30-day revenue per install: the metric that decides a test
30-day revenue per install is the money still there a month later, divided by the cohort's installs. It is what says whether we keep the variant.
Article summary
- 30-day revenue per install is a cohort's retained money, divided by its installs. Refunds are already out.
- In the example in the article, not a measurement: €800 across 1,000 installs is €0.80. €600 across 1,000 is €0.60. Per payer, the variant looks better, €15 against €10.
- We iterate on the trial rate. We only call the new baseline after this revenue. If it falls while trials rise, the offer promised too much.
- It is not the month's ARPU, and it is not trial-to-paid. The window is thirty days after install, and the base is the install.
- Realized LTV is the reading after that, for apps that can wait. It is in the next article.
Summarize this article with
30-day revenue per install is the money still there thirty days after install, divided by that cohort's installs. The trial rate says who started. The trial-to-paid says whether the promise held. This one says whether the app won.
What is 30-day revenue per install?
Take the people who installed on the same day, or during the same test window. Wait thirty days. Add up what they paid that is still there, then divide by the number of installs. Not by payers, not by active users.
Retained means refunds are already out: a payment given back is not money earned. An open trial does not count either. At thirty days, a short trial is over, and the person has paid or left.
firstapp reads it on subscription apps that are already live. It is the third step on the Method page, after the trial rate and trial-to-paid, and before LTV. The offer is detailed on Growth.
Newsletter
Once a week.
Last issue
We opened 10 apps to see how they sell a subscription.
+534 people read us each week
How do you calculate 30-day revenue per install?
The formula:
30-day revenue per install = cohort retained revenue at D30 / cohort installs
The cohort is the set of installs we compare: control on one side, variant on the other. The denominator is fixed at install. It does not shrink when people leave. If it did, a base that melts would push the figure up without a euro coming in. That is the trap of ARPU, applied to a test.
💡 Example, not a measurement. The two rows are the same kind of app, the same price, the same day. This is not a firstapp figure.
| Installs | Trials | Payers | Retained revenue | Per install | |
|---|---|---|---|---|---|
| Control | 1,000 | 400 | 80 | €800 | €0.80 |
| Variant | 1,000 | 560 | 40 | €600 | €0.60 |
Trials go from 400 to 560, and it looks like a gain. Payments go from 80 to 40. Retained money goes from €800 to €600. Per install, the variant falls from €0.80 to €0.60.
Per payer, though, it looks better. €800 across 80 payers is €10. €600 across 40 payers is €15. The basket goes up, and the app took in less. Both sentences are true. Only one decides.
Why this is what decides a test
We read a test in an order, from the fastest to the most solid. The trial rate can be inflated by an offer people will not pay for. Trial-to-paid says whether the paywall promised too much. Retained revenue at thirty days, per install, is the money. The realized LTV comes after, for apps that can wait.
We iterate on the trial rate, because it answers quickly. We do not call the new baseline off it. If the variant that wins on trials loses on this revenue, it promised too much. In the example, it started 160 extra trials, and it lost €200.
We reread the winner by channel and by country. A figure that holds overall can lose on people who came from ads, often the ones who weigh the most. We also set aside what is not the test: a season, a spend change, a store feature, a price changed on the side.
On a rate, Fisher's exact test says whether the gap could be chance. The decision is read here, on the date written before launch. The frame is on the Method page. This article does not publish a threshold.
What this revenue is not
It is not trial-to-paid. Trial-to-paid starts from trials. This one starts from installs, and it counts euros. A rate that falls can come with revenue that rises, if the price is higher. A rate that rises can come with revenue that falls, as in the example. The article on trial-to-paid says how to read the rate. It does not keep the variant on its own.
It is not the calendar month's ARPU either. A month's ARPU mixes people who installed yesterday and people who installed last year, plus the season. 30-day revenue does not mix cohorts. It answers a precise question: did this wave of installs pay off, one month later.
An annual payment drops several months into the first thirty days. The figure goes up, and it has not said what the install will be worth after that. So we set the mix of plans beside it, as with ARPU. Weekly and annual are in weekly or annual.
When to read 30-day revenue
We only read installs that have reached the thirtieth day. Yesterday's are not this figure yet. Counting them means treating an open trial as money already retained.
The test window is written before launch. Two weeks at minimum when the decision is about this revenue: weekdays and the weekend do not bring the same people. We stop on the planned date. Cutting on the day the trial rate "is significant" manufactures false winners. The money is read after, once the cohort has thirty days.
What comes next, for apps that can wait, is realized LTV. Thirty days are enough to decide most tests. Further out, we check that the money holds.
The mistakes that distort 30-day revenue
Keeping the variant because trials go up. The trial rate is for iterating. In the example, trials go up, and the app loses €0.20 per install.
Dividing by payers. Revenue per payer says the basket. It goes up when small payers drop out. Revenue per install says whether the wave paid off. In the example, €15 against €10, and yet the variant lost.
Reading before the thirtieth day. A three-day trial chose early. Refunds and cancellations from the first weeks have not finished moving. The figure at D10 is not the figure at D30.
Comparing two cohorts that do not have the same mix of plans. An annual on one side, a weekly on the other, and D30 is no longer talking about the same thing. We write the mix next to the amount.
FAQ
What is 30-day revenue per install?
It is the money still there thirty days after install, divided by the installs in the same cohort. Refunds are out. The base does not shrink when people leave.
Why not keep a test on the trial rate?
The trial rate answers quickly, and it can be inflated. We iterate on it. We only call the new baseline after retained revenue. If trials rise and this revenue falls, the offer promised too much.
How is it different from ARPU?
A month's ARPU mixes every active person in the period. 30-day revenue follows a cohort of installs, thirty days later. Dividing by payers gives a different figure: in the example, €15 against €10, while the app took in less.
When do you call the winner?
When the installs we compare have reached the thirtieth day, on the date written before launch. The test covers at least two weeks. We reread by channel and by country before keeping the variant.
What do we read after thirty days?
Realized LTV, for apps that can wait. Thirty days decide the test. Further out, we check that the money holds. It is the article on realized LTV.
How do you calculate it?
Cohort retained revenue at D30 / cohort installs. In the example in the article, not a firstapp measurement: €800 across 1,000 installs is €0.80, and €600 across 1,000 is €0.60.