Cohort Analysis for Subscriptions: Reading the Curve

Cohort retention is the only subscription metric immune to your growth rate. Everything else moves when your mix changes; this does not.

Building the table

Group subscribers by signup month. For each group, track the percentage still active at each subsequent cycle.

Cohort	C1	C2	C3	C4	C5	C6
Jan	100%	84%	76%	71%	68%	66%
Feb	100%	86%	79%	74%	71%	—
Mar	100%	81%	71%	65%	—	—
Apr	100%	88%	82%	—	—	—

Read it two ways.

Down a column — is retention at a given cycle improving over time? The C2 column above goes 84, 86, 81, 88. March was bad; April was the best yet. Something changed in March worth investigating, and something in April worth repeating.

Along a row — where does each cohort lose people? Every row above loses most of its subscribers between C1 and C2. That is where the leverage is.

Cycles, not calendar months

Use cycle number, not elapsed months, for the columns.

A subscriber on an eight-weekly plan reaches cycle 3 after six months; a fortnightly subscriber reaches it in six weeks. Putting both in a “month 3” column compares different points in their lifecycle and blurs exactly the signal you want.

Reading the shape

A curve that flattens is healthy. Steep early loss then a stable plateau means you have a committed core. Your work is at the front of the curve.

A curve that keeps declining is a problem. If you are still losing 5% per cycle at cycle twelve, there is no committed core and lifetime value is much lower than a flattening curve suggests.

A cliff at a specific cycle is diagnostic. A drop at cycle 3 on a prepaid 3-month plan is the prepaid term ending — you need a renewal sequence. A drop at cycle 6 might be when an introductory discount expires. Cliffs almost always have a specific, findable cause.

Segment the cohorts

The table becomes far more useful split by:

• Acquisition channel — some channels bring subscribers who churn immediately. Comparing CAC across channels without retention is how you scale spend on your worst channel.
• Entry product — the first product predicts retention more strongly than most merchants expect. If one entry product retains twice as well, that is what your ads should feature.
• Plan type — prepaid versus pay-as-you-go, and frequency chosen.
• Discount level — did the deeply discounted cohort retain, or did the discount buy subscribers who were never going to stay?

That last one settles a lot of arguments about promotions.

Using it for LTV

Cohort survival is the honest input to lifetime value. Rather than dividing contribution margin by a blended churn rate, sum expected contribution across the actual survival curve.

The result is usually lower than the naive formula for early cycles and higher for the tenured core — and the difference changes what you can afford to pay for acquisition. LTV.

The practical use

Cohort analysis answers “where should I spend my next effort?”

If the curve collapses between C1 and C2, onboarding and first-delivery experience are your priority, not win-back campaigns. If it flattens nicely but the plateau is low, acquisition quality is the issue, not retention. If a specific cohort broke, find what changed that month.

Without the table you are guessing, and the blended churn number will actively mislead you.