How do you build a retention curve and read it?
A retention curve plots the percentage of users from a given cohort who are still active at each time interval after their first use. If the curve flattens above zero, you have a product people genuinely want; if it slopes to zero, you're refilling a leaky bucket. Building and reading one correctly is the fastest way to know whether you have product-market fit before you waste money on growth.
Step 1: Define your cohort and your 'active' event
Before you touch a spreadsheet, you need two definitions locked down: what groups users into a cohort, and what counts as active. A cohort is typically the week or month a user first performed your core action — not the week they signed up, because signups include people who never meaningfully engaged. Your 'active' event should be the one action that delivers the core value of your product: a message sent, a file exported, a booking completed. Using a weak proxy like 'logged in' inflates retention artificially and hides real decay.
Once those are defined, group every user by their acquisition week (Week 0). Then for each subsequent week (Week 1, Week 2, Week N), count how many users from that original cohort performed your active event again. Divide by the original cohort size and multiply by 100 to get a percentage. That percentage at each week is one point on your retention curve.
Do this for multiple cohorts — at least 8–12 weeks of cohorts if your product is weekly, or 6–8 months if it's monthly. Stacking them lets you compare whether newer cohorts retain better or worse than older ones, which tells you whether your product improvements are actually working.
Step 2: Plot the curve and find the flatten point
Put time intervals on the X-axis and percentage retained on the Y-axis. Most curves drop sharply in the first two or three periods — this is normal, because any product loses a portion of users who tried it once and found it wasn't for them. The critical diagnostic is not the steepness of that early drop; it's whether the curve eventually flattens into a horizontal line.
A curve that flattens at, say, 25% at Week 6 and then stays near 25% through Week 12 means roughly a quarter of users found genuine, repeating value. That is the signal of a retained core. A curve that keeps declining toward zero at every time step means no segment of users has made the product a habit — you have an acquisition business, not a retention business.
The level at which your curve flattens matters enormously by category. A social network or daily utility needs to flatten somewhere above 20–30% at 8 weeks. A weekly SaaS tool might target 40–50% at 8 weeks. A monthly workflow product may aim for 60–70% at 6 months. There is no universal number, but you should benchmark against direct competitors and analogous products, not against abstract ideals.
Step 3: Segment to find what causes retention
A single aggregate curve lies to you. The real insight comes when you break the retained population apart. Segment by acquisition channel, by the feature used in the first session, by company size, by geography, or by whatever your product's natural dimensions are. You'll almost always find that one segment retains at 2–3x the rate of another. That high-retaining segment is your real customer — the one for whom the product already works.
Once you've found the high-retaining segment, work backwards: what did those users do in their first session that the churned users did not? This is the 'aha moment' analysis. It is not that the aha moment caused retention; it's that reaching it correlates with users who already had the problem your product solves well. Identifying it lets you redesign onboarding to help more new users reach that experience quickly rather than drift away before they ever see the product's real value.
Also run a cohort-over-cohort comparison: plot Week-8 retention for your January cohort, February cohort, March cohort, and so on as a trend line. If that number is rising steadily, your product changes are improving fit. If it's flat or falling despite shipping features, the features are not solving the core retention problem.
Step 4: Use the curve to make a concrete decision
A retention curve is only useful if it changes what you do next. There are essentially three situations and three corresponding responses. First: the curve is still sloping downward and you don't yet have 8–12 weeks of data on enough cohorts — in this case, don't scale paid acquisition yet and focus entirely on talking to users who churned at Week 2–4 to understand what failed. Second: the curve flattens but at a much lower level than comparable products — this means you have some real users, but the value is weak; the right move is to double down on the retained segment, understand what they use, and cut or deprioritize features the churned majority touched but didn't retain from. Third: the curve flattens at a healthy level — now you can model the business. Multiply your flatten level by new-user volume to project steady-state active users, and that gives you the denominator for monetization and growth planning.
Garry Tan's framing from his work at Y Combinator is that retention is the foundation everything else rests on. Without a flattening curve, growth spend accelerates churn rather than compounds value. The curve doesn't just tell you about retention — it tells you whether your product deserves to grow at all.
“A great deal of knowledge is unconscious, and experts have if anything a higher proportion of unconscious knowledge than beginners.”
— Paul Graham, source
The one thing to do
Plot your Week-8 retention rate for your last six cohorts right now — if that number is not trending upward and flattening above your category benchmark, fix retention before spending another dollar on acquisition.
Frequently asked questions
How many users do I need in a cohort to trust the retention curve?
You generally want at least 100–150 users per cohort before drawing conclusions, and 200+ for segment-level analysis. Smaller cohorts produce curves that are too noisy to act on — a single churned user can move your percentage by several points.
What's the difference between Day-N retention and Week-N retention?
Day-N retention measures whether a user came back on a specific day after signup, while Week-N measures whether they were active at any point during that week. For most products, weekly buckets are more reliable because they smooth out day-of-week effects — a Monday signup and a Friday signup are both counted fairly in the same Week 1 window.
My curve flattens but at 4%. Is that good enough?
It depends entirely on your category. For a high-frequency consumer app, 4% is a serious problem. For a niche B2B workflow tool with a very small total addressable market, 4% retention of a paid cohort might still be a viable business — run the unit economics to find out. The number only has meaning relative to your customer acquisition cost, average contract value, and market size.
Should I use absolute or relative retention?
Use relative retention (percentage of original cohort) for product-market fit diagnosis, because it normalizes for cohort size. Use absolute user counts when planning infrastructure or forecasting revenue, because percentages alone don't tell you whether your retained base is 50 people or 50,000.
Sources
- gstack: skillify/SKILL.md — Garry Tan
- Putting Ideas into Words — Paul Graham