Reading a Retention Curve
The only question that matters is whether the curve flattens, and at what height.
Someone puts a retention chart on the screen. Day 1 is 62 percent, day 7 is 31 percent, day 30 is 14 percent. The room reacts to the drop from 62 to 31, because it is the biggest number on the chart. The drop is not the interesting part. Every product loses most of its signups quickly. What matters is what the line is doing at day 30, day 60 and day 90, and whether it is heading toward a floor or toward zero.
A retention curve: The share of a cohort still active plotted against time since signup. either flattens or it does not. If it flattens, some group of people found something they keep coming back for, and you have a business whose users accumulate. If it keeps sliding toward zero, every user you acquire eventually leaves, and growth is a treadmill: you must acquire faster than you lose, forever, and the faster you grow the more you must acquire.
The only question: does it flatten
Three shapes
- Smile
- Falls, flattens, then rises as the remaining users deepen usage or bring others in. Rare, and usually a sign of a product with a network or a strong habit.
- Flattening
- Falls steeply, then settles at a stable floor. Healthy. The floor is your real PMF: Product-market fit: the point where a market clearly wants what you built. number for that cohort: A group of users bucketed by when they joined, tracked over time rather than blended together., and every improvement to the floor compounds across all future cohorts.
- Sliding
- Keeps declining at every measured point with no sign of a floor. The product solved a one time need, or it never became a habit. No amount of acquisition spend fixes this shape.
Worked example
Hypothetical: Tidemark, a habit tracking app
Cohort of 1,000 signups. Week 1: 400 active. Week 4: 180. Week 8: 150. Week 12: 145. Week 16: 143. The line has clearly flattened at roughly 14 percent. Now the arithmetic that matters. If the floor is 14 percent and you acquire 10,000 users a month, your retained base grows by about 1,400 a month and keeps growing as long as acquisition holds. If instead the curve had kept sliding, say to 9 percent at week 12 and 5 percent at week 24 with no floor in sight, then the same 10,000 signups a month produces a base that plateaus and then decays the moment acquisition slows. Same top of funnel, entirely different business. This is why a two point improvement in the floor is worth more than a large improvement in day 1, even though day 1 is the number that moves fastest and feels most controllable.
Three things you have to get right for the chart to mean anything. First, define active as the action that represents value, not as opened the app. A retention curve built on app opens flatters you and will not correlate with revenue. Second, plot cohorts separately rather than blending everyone, otherwise recent signups drag the early points and long tenured users prop up the later ones. Third, give the curve enough time. You cannot tell a flattening curve from a sliding one at day 14. You need to see several periods where the slope stops changing.
Chasing the drop
- Focus on day 1 and day 7
- Onboarding tweaks, tours, nudges
- Wins are large and fade quickly
- Improves the height of the cliff
Chasing the floor
- Focus on weeks 8 to 16
- Core value, habit triggers, accumulated user data
- Wins are small and compound across every future cohort
- Improves the level the curve settles at
Segment the curve before you draw conclusions. A blended floor of 14 percent might be 40 percent for accounts that connected their calendar and 4 percent for those that did not, which turns a vague retention problem into a specific activation: The moment a new user first gets the value the product promised. problem you can work on. Andrew Chen's point about the flattening curve is that it tells you a real group exists. Segmenting tells you who they are.
Finally, resist the urge to compare your floor to a number you read somewhere. Retention benchmarks vary enormously by category, frequency of the underlying need, and how you define active. A weekly tool and a monthly tool should not be judged by the same curve. Compare your cohorts to your own earlier cohorts. That comparison is the one you can act on.
Quick check
Why does the floor of a retention curve matter more than the size of the early drop?
The takeaway
A curve that flattens means a group of users genuinely stuck, and lifting that floor compounds across every cohort you ever acquire.
Try this tomorrow
Plot your last six monthly cohorts on one chart using a value action as active, and mark where each one flattens.
Answer the check above, then bank the day.
Where this comes from
- The Cold Start Problem, Andrew Chen
- Lenny's Newsletter, on retention benchmarks, Lenny Rachitsky
- Hacking Growth, Sean Ellis and Morgan Brown
Metrics is one of six tracks. These lessons summarise and build on the work above, they do not reproduce it. Buy the books, they are better.