A subscriber total is a stock: new people flow in, others flow out, and the number in the middle tells you little about either. Cohort analysis for subscriptions fixes that. You group subscribers by when or how they joined, then follow each group over time. This guide explains how to measure subscriber retention by cohort, build a retention table from your own billing data, and read what it tells you.
What a cohort is
A cohort is a group of subscribers who share a starting point. Because everyone in the group started together, you can compare like with like: how many of the people who joined in March were still paying three months later, against the same figure for April.
Two ways of cutting cohorts do most of the work.
Cohorts by join month
This is the default. Everyone who first paid in a given calendar month belongs to that month’s cohort. Join-month cohorts show whether retention is improving over time, and they make the effect of a change visible: if you redesigned onboarding in May, the May and June cohorts should behave differently from the ones before.
Cohorts by entry point
Join month tells you when. Entry point tells you how. Useful entry points for a media brand include:
- the piece, episode or video someone engaged with just before subscribing
- the offer: full price, introductory discount, gift or bundle
- the channel: newsletter upgrade prompt, paywall, podcast mention, social post or referral
- the plan: monthly or annual, individual or group
Entry-point cohorts answer the questions that shape a business, such as whether people who arrive through the podcast stay longer than people who arrive through search, or whether a launch discount brought in subscribers who leave the moment it ends. To use them, capture the entry point at sign-up and store it against the subscriber. It is very hard to reconstruct later.
How to build a subscription retention table
You can build a first version in a spreadsheet from a billing export. You need one row per subscription with:
- a subscriber ID
- the date of first payment
- the plan and billing interval
- the entry point, if you have it
- the date the subscription ended, or blank if it is still active
Then work through these steps:
- Define “retained”. The cleanest definition is “has paid for, and has access to, month N”. Decide whether a subscriber in a failed-payment grace period counts, then apply that rule everywhere.
- Assign each subscriber to a cohort by the month of their first payment.
- Count each cohort’s starting size. This is month zero, always 100%.
- For each later month, count how many are still retained and divide by the starting size.
- Lay it out as a triangle. Cohorts run down the side and months since joining run across the top. Recent cohorts have fewer columns because they have not had time to age.
Here is a small, entirely hypothetical example for a monthly plan, measured at the start of August. It is illustrative only, not benchmark data.
| Cohort | Starting subscribers | Month 1 | Month 2 | Month 3 | Month 6 |
|---|---|---|---|---|---|
| January | 400 | 82% | 74% | 70% | 63% |
| February | 350 | 80% | 71% | 67% | not yet |
| March (launch discount) | 900 | 66% | 52% | 45% | not yet |
| April | 380 | 84% | 77% | 73% | not yet |
Read across a row to see one group age. Read down a column to compare groups at the same age. In this example, March brought in more than twice as many subscribers as any other month, and a chart of total subscribers would have looked like a triumph. The table shows that fewer than half were still paying by month three. March still kept the most people in absolute terms, 405 at month three, so the discount was not a failure. It brought in a different kind of subscriber, and its cost should be judged on that basis.
Why month-three retention matters
If you track one cohort figure, make it month-three retention for monthly plans: the share of a cohort still paying after three billing cycles. By then, people who subscribed on impulse, for a single piece or a discount, have usually made their decision. The ones left have started a habit.
Month three is early enough to act on. You can judge a change to onboarding or pricing three months after making it, rather than waiting a year. It is also late enough to be meaningful, unlike month-one retention, which is heavily shaped by people who only came for one thing.
Reading the curve: where it flattens
Plot each cohort’s retention against months since joining and you get a curve that drops steeply at first and then levels off. The point where it flattens is the most important thing on the chart.
- A curve that flattens means you have a core of subscribers who stay. The height of that floor, and how soon you reach it, tells you what a new subscriber is worth over time.
- A curve that keeps falling at a steady slope means even long-standing subscribers are leaving. That usually points to the product itself: the cadence has slipped, the focus has drifted, or the value has stopped renewing.
- Newer cohorts flattening higher than older ones means your changes are working. This is the clearest evidence you will get.
Early drop-off and long-term decay have different causes and different fixes. Steep early losses are usually about onboarding and expectations, the gap between what people thought they were buying and what arrived. Slow ongoing losses are about habit and value, which is where engagement metrics help explain what the cohort table shows.
Common pitfalls in cohort analysis
Mixing annual and monthly plans
An annual subscriber cannot churn in month two, because they have already paid for twelve months. Put annual and monthly subscribers in the same table and annual plans will flatter every early column, then produce a cliff at month twelve. Keep separate tables, and for annual plans measure the renewal rate at each anniversary.
Counting free trials as subscribers
Start the cohort at first payment, not trial start, and track trial-to-paid conversion as its own figure. If cohorts start at trial sign-up, month one becomes a measure of trial conversion and hides what happens to people who actually paid.
Ignoring failed payments
Part of every cohort’s decline is involuntary churn from failed payments rather than choice. If you can tag each ending as cancelled or payment-failed, do it. The fixes are entirely different.
Reactivations
Someone who cancels in March and returns in June is easy to double-count. Pick a rule, usually that they stay in their original cohort and count as retained only in months they paid, and keep to it.
Reading too much into small cohorts
With 40 subscribers in a cohort, one person is 2.5 points of retention. Group small months into quarters, or wait until several cohorts agree before drawing a conclusion.
Turning the table into decisions
A retention table earns its keep when someone reads it on a schedule. Update it monthly and look for three things: whether the newest cohort started better or worse than the last few, which entry points produce cohorts that flatten highest, and what changed in the product to explain any difference. The answers should shape where you spend acquisition effort, how you think about pricing and discounts, and what you fix first. They also make the case for subscriber retention work far more clearly than a single churn percentage.
Malahide Studio sets up cohort reporting as part of running subscription products with independent media brands, so each month’s decisions rest on how real groups of readers, listeners and viewers behave. If you would like a second pair of eyes on your first retention table, get in touch.