Quick Summary
- A cohort is a group of users who started in the same time period; tracking them separately reveals retention patterns hidden in aggregate metrics.
- Retention curves flatten over time; the goal is to raise the floor, not just the early slope.
- 2025 benchmarks show NRR varies sharply by ARR tier: $1M-$10M SaaS averages 98% while $100M+ companies hit 115%.
- Feature adoption is the strongest leading indicator: customers using 70%+ of core features retain at 2x the rate of light users.
- Week 1 activation is the highest-leverage moment: a 5% improvement there produces roughly a 20% improvement at Week 10.
What Is a Retention Cohort?
A retention cohort groups users by the date they first performed a key action, usually signup or first purchase, and then tracks what percentage of each group returns in subsequent periods.
Instead of asking "how many users are active this month?" a cohort analysis asks "of everyone who joined in January, how many were still active in February, March, and April?"
This distinction matters because aggregate active-user counts can grow while underlying retention decays. A product adding 1,000 new users per month can show a stable DAU even if 80% of every cohort churns in week one.
The Retention Curve
Plot cohort retention on a chart with time on the x-axis and percentage retained on the y-axis. The resulting line is the retention curve.
Three zones matter:
- The cliff (Day 1-7): the largest absolute drop. Most users who will churn do so immediately.
- The slope (Week 2-8): continued decay at a decelerating rate.
- The floor (Week 8+): the cohort stabilizes. Users who reach the floor tend to stay.
A healthy SaaS product sees its curve flatten above zero. A broken product sees curves that trend toward zero for every cohort.
2025 Benchmark Data
Understanding whether your retention is good requires external reference points.
NRR by ARR tier (Wudpecker, 2025 B2B SaaS report):
| ARR Tier | Median NRR |
|---|---|
| Under $1M | 91% |
| $1M, $10M | 98% |
| $10M, $100M | 107% |
| Over $100M | 115% |
NRR (Net Revenue Retention) above 100% means expansion from existing customers outweighs churn. Companies below $10M ARR should target 95%+ NRR before scaling paid acquisition.
Feature adoption and retention (Pendo, 2025): Customers who adopt 70% or more of a product's core feature set retain at 2x the rate of customers who use fewer features. This finding shifted the growth focus from "getting users to come back" to "getting users to go deeper."
Pricing model impact (Vitally, 2025): Products using usage-based pricing see 46% lower churn than equivalent flat-rate products. The mechanism is alignment: customers who pay for what they use are less likely to feel they are overpaying during low-activity months.
Early activation leverage (Pushwoosh, 2025): A 5% improvement in Week 1 retention produces approximately a 20% improvement in Week 10 retention. Week 1 is the highest-leverage window in the entire retention curve.
NRR and GRR measure different things. Gross Revenue Retention (GRR) can never exceed 100% because it excludes expansion revenue. NRR includes upsells and cross-sells, so it can exceed 100%. Track both: GRR reveals your churn floor, NRR reveals your revenue health.
Building a Cohort Table
A cohort table is a matrix where:
- Each row is a cohort (users who joined in a specific month or week)
- Each column is a time period after joining (Week 1, Week 2, etc.)
- Each cell contains the percentage of that cohort still active
Example cohort table (weekly, by signup week):
| Cohort | Week 0 | Week 1 | Week 2 | Week 4 | Week 8 |
|---|---|---|---|---|---|
| Jan W1 | 100% | 42% | 31% | 24% | 21% |
| Jan W2 | 100% | 45% | 33% | 26% | 23% |
| Feb W1 | 100% | 51% | 39% | 31% | 28% |
| Feb W2 | 100% | 53% | 41% | 33% | 30% |
Reading down each column tells you whether retention is improving over time. If Week 4 retention is rising across successive cohorts, your product improvements are working. If it is flat or declining, they are not.
How to Run a Cohort Analysis
Step 1: Define your cohort event
The cohort-entry event should be meaningful. Options include:
- Account creation (broad, includes tire-kickers)
- First value action (first post published, first file uploaded, first report generated)
- Payment completion (most meaningful for revenue cohorts)
For most SaaS products, "first meaningful action" produces more actionable cohorts than raw signups.
Step 2: Define your retention event
What counts as "retained"? This must match how your product creates value:
- A messaging app: sent a message
- A project tool: created or updated a task
- An analytics product: viewed or exported a report
Avoid using login as the retention event. Login does not confirm value delivery.
Step 3: Choose your time grain
- Consumer apps: daily or weekly cohorts
- B2B SaaS: weekly or monthly cohorts
- E-commerce: monthly or quarterly cohorts
Step 4: Segment cohorts by acquisition source
A cohort table that mixes organic signups with paid traffic hides important signals. Paid cohorts often show steeper early decay than organic. Segmenting by source lets you evaluate channel quality beyond CPA.
Step 5: Look for cohort-over-cohort improvement
Compare the same period (e.g., Week 4) across successive cohorts. A rising number means your product is retaining better. A flat or falling number means changes you made between cohorts have not improved retention.
Real Company Examples
Duolingo: DAU/MAU as a retention proxy
Duolingo's Q4 2024 earnings showed DAU/MAU at 37%, up from 26% in Q4 2022. The company attributed this to streak mechanics, push notification personalization, and the "Heart" system redesign. Their cohort analysis showed that users who completed three consecutive daily sessions in Week 1 had a 3x higher 30-day retention rate than users who completed only one session.
Netflix: Cohort thinking at scale
Netflix's Q4 2024 added 19 million net subscribers, its largest quarter. The company's retention strategy is driven by cohort-level analysis of content consumption in the first 30 days. New subscribers who watch at least one piece of content in week one and at least three pieces by day 30 churn at roughly half the rate of subscribers who do not. This drove their investment in "Day 1 release" strategies for tent-pole content.
AI-native SaaS: The 'tourist' cohort problem
Across AI-native SaaS products in 2024 and 2025, cohort analysis revealed a recurring pattern labeled the "tourist" problem: a large initial cohort signs up after a product launch or viral moment, uses the product once or twice, and disappears. Gross Revenue Retention for many AI tools started 2025 at around 27% and climbed toward 40% by mid-year as companies built onboarding flows specifically targeting tourist segments with guided first-use experiences.
If your cohort table shows a large drop between Day 0 and Day 1, and then a smaller drop from Day 1 to Day 7, your acquisition funnel is attracting the wrong users. If the large drop is between Day 1 and Day 7, your onboarding is failing users who had genuine intent.
Leading Indicators to Track
Retention is a lagging metric. By the time it shows decay, you have already lost users. These leading indicators predict retention before the curve turns down:
| Indicator | Why It Predicts Retention |
|---|---|
| Time to first value action | Users who reach value faster retain better |
| Features adopted in Week 1 | Breadth of activation correlates with long-term retention |
| Session frequency in Days 2-7 | Habit formation happens early or not at all |
| Support tickets in Week 1 | Confusion in onboarding predicts churn |
| Invite or share action | Social activation creates switching costs |
Build a health score combining these signals. Flag accounts scoring below threshold for proactive outreach in the first two weeks.
Common Mistakes
1. Treating aggregate retention as a product metric
Aggregate metrics blend cohorts. A growing product with improving retention can show a declining aggregate rate if early, large cohorts age out. Always segment by cohort.
2. Optimizing the slope instead of the floor
Reducing early churn is important, but the retention floor (the percentage who stay long-term) is more valuable. A product with a 25% floor beats one with a slower slope that still trends toward zero.
3. Using login as the retention event
Login is not value delivery. A user who logs in and immediately closes the app is not retained. Use an action that confirms the user received value.
4. Ignoring acquisition source in cohort segmentation
Paid and organic cohorts behave differently. Mixing them obscures channel-level retention problems and can lead to overspending on channels with poor long-term retention.
5. Acting on a single cohort
One cohort is anecdote. Three or more cohorts showing the same pattern is signal. Wait for consistent directional movement before changing the product based on cohort data.
6. Measuring retention without a defined activation milestone
If you have not defined what "activated" means for your product, your retention curve measures the wrong thing. Activation is the point at which a user has experienced enough value to have a reason to return. Define it, instrument it, and use it as your cohort entry event.
Tools for Cohort Analysis
- Mixpanel: best-in-class retention and cohort reports; free tier available
- Amplitude: strong for product analytics and cohort comparison
- PostHog: open-source, self-hostable, built-in retention tables
- Heap: auto-captures all events; useful when you do not have pre-instrumented events
- Looker / Metabase: SQL-based cohort tables for teams with a data warehouse
- Google Sheets / Excel: sufficient for small datasets; use pivot tables on exported event logs
For early-stage products, PostHog or Mixpanel's free tier is enough. Do not spend on analytics tooling before you have enough weekly active users to produce statistically meaningful cohorts.
