, feature releases, and price tests. Per Amplitude's retention playbook, cohorts feed directly into Lifecycle and Compass reports.
Looker, Tableau, and BI tools
When the data lives in a warehouse, cohort tables get built in SQL and rendered in a BI tool. More work upfront. Total flexibility on cohort definition.
Custom SQL
The escape hatch. A GROUP BY on signup week joined to an event table on user ID. Slow to build, but every cohort question becomes answerable.
Reading a cohort heatmap
A cohort heatmap encodes the table as color. Higher retention shows darker (or warmer) cells. The eye scans patterns faster than it scans numbers.
Three patterns to look for:
- Vertical bands. A column gets darker over time. Newer cohorts retain better at the same elapsed period than older ones. Product is improving.
- Horizontal cliffs. A row drops sharply at week 1 or week 4. Onboarding or first-value friction. Fix it and every future cohort benefits.
- The smile curve. Retention drops, flattens, then ticks back up. Power users form a stable core. Common in social and creator tools.
[PERSONAL EXPERIENCE] Run the heatmap weekly. Print it. Stick it on a wall. The product team starts arguing about the right things, like which cohort change correlates with which release, instead of arguing about blended numbers that hide everything.
Real-world example with numbers
A subscription DTC brand looks at three monthly cohorts of paid signups.
| Cohort | Size | Month 1 | Month 3 | Month 6 |
|---|---|---|---|---|
| January | 1,200 | 78% | 54% | 41% |
| February | 1,450 | 81% | 59% | 47% |
| March | 1,600 | 84% | 63% | (pending) |
[ORIGINAL DATA] Reading the table, two things jump out. Month-1 retention climbs from 78 to 84 percent across three cohorts. Month-3 retention follows the same lift. The team shipped a new onboarding flow in late January. The cohort math confirms it worked. Blended retention barely moved over the same window because new signups masked the gain.
Multiplied across a $40 ARPU and 70 percent gross margin, the 6-point Month-3 lift is worth roughly $24,000 in incremental contribution per cohort over the next year. The onboarding fix paid for the next two quarters of growth experiments.
Cohort analysis in 2026
Two shifts are reshaping cohort work this year. AI-assisted querying and privacy-driven attribution gaps.
Warehouse-native AI assistants now write the GROUP BY queries on demand. Marketers who could not write SQL twelve months ago are now slicing cohorts by acquisition channel, creative variant, and lifecycle stage on their own. The bottleneck moved from query writing to question framing.
Privacy changes pushed cohort analysis up the stack. With third-party signals weakening, first-party cohort behavior is the most reliable input to ad platforms. High-retention cohorts get exported as seed audiences for DAU lookalikes. Low-retention cohorts feed negative audiences. The cohort table stopped being a product report. It became a media-buying signal.
The teams pulling ahead in 2026 treat cohort analysis as the connective tissue between product, finance, and paid acquisition. One number, three departments, same source of truth.
Related terms
Frequently asked questions
What is the difference between cohort analysis and segmentation?
Segmentation slices users by attribute, such as country or plan. Cohort analysis slices by time of joining and follows each group forward. Segmentation answers 'who are they.' Cohort analysis answers 'are they getting stickier.' Most teams need both, but cohorts are the sharper tool for retention questions.
How many users do you need for a cohort to be meaningful?
At least 100 users per cohort for a stable retention curve. Below 50, weekly noise dominates. Per Mixpanel's cohort guide, small cohorts can still surface directional signal, but trust the curve only after the cohort has stabilized for two to three reporting periods.
What time period should a cohort cover?
Match the cohort window to the natural usage cycle. Daily cohorts for high-frequency apps like social or news. Weekly cohorts for SaaS and ecommerce. Monthly cohorts for low-frequency or B2B products. Mixing windows hides patterns. Pick one and hold it across reports.
Can you do cohort analysis in GA4?
Yes. GA4 ships a built-in Cohort Exploration template under Explore. It lets you group by acquisition date, audience, or first-touch source, and chart retention or revenue over weeks. Per Google's GA4 cohort docs, the report supports up to 60 cohort buckets at a time.
What is a behavioral cohort?
A behavioral cohort groups users by an action they took, not by signup date. Example: everyone who completed onboarding in their first session. Behavioral cohorts isolate the impact of a single feature or flow on long-term retention, which acquisition cohorts alone cannot answer.