Cohort Analysis
Average retention numbers lie. They blend last year's loyal users with this week's new signups, smooth out the cracks, and hide the fact that your product may be silently deteriorating.
Cohort analysis is the antidote: it groups users by when they joined (or what they did), then tracks each group separately over time. That separation is what reveals whether the changes you ship are actually working.
Quick Summary
- A cohort is a group of users who share a starting event in the same time window, usually the week or month they signed up.
- Cohort analysis shows retention curves per group, so you can spot whether newer users behave better or worse than older ones.
- Blended "average retention" hides the truth; cohort tables expose it.
- The goal is a curve that drops fast early, then flattens, that flat tail is the sign of product-market fit.
- Segment cohorts by acquisition channel and plan tier for the most useful insights.
What It Actually Is
A cohort (from the Latin word for a group of soldiers) is just a label for people who share a defining event at roughly the same time. In marketing and product analytics, the most common defining event is "signed up" or "made their first purchase."
You then measure one specific behavior across time periods after that event. The output is a triangle-shaped table. Each row is a cohort. Each column is a time period after they joined. Each cell shows what percentage of that cohort is still active.
Cohort analysis is not the same as segmentation. Segments are static slices of your audience (age, location, plan). Cohorts are dynamic groups tracked across time from a shared starting point. Both are useful, but cohorts answer the question "are things getting better?" while segments answer "who are these people?"
Here is what a basic retention cohort table looks like:
| Cohort | Week 0 | Week 1 | Week 4 | Week 8 | Week 12 |
|---|---|---|---|---|---|
| Jan users | 100% | 48% | 31% | 26% | 22% |
| Feb users | 100% | 51% | 34% | 28% | 24% |
| Mar users | 100% | 55% | 38% | 32% | 27% |
Reading down each column shows progress: March users are outperforming January users at every time period. That improvement is real signal. A blended "30% average retention" would have hidden it completely.
Why It Matters: The Numbers
Retention has gotten harder, not easier. According to 2025 benchmarks, the average SaaS product retains roughly 46.9% of users after one month and about 35% after eight weeks. Net Revenue Retention (NRR, which measures whether existing customers are spending more or less over time) sits at a median of 101% for SaaS companies, barely above flat.
But here is the critical split: companies with NRR above 100% grow 48% year over year. Companies below 100% grow at less than half that rate. The entire difference comes down to how well you retain and expand existing customers, and cohort analysis is the only reliable way to measure that.
Industry retention benchmarks by sector (2024-2025 data):
- Enterprise B2B SaaS: 90-95% gross retention, 110-130% NRR
- Vertical SaaS (niche industries): 5-10 percentage points higher than horizontal SaaS, because switching costs are higher
- E-commerce: 30-40% customers make a second purchase within 90 days on average
- Consumer mobile apps: Under 25% of users return after day 1; under 10% stay after day 30
None of these numbers are visible in a single blended chart. You only see them in a cohort table.
The Three Types of Cohort Analysis
Acquisition cohorts are the most common. Group users by when they joined, then track what percentage return each week or month. This tells you whether retention is improving across your user base over time.
Behavioral cohorts group users by an action they took. For example: "users who completed onboarding in under 5 minutes" versus "users who took longer than 30 minutes." Comparing these groups reveals which behaviors predict long-term value.
Predictive Cohorts
Predictive cohorts use machine learning to group users likely to churn or convert before it happens. Tools like Amplitude and Mixpanel now offer this out of the box. You segment by predicted behavior, then intervene early with targeted campaigns.
How to Run a Cohort Analysis: Step by Step
Step 1: Pick one north-star action. This is the behavior you want users to repeat. For a media app it might be "watched 3 or more videos." For SaaS, "logged in and completed a core task." Do NOT use vanity events like "opened the app." Use the action that actually predicts whether someone will pay you next quarter.
Step 2: Choose your cohort window. Use weekly cohorts for fast-moving consumer apps with high daily activity. Use monthly cohorts for SaaS products. Use quarterly cohorts for enterprise deals with long sales cycles.
Step 3: Build the triangle table. Rows equal signup week or month. Columns equal time periods since signup. Cells equal the percentage of that cohort that performed your north-star action. Most analytics tools (GA4, Amplitude, Mixpanel) build this table automatically.
Step 4: Read the curve shape, not just the number. A healthy retention curve drops steeply in the first few periods, then flattens out. The flat tail means you have a loyal core user base. A curve that never stops falling means you have no retention at all, you are refilling a leaky bucket.
Step 5: Segment before you celebrate. One cohort table for all users is almost always misleading. Split by acquisition channel (organic search vs paid ads vs referral), plan tier (free vs paid), and onboarding variant if you have one. These subgroups behave completely differently.
Step 6: Compare cohorts vertically. If the cohort that went through your new onboarding flow shows 38% retention at week 4, versus 28% for the previous cohort, the change worked. That vertical comparison is the entire point.
Spotify and Netflix: Real Cohort Wins
Spotify ran cohort analysis on acquisition channels and found that users acquired through social media campaigns had 25% lower six-month retention than users coming from search advertising. They reallocated budget toward search and reduced social spend. The result was a lower volume of signups, but a significantly higher percentage who became long-term paying subscribers.
Netflix applied cohort analysis to content launches and found that subscribers who joined during a major tentpole release (like the "Stranger Things" Season 2 premiere in 2017) had 18% higher year-one retention than subscribers who joined during quieter periods. This finding directly shaped how Netflix sequences big releases: use them as acquisition drivers, then measure whether the cohort that arrives actually sticks around. Both examples are documented in FasterCapital's retention modeling research.
How to Do This in Google Analytics 4 (GA4)
GA4 includes a built-in cohort exploration tool under the Explore tab. Here is the fastest path to your first cohort table:
- Open GA4 and click Explore in the left sidebar.
- Click Template gallery and select Cohort exploration.
- Set your Cohort inclusion criteria, this defines who enters the cohort (for example: users who made a purchase, or users who triggered a specific event).
- Set your Return criteria, this defines the behavior you are tracking over time (usually session_start or a key action).
- Choose your granularity: daily (consumer apps), weekly (most products), or monthly (SaaS or e-commerce).
- Choose your calculation method: Standard (how many users were active in each specific period), Rolling (active in the current period AND all prior periods), or Cumulative (unique total from start through current period).
Important GA4 limitation: it tracks by device ID, not user ID. Someone who uses your site on both a phone and a laptop counts as two users. For more accurate data, connect GA4 to BigQuery or use a dedicated analytics tool like Amplitude or Mixpanel.
Common Mistakes
The four mistakes that waste your cohort data:
- Looking at one cohort in isolation. A single retention curve tells you nothing. You need at least three or four cohorts side by side to see direction: improving, flat, or decaying.
- Blending segments. Free trial users, self-serve paid users, and enterprise accounts have completely different retention shapes. Averaging them produces a chart that describes nobody accurately.
- Picking the wrong action. "Opened the app" or "visited the site" inflates everything. Choose the action that correlates with actual revenue or long-term engagement.
- Calling a winner too early. Cohort curves need 8 to 12 time periods before the flattening point becomes clear. A decision at week 2 is reading noise, not signal.
- Ignoring seasonality. A November e-commerce cohort always looks stronger than a February cohort, because holiday shoppers inflate both acquisition and early spend. Compare like-for-like time periods when possible.
Metrics Worth Tracking in Your Cohort Table
Beyond the basic "what percentage came back," here are the metrics that convert cohort data into business decisions:
- Day 1 / Day 7 / Day 30 retention: Industry shorthand for early activation quality. Poor day-1 retention usually means your onboarding fails, not your product.
- Cumulative revenue per user (ARPU by cohort): Which acquisition channel delivers users who spend more over 6 months? This reveals true CAC payback.
- First-to-second purchase rate: For e-commerce, getting a customer to buy twice is the most predictive indicator of lifetime loyalty. Industry average is around 30-40% within 90 days.
- Expansion rate: For SaaS, the percentage of users in a cohort who upgraded to a higher tier over time. This is the engine behind NRR above 100%.
- 6-month survival curve: The cohort percentage still active after 6 months. This is the clearest single metric for product-market fit.
The One-Line Takeaway
If your retention curves are getting higher each month, you are building something people want. If they are not, no amount of new user acquisition will save the business.
