If you’re trying to make sense of user behavior in your product, you’ve probably heard about “cohorts.” They’re not magic, but they are useful—if you set them up right. This guide is for product managers, analysts, or anyone who wants to cut through the noise and actually get answers from Amplitude. You’ll get a step-by-step approach, some honest advice about what’s worth your time, and a few pitfalls to skip.
Why bother with user cohorts?
Cohorts let you group users by shared traits or behaviors, so you’re not just looking at blobs of data. Want to see if new signups from the last campaign stick around longer than your average user? That’s a cohort. Need to know if features are hitting with your power users or falling flat with newbies? Cohort. Without cohorts, you’re mostly guessing.
But fair warning: It’s easy to overcomplicate this. Don’t create a dozen fancy segments just because you can. Start simple, test, and iterate.
Step 1: Understand what a cohort actually is in Amplitude
Before you dive in, here’s what you need to know:
- Cohorts are saved groups of users defined by events (what people did), properties (who they are), or both.
- They update automatically if you set them to be dynamic (recommended for most cases).
- You can use cohorts in funnels, retention, or any analysis where you want to compare apples to apples.
Pro tip
Cohorts in Amplitude aren’t retroactive for all data—if your event tracking is spotty, your cohort will be too. Garbage in, garbage out.
Step 2: Decide what you want to learn (and keep it specific)
Don’t just create a cohort because you read about it. Ask: “What real question do I have?” Some good starting points:
- Are users who completed onboarding more likely to return after a week?
- Do people from the latest marketing campaign convert better than last month’s group?
- Are power users (5+ sessions/week) using the new feature?
Pick one question. If you can’t phrase it in a sentence, it’s probably too vague.
Step 3: Log into Amplitude and find the Cohorts section
- Log into your Amplitude project.
- On the left sidebar, find “Users” and click “Cohorts.”
- Hit “Create Cohort” in the top right.
You’ll see a cohort builder with filters and options. This is where the magic happens—or the confusion, if you’re not careful.
Step 4: Define your cohort—choose the right filters
You’ll build your cohort with conditions based on:
- Event behaviors (e.g., “did Purchase at least once in last 30 days”)
- User properties (e.g., “Country is United States”)
- Combination logic (AND/OR)
Example 1: Users who signed up in the last 14 days
- Add filter: “Performed event” → “Sign Up” → “in the last 14 days”
- Set “most recent” if you only care about their latest signup
Example 2: Power users who used a key feature 5+ times
- Add filter: “Performed event” → “Used Feature X” → “at least 5 times in last 7 days”
Example 3: Users from a specific marketing campaign
- Add user property filter: “utm_campaign equals ‘spring_launch’”
Don’t overdo it. Every filter you add makes your cohort smaller and sometimes less useful. Start broad; narrow only if you have a reason.
Step 5: Set cohort type—dynamic vs. static
- Dynamic cohorts update automatically as user data changes. Great for ongoing analysis, like “active last 7 days.”
- Static cohorts are a snapshot in time. Use them if you want to freeze a group for a specific experiment.
Most of the time, dynamic is what you want. Static is helpful only for historical experiments.
Step 6: Name, save, and (optionally) share your cohort
- Give your cohort a clear, boring name. “Signed up last 14 days” beats “Superstar Segment Alpha.”
- Save it.
- Share with teammates if they’ll use it; otherwise, keep it to yourself. Too many shared cohorts just clutter things up.
Step 7: Use your cohort in analysis
Now the real value shows up. Here’s how to make that effort pay off:
Funnels
- Compare drop-off rates for “All users” vs. your cohort.
- Example: Are users from the new campaign bailing out at Step 2 more than regulars?
Retention
- See if “Completed Onboarding” cohort sticks around longer than average.
- If not, maybe onboarding isn’t as magic as you thought.
Event Segmentation
- Compare feature adoption between cohorts.
- Spot usage patterns that aren’t obvious in the aggregate.
Honest take
Most people waste time slicing and dicing data into a million cohorts and never draw any real conclusions. Focus on a couple of cohorts that actually tie to business outcomes. If you’re not acting on the insight, it’s just a vanity exercise.
Step 8: Iterate, clean up, and avoid common traps
- Delete unused cohorts once you’re done with an analysis. Otherwise, you’ll end up with a graveyard of confusion.
- Review definitions often. If your events or properties change, your cohort logic can break. Keep things tidy.
- Don’t get too fancy. Fancy “nested” cohorts (cohorts of cohorts) are rarely worth it unless you’re doing very advanced analysis.
Step 9: Advanced tips (only if you need them)
If you’re comfortable and actually need to go deeper, here’s what you can try:
- Import cohorts from CSV if you have user IDs from somewhere else (like a CRM).
- Combine cohorts using AND/OR logic for complex targeting.
- Sync cohorts to tools (like marketing platforms) if your plan supports it—but only if you have a real use case.
But honestly, most teams get the most value from a handful of simple, well-defined cohorts.
Summary: Keep it simple, revisit often
Cohorts in Amplitude are powerful, but only if they help you answer real questions and actually change what you do. Start simple, stay specific, and don’t be afraid to delete the stuff you don’t use. The best analysis is the kind you actually act on. So make a cohort, run the analysis, and move on. There’s always another question to explore.