Subscription and media analytics
Subscription analytics shows which offers bring subscribers who stay at full price, and what you can spend to win each. Publishing analytics adds which articles turn anonymous readers into registered ones, then subscribers. Both need retention read by cohort and by offer.
Retention by acquisition offer, example
220 of 810 discount subscribers, 27.2%, left when the offer ended
- Month 4
- Discount cohort810 to 590, 27.2% left
- Full price800 to 775, 3.1% left
- Blended, both cohorts1,610 to 1,365, 15.2% left
Show the numbers
| Month | Full price | 50% off for three months |
|---|---|---|
| 0 | 1,000 | 1,000 |
| 1 | 900 | 940 |
| 2 | 840 | 875 |
| 3 | 800 | 810 |
| 4 | 775 | 590 |
| 5 | 750 | 545 |
| 6 | 728 | 510 |
| 7 | 708 | 482 |
| 8 | 690 | 460 |
| 9 | 675 | 443 |
| 10 | 662 | 430 |
| 11 | 650 | 419 |
| 12 | 640 | 410 |
Subscription and media metrics, with formulas
Eight KPIs with formulas, built on contribution.
Eight media metrics and subscription KPIs, each with the formula that goes into the model. "Contribution" in every one means subscription revenue minus payment fees and the content cost you've agreed to allocate per subscriber.
| KPI | Formula | A good reading | The common mistake |
|---|---|---|---|
| Cohort retention at month n | Subscribers from the cohort still paying at the end of month n ÷ subscribers who started in the cohort | A curve that flattens after the first full-price renewal | One blended curve across every offer, which hides a discount cohort leaving at the price step |
| Monthly churn rate | Subscribers lost in the month ÷ subscribers paying at the start of the month | Read per cohort and offer, with the price step months marked | Adding the month's new joiners to the denominator, which makes churn look lower in a heavy acquisition month |
| Involuntary churn share | Subscriptions ended by a failed payment after all retries ÷ all subscriptions ended | Falling, and read again after any change to how failed payments are retried or chased | Counting a failed payment as churn on the day it fails, before the retries have run |
| Contribution LTV | Sum of a cohort's contribution over a stated horizon, such as 24 months ÷ subscribers who started in the cohort | Quoted with its horizon every time | Using revenue, which leaves out payment fees and content cost, or giving no horizon at all |
| CAC | Acquisition spend ÷ new paying subscribers | Below contribution LTV for the same offer | Dividing by free-trial starts, when many trials never pay |
| LTV to CAC | Contribution LTV over the stated horizon ÷ CAC | Above 1 within the horizon, with the horizon written next to the ratio | Quoting the ratio with no horizon, so it grows as far as anyone wants by stretching the months |
| Payback months | The month in which a cohort's cumulative contribution per starting subscriber first reaches CAC | Shorter than the horizon used for LTV | Dividing CAC by full-price monthly contribution, which ignores the discount months and the subscribers who left |
| Paywall conversion per exposure | Subscriptions started after a paywall view ÷ paywall views, per article | Read per article and per section, for anonymous and registered readers apart | Dividing by all page views or all users, which mixes readers who never met the paywall into the base |
More worked examples are in the KPI and dashboard library.
Where subscription data usually lives
Billing, app stores and more, and where joins break.
A blended churn rate of 15.2% is how a 27.2% exit at the end of an offer goes unnoticed.
Churn analysis by cohort and offer
Retention read by the offer each subscriber joined on.
A cohort is every subscriber whose first paid month is the same, split by the offer they joined on. Each cohort gets its own retention curve, and each offer's price step is marked on it. In the example, the half-price cohort keeps more subscribers than full price for three months. Then the offer ends and 220 of 810 leave in a single month. Read together, the two cohorts lose 15.2% that month, which looks like an ordinary bad month. Read by offer, the discount is the obvious cause.
The decision this changes is the offer itself. A discount is worth running if the subscribers still paying after the price step bring enough contribution to cover the discount and the extra cancellations. That sum needs month-12 retention, because at month 3 nobody on the offer has paid full price yet.
LTV to CAC on contribution, with payback months beside it
How to build LTV to CAC that holds up.
LTV to CAC only means something when LTV is built on contribution and has a horizon. Revenue LTV leaves out payment fees and the content cost each subscriber carries. An LTV with no horizon can be made as large as anyone likes by assuming subscribers stay for years. So the ratio is quoted with its months, for example 24, and payback months sits next to it, because payback shows how long the business funds each new subscriber.
Subscription analytics services that fit
The services subscription businesses need most, and why.
Read retention by cohort and by offer before you judge a channel or a discount. A blended churn rate hides the cohort that leaves at renewal.
Publishing analytics: registered and anonymous readers
How to count anonymous and registered readers.
AI in subscription and media analytics
What an AI needs before it can explain churn.
Ask an AI assistant "why did churn rise in March?" and it can only give a useful answer if cohort, offer and price step are columns it can query. With them, it can say whether March was the month a discount cohort reached full price. Without them it reports a blended rate and guesses at a cause. AI-led analytics builds those definitions first and shows the query behind every answer.
Questions people ask
Answers on churn, paywalls, cancellations and GA4.
Do you build churn dashboards for subscription businesses?
Yes. The core of one is a retention curve per monthly cohort, split by the offer each subscriber joined on, with the price step for every offer marked on the chart. Blended churn can sit beside it for the board.
Can you run paywall A/B tests for publishers?
Yes. The sample size is fixed before the test starts, from the current conversion per paywall exposure and the smallest lift worth having. A test that brings in more discounted subscribers is also read at the first full-price renewal, because that's where the difference shows.
How do you measure marketing for subscription businesses?
On subscribers who stay past the end of their offer. A channel that brings cheap sign-ups who leave at the price step can look like the best channel for three months. Where spend is large enough, a geo holdout test shows how many subscriptions it actually added.
How should we count a subscriber who has cancelled but still has access?
Pick one rule and write it down. We count a subscriber as lost at the end of the period they've paid for, because they can still read until then. Whichever rule you choose, churn, retention and LTV all have to use the same one.
Does GA4 work for publishers?
For anonymous reading, yes, with care. GA4 tells users apart by first-party cookie, so one reader on a phone and a laptop counts twice. Registered and subscribed readers are better measured on their own login ID, joined to the billing system.
Want retention split by offer?
Bring a year of subscription data with the offer codes. The first call works out whether your discount cohorts pay for themselves.