The skill: Cohort analysis groups users by when they arrived (or what they did) and tracks how they behave over time. It's the only reliable way to tell whether your product is actually getting better or whether growth is just masking churn.
Benchmarks
In a Nutshell
- Pick one cohort definition per analysis. Signup week is the most common starting point. Don't mix signup-based and action-based cohorts in the same chart.
- Use relative time, not calendar dates. "Day 7 after signup" lets you compare cohorts fairly. "January 15th" doesn't.
- The shape of the curve matters more than any single number. A curve that flattens early means you have a retained core. A curve that never flattens means you have a leaky bucket.
- Three cohorts is the minimum for a trend. One cohort going up could be noise. Three in a row is a signal.
- Segment by channel before celebrating. A retention improvement might just mean you shifted ad spend from low-quality to high-quality channels. The product didn't change, the audience did.
- Small cohorts lie. A 50-user cohort showing 40% Day-7 retention has a confidence interval so wide it could be anywhere from 26% to 55%. Don't make strategy changes on small samples.
- Always show the full curve. Reporting "30% Day-30 retention" without the curve hides whether users churned gradually or all dropped off on Day 3.
- Compare to your own baseline first, benchmarks second. Improving from 8% to 12% Day-30 retention is a 50% improvement, even if the benchmark says "good" is 15%.
Reading the Curve
The retention curve tells you three things if you know what to look for.
Where it drops steepest is where you're losing people fastest. If Day 1 to Day 3 is a cliff, your activation flow has a problem. Users showed up, tried something, and left. If the drop happens between Day 7 and Day 14, users activated but didn't build a habit.
Where it flattens is where you've found your retained audience. This is the floor, the users who actually get value from your product. The earlier this flattening happens, the faster you know who sticks around. If the curve hasn't flattened by Day 30, you might not have product-market fit for that segment.
Whether newer cohorts flatten higher is the real measure of product improvement. If January's curve flattened at 8% and March's curve is flattening at 12%, something you did between January and March is working. That's the signal worth investigating. Was it a product change? A different acquisition channel? A pricing tweak? The cohort chart tells you that something changed. Your job is to figure out what.
Do's and Don'ts
Written with ❤️ by a human (still)