Loading tutorials…
Loading tutorials…
Retention is the single most important SaaS metric. But Amplitude has three retention models and they answer different questions. Pick the wrong one and you'll think you're growing when you're actually leaking users.
Who this is forSaaS teams, product-led growth companies, and consumer apps where repeat usage is the core success metric. Especially relevant for teams trying to figure out if a recent product change actually improved retention.
What you'll need
Step 1
Amplitude → Charts → "New Chart" → "Retention Analysis." The 2026 UI groups all retention models under one chart type.
Open Amplitude → Charts. Click "New Chart" → "Retention Analysis."
You'll see two event slots: "Start Event" and "Return Event." Plus a "Retention Model" selector with N-Day, Bracket, and Unbounded options.
Name the chart from the start: "Web users — Signup to Active L30, N-Day." Generic names lose their meaning fast.
Save to a Notebook. Retention charts get referenced more often than funnels — make them easy to find.
Step 2
Start Event = the moment the user enters the cohort. Most teams use "Signup Submitted." But "First Active Session" is often more honest for product-led growth.
"Signup Submitted" is the classic start event. Counts everyone who signs up, including those who never come back. Honest for measuring acquisition quality.
"First Active Session" or "First Feature Used" filters to users who actually engaged. Shows retention conditional on activation — more useful for product decisions.
For paid SaaS, "Subscription Started" as start event measures paid-user retention specifically. Different question than overall retention.
Filter the start event by property if needed. Example: `signup_source = paid_search` shows retention of paid acquisition specifically.
You can only pick ONE start event per chart. Build multiple charts for multiple cohorts.
Step 3
Return Event = the action that proves the user is still active. "Any Event" makes retention look great but is meaningless. Pick a value-receiving action.
Bad choice: "Any Event." Includes users who opened an email tracking pixel — not actually using the product. Inflates retention by 30-100%.
Better choice: a specific value-receiving action. SaaS: "Workspace Opened," "Feature Used," "Report Generated." Content products: "Article Read." E-com: "Product Viewed."
Best choice: your North Star action. The one event whose absence means the user is churning even if they're logging in.
Test: pick a return event, then check your retention number against your subscription churn rate. They should be related. If your retention chart shows 70% Week 4 but you have 30% monthly subscription churn, the return event is too loose.
You CAN use multiple events as Return — set "performed [event A] OR [event B]." But avoid this until you understand single-event retention first.
Step 4
N-Day: user must return on exactly day N. Bracket: user must return within a day-range (e.g., days 7-13). Unbounded: user must return on or after day N (most lenient).
N-Day retention: "% of users who returned on EXACTLY Day 7." Strictest. Use for daily-use products (Slack, Notion) where consistent return matters.
Bracket retention: "% who returned in Days 7-13." More forgiving. Use for weekly-use products (Linear, GitHub) where the user returns weekly-ish, not on a specific day.
Unbounded retention: "% who returned ON or AFTER Day 7" (cumulative through today). Most lenient. Use for monthly-use products or when you want the most positive-looking chart.
Most B2B SaaS dashboards use Bracket because the work cadence is weekly, not daily.
Switch between models on the same chart — Amplitude lets you toggle. Compare; the gap between N-Day and Unbounded tells you how habitual your product is. Big gap = users come back, but irregularly. Small gap = users either return on schedule or churn.
Step 5
Retention curves drop fast at first, then ideally flatten. The flat portion is your "retained user" baseline. The steeper the decay, the worse retention is.
Day 0 to Day 7 drop: usually 40-70% for SaaS. This is the activation cliff — users who signed up but never came back.
Day 7 to Day 28: a steep continued drop here means activated users still aren't finding value. Investigate onboarding.
Day 28 onward: this should approach flat. A flat retention curve at 20% means 20% of users become long-term users. That's your retained base.
Bad shape: continuously declining curve, never flattens. Means even your "retained" users churn over time. The product has a retention problem at every stage.
Good shape: steep drop early, then flat. Means once users get past activation, they stay. Focus all eng effort on activation.
Step 6
Group retention by signup month, plan tier, or acquisition source. The aggregate hides the actionable insight.
Group by signup cohort (Day/Week/Month) to see if recent changes are improving retention. Look at the curves side-by-side — newer cohorts at the top is the goal.
Group by `plan_tier` to see if paid users retain better than free (usually yes, by 2-3x). The gap quantifies the value of converting free → paid.
Group by `utm_source` to see retention by acquisition channel. Paid social usually retains 20-50% worse than organic. Worth knowing before scaling that channel.
Group by `feature_first_used` (first feature the user touched) to see which onboarding paths lead to better retention. The winning feature should be the default in your onboarding flow.
Compare two cohorts: "Pre-product-change" vs "Post-product-change." If retention diverges 14+ days post-change, the change is working.
Step 7
Add annotations for product/marketing changes. Set alerts on retention shifts. Six months later you'll thank yourself.
Settings → Annotations → "Add annotation" on every product change. "May 12 — removed credit card requirement on free trial." Becomes searchable forever.
Alerts → "Notify when Week 4 retention drops by 10%." Slack alert lands in your inbox 24 hours after the drop, not three months later.
Save the retention chart to a Notebook with a written narrative: "What we measure, what 'good' looks like, what we change when retention drops." This is the institutional memory.
Share the Notebook in the team Slack channel monthly. Retention charts that aren't seen don't get acted on.
Common mistakes
Using "Any Event" as the return event
What goes wrong: Retention looks 30-100% higher than reality. The team thinks the product is sticky when 50% of "retained" users are just opening tracking-pixel emails. Strategy decisions ($50K-300K of eng investment) get made on a phantom metric.
How to avoid: Pick a specific value-receiving action ("Workspace Opened," "Report Generated") as Return Event. Compare retention numbers before/after the change — the drop is the lie you were telling yourself.
Reading retention before enough data has accumulated
What goes wrong: You compute Week 12 retention on a cohort that signed up 4 weeks ago. The number is artificially zero (no one has had 12 weeks yet). You panic and rebuild onboarding. Wasted: $15K-40K in eng.
How to avoid: For Week N retention, you need at least N weeks of data per cohort. Use Bracket retention for partial data — it surfaces only what's actually measurable.
Single retention chart for all users
What goes wrong: Free users tank retention. Paid users have great retention. Aggregate looks mediocre. Team focuses on the wrong intervention because they can't see the segments. Lost product velocity: 3-6 months.
How to avoid: Always group by plan tier or acquisition source. Run separate charts for free vs paid. Decisions are made at the segment level, not the aggregate.
Comparing N-Day across different products
What goes wrong: You benchmark "20% Day 30 retention" against a SaaS peer. They're using Unbounded retention; you're using N-Day. You're actually doing better than them but the metric mismatch makes you think you're behind. Drives bad strategic decisions.
How to avoid: Confirm the retention model before benchmarking. When in doubt, ask "is that N-Day, Bracket, or Unbounded?" If they don't know, the benchmark is unreliable.
No annotations on the retention chart
What goes wrong: Retention dropped in March. The team spends 2-3 weeks investigating, only to remember a pricing change shipped that same month. 40-60 hours of analyst time spent re-discovering what an annotation would have made obvious.
How to avoid: Every product/marketing change gets an Annotation on the retention chart. 30 seconds per change. Pays for itself the first time you avoid a forensic investigation.
Recap
Done — what's next
How to build Amplitude funnels that actually answer business questions
Read the next tutorial
Hand it off
Retention is the metric that compounds — improving it 5% turns into massive LTV gains over a year. A specialist who has lived inside dozens of SaaS retention curves brings pattern-matching you can't shortcut. Initial retention audit + chart build typically $200-500; ongoing analysis $400-900/mo at $14-16/hr.
See specialist rates
Minimum 8-12 weeks of data to read a 30-day retention chart trustworthily. For 90-day retention, wait 4-6 months. Earlier reads have right-truncation bias and will mislead. Use Bracket retention to surface only the truly-measurable windows.
Amplitude defaults to classic cohort retention (X% of Day-0 users return on Day N). Rolling retention (X% of users active in week N-1 are active in week N) is available on Plus plan via custom SQL. Classic is fine for 90% of use cases.
Two different metrics. Retention is users who keep using the product; MRR churn is users who keep paying. A user can keep paying without using (low engagement, will eventually churn) or use without paying (free tier). Track both — divergence is signal.
Highly variable by category. B2B SaaS: 30-50% Day 30 is healthy. Consumer apps: 10-25% Day 30. Developer tools: 50-70% Day 30. Anything under 10% Day 30 means activation is broken. Never benchmark without confirming the return event + retention model.
Click any data point in the retention chart → "Users at this point" → list of user_ids. Click into User Lookup to see their event stream. Useful for qualitative analysis of "what did the churning users do differently from retained users?"
Amplitude
Funnels are the most-used chart type in Amplitude and the most-misused. The defaults assume an attribution window that doesn't match most SaaS sales cycles. Here's how to build funnels that match reality.
Amplitude
Cohorts and Personas are where Amplitude beats GA4 by a mile. But most teams build sloppy cohorts and end up with overlapping segments that conflict. Here's the discipline that keeps them clean.
Amplitude
Funnels show you the path you expected. Pathfinder shows you the paths users actually take. The gap is where 60% of product insight lives — and where most teams never look.
Amplitude
DIY Amplitude is a great idea — until your taxonomy gets out of control or your charts disagree with reality. This is the honest framework for when the math flips toward hiring.