Retention Curves
The shape tells the story before the numbers do
A retention curve plots retention over time. Its shape — steeply decaying, flattening, or flat — is the product's signature, and category curves give every company a truth-teller in the room.
What you will learn
- Plot and read a retention curve for a cohort
- Distinguish healthy flattening from linear decay
- Benchmark against category curves and fit indexes
Retention curves show % of a cohort active over time. A curve that flattens to a stable plateau signals product-market fit; a linearly decaying one means users try, stop, and never harden into a habit.
What is this?
A retention curve plots a cohort's retention (Day-N or rolling) against days/weeks since entry. The shape encodes product health: the classic healthy curve drops fast in week one, then flattens onto a plateau that holds; the unhealthy one keeps sliding down a straight line, meaning no subset of users is forming a habit. Reading curves next to category benchmarks (messaging apps hold high plateaus; marketplace supply/demand removes; B2B tools are weekly) prevents both despair and delusion.
Why do you need it?
- The plateau level and its steepness are the numeric definition of product-market fit.
- Curves reveal cohorts at a glance: a curved-shift is instantly noticeable before any average changes.
- Decay curvature is diagnostic — monthly-decayers differ from weekly-commerce-decayers.
- Once benchmarked, curves remove 'is this good or bad?' from retention reviews — the category norm is the referee.
Mental model
Ice melting vs ice holding
Decaying vs flattening retention curves
A cube left out slides into a shallow puddle (bad curve). A thermos jug with a stable temperature holds its level (good curve). Melting is linear; holding is horizontal, and the thermometer read says which.
How it works
- Take one cohort (or overlay several vintages) and compute retention per period.
- Plot period-since-entry on x, % active on y.
- Fit or eyeball the shape: steep early slope → onboarding/expectations; continued slope → lack of repeat value.
- Compute a flattening index (e.g. D30/D7 ratio) or benchmark curve from category data.
- Read the cohort trend: overlaid vintages that flatten higher show the product steadily improving.
Three curve shapes.
In production
- A messaging startup presents its D30 curve vs the category band to a board: above-band = retention story, below = honest risk disclosure.
- A fitness app watches its weekly curve flatten after adding streak mechanics — a second derivative change, not just slope.
- An e-commerce merchant sees a December-vintage curve spike (holiday buyers) and learns to exclude that vintage from the fit benchmark.
Best practices
- Plot curves per vintage and overlay them; a single averaged curve hides the trend.
- Use the category benchmark band before drawing conclusions from any shape alone.
- Report the flattening index explicitly (e.g. D30/D7 = 0.60) so 'flatness' is a number, not a vibe.
- Choose Day-N vs rolling deliberately per product type and keep the choice constant in comparisons.
Common mistakes
- Judging a curve 'bad' without the category norm — a 0.5% D30 can be elite in enterprise but fatal in consumer.
- Averaging vintages with different definitions and producing a nonsense blended shape.
- Calling the early steep drop 'onboarding failure' when it's simply the natural win-now of weak intent.
- Reading curves off different units (days vs weeks) side by side.
Debugging tips
- A curve that doesn't drop at all in week one means your active definition includes first-open-only events.
- When vintages diverge, check the acquisition channel mix per vintage before blaming product.
- Flattening index drifting month to month usually tracks a release or pricing event — annotate the axis.
- Recompute the curve from raw events when two tools disagree; the shape is sensitive to dedupe details.
Interview checkpoint
Hands-on lab
- Generate a cohort and compute D0–D6 retention
- Plot the curve and compute a flattening index
- Write a fit verdict with the numbers
Plot and Interpret a Retention Curve
30 min · beginner
Quick quiz
A D30 retention curve that keeps sliding down linearly most likely means:
Interpreting a retention curve without its category benchmark is generally safe.