A first paid test costs less than most teams fear, as long as you buy the answers in order. A few hundred dollars tells you which ads get attention. $2,000 to $3,000 gets you first results on whether those people pay. About $10,000 a month on Meta, or closer to $20,000 on Google, tells you whether the channel holds at real volume. Three months and about $20,000, up to $50,000 if it goes well, tells you whether to keep going.
Each stage has its own stop rule, so a failed test costs you the price of that stage and no more. These are ad budgets only. Making the ads and paying whoever runs the campaigns come on top.
What are the four stages of a first paid test?
| Stage | What it answers | Budget | Stop when |
|---|---|---|---|
| Install test | Which ads get attention? | $20 to $30 a day per test, a few hundred dollars in total | An ad has spent $50 and has not earned its place |
| Purchase test | Do those people pay? | $2,000 to $3,000 for first results, up to $10,000 to reach 100% ROAS | $200 to $300 over about 3 days, decent clicks and installs, no trials or purchases |
| Channel test | Does the channel work at real volume? | About $10,000 a month on Meta, closer to $20,000 on Google | Cohorts of the same age miss your CPA ceiling on reconciled numbers |
| Three months | Is it worth continuing? | About $20,000, up to $50,000 if it goes well | Three months are up with no path to payback. You can pause at any point before that |
These are my planning numbers from running these tests, and your cost per purchase moves all of them. They cover Meta and Google on a consumer app. They are a way to plan a budget, and they make no promise about a result.
What should be ready before the first dollar?
Three things, because each one can waste the whole test.
- Events that measure revenue. If trials and purchases do not reach the platform and your MMP, no stage can give you an answer.
- A CPA ceiling. The most a paying user may cost you comes from your proceeds and how long you can wait for them. Without it, the third stop rule has nothing to compare against.
- Ads to test. A test with one ad tells you about that ad.
If retention falls away in the first few days, fix that first. More spend only scales the problem.
What does the install test tell you?
It tells you which ads people respond to, and it is the cheapest answer you can buy. An install campaign at $20 to $30 a day per test answers the question even when the account behind it spends six figures a month. On US install campaigns I read IPM, installs per 1,000 impressions, alongside spend. Below 2, the ads are not ready.
Spend picks the winners here. An ad that has spent $50 without earning its place stops, and the money goes to the ads that are working.
Do not stop at this stage and call it a result. IPM on its own can pick ads that win the auction and lose at the paywall. The install test only tells you what is worth taking to the next stage.
What does the purchase test tell you?
It tells you whether the people those ads bring will pay. The campaign now optimizes toward a trial or a purchase, and strong campaigns move fast. You should see first trials or purchases within 1 to 3 days.
This stage has the earliest stop rule. If you have spent $200 to $300 over about 3 days, clicks and installs look decent and there are no trials or purchases, stop. More budget will not fix it. Check the creative, the paywall, the tracking and whether the product has demand.
If purchases come in, keep going to $2,000 to $3,000 for first results. Reaching 100% ROAS can take up to $10,000.
Treat those first results as a direction and leave the verdict for later. Take an illustrative $2,500 at $30 per trial, which buys about 83 trials. At the 37.4% trial to paid rate in my CPA ceiling article, about 31 of those pay. That is below the 50 purchases I want before I read a cell, and on one account ROAS only predicted the next week best from 50 purchases, at a correlation that was still moderate.
Why does a channel test cost about $10,000 a month?
Because the platforms need volume to learn, and volume has a price. $10,000 a month is about $333 a day. At an illustrative $33 per purchase, that buys about 10 purchases a day. I aim for at least 5 to 10 optimization events a day per campaign, and Meta says ad sets usually leave learning after about 50 results in the week after the last significant edit, which is about 7 a day.
So the sum follows from your cost per purchase. If yours is $60, the same $10,000 buys about 5 a day, the bottom of that range. Above that, you need more budget, one fewer split in the account or an event higher in the funnel.
I plan closer to $20,000 a month for Google. Many teams wait weeks before it reaches stable performance, and Google’s own guide asks for an in app action completed by at least 10 different users a day and a daily budget of at least 10 times your target CPA. A $50 target means $500 a day, or $15,000 a month, before you count the slow start.
The stop rule here is slower on purpose. Young cohorts look worse than old ones until their revenue matures, so compare cohorts of the same age, on numbers reconciled between the platform, the MMP and your billing data. I cut when those cohorts miss the CPA ceiling. A bad week in the ad platform is not enough.
What does three months add up to?
About $20,000, up to $50,000 if it goes well. Creative tests take $2,000 to $3,000 in the first month. Two months near $10,000 on Meta brings you to about $22,000. Spend past that only makes sense when the numbers have earned it.
Three months is the cap, and you can pause at any point before it. If there is still no path to payback at the end, stop and work out why. Keep your money for the next attempt, because the test has already told you something.
Can you afford to wait for the money?
Work this out before you start, because ad platforms bill you long before the stores pay you. Apple pays within 45 days of the last day of its fiscal month. A purchase on the first day of that month can take more than two months to reach your bank.
For a team without a known renewal curve, I would recoup everything in the same month. Year two is a loan you take from yourself until you have the data to back it.
Advances can close the gap once the numbers work. RevenueCat Early Payouts advances up to 80% of store proceeds at a 2.5% fee, by application. Adapty Finance advances up to 85%, with a fee it puts at 2.5% to 3.25% depending on the app, and your app has to track revenue through Adapty. I am an Adapty Berlin ambassador, so weigh that mention accordingly. An advance speeds up money you have already earned. It does nothing for a test that is losing money.
What does a stopped test tell you?
The stage where it stopped points to the problem.
| Where it stopped | What it points to | Where to look next |
|---|---|---|
| Install test | People do not respond to the ads | New concepts, and a different promise in the first seconds |
| Purchase test, no purchases | Something after the click is broken, or demand is weak | Tracking, the paywall, the first minute in the app |
| Channel test, cost above the ceiling | The ads work but the economics do not, yet | The price and the trial, cheaper markets, the audience |
| Three months, no path to payback | Often the product and the audience | Whether this channel is your bottleneck at all |
I have stopped one myself. On a fitness app spending $1,500 a day, cost per trial fell from $60 to $90 down to $30 to $40 and then stayed there, against a $15 target, after more than 40 creatives. On day 25 we asked whether to keep spending, and the answer was not yet. The app had two audiences and one product, a product problem showing up in ad performance.
Sources and scope
The platform and payout pages were checked on 31 October 2026, and they change without notice. The budgets and stop rules are my planning numbers from my own work, and they are not benchmarks. The $33 and $60 per purchase, the $30 per trial and the $50 target are illustrations, and the sums on them are my arithmetic. The 37.4% is RevenueCat’s median for 5 to 9 day trials with all plans pooled, from the CPA ceiling article.
- Meta, About the learning phase, Meta Business Help Center
- Google, Best practices guide: Setting up your App campaigns, Google Ads Help
- Apple, Overview of receiving payments, App Store Connect Help
- RevenueCat, Early Payouts
- Adapty, Adapty Finance
Questions people ask
What is the cheapest way to test paid ads for an app?
An install test at $20 to $30 a day per test, a few hundred dollars in total. It shows which ads get attention. It says nothing yet about whether those people pay, so the next stage is a purchase test.
How much budget do I need to test Meta ads for a subscription app?
I plan $2,000 to $3,000 for first results on whether people pay, and up to $10,000 to reach 100% ROAS. A meaningful test of Meta as a channel takes about $10,000 a month. These are my planning numbers, and your cost per purchase moves all of them.
How much budget do I need to test Google App campaigns?
I plan closer to $20,000 a month, because many teams wait weeks before Google reaches stable performance. Google asks for an in app action completed by at least 10 different users a day and a daily budget of at least 10 times your target CPA.
When should I stop a paid app test?
Stop an ad in an install test once it has spent $50 and has not earned its place. Stop a purchase test at $200 to $300 over about 3 days if the top of the funnel looks decent and there are no trials or purchases. Stop a channel test when cohorts of the same age miss your CPA ceiling on reconciled numbers. Stop the whole plan after three months with no path to payback.
Can I get my app store money faster to fund a test?
Sometimes. Apple pays within 45 days of the end of its fiscal month. RevenueCat Early Payouts advances up to 80% of store proceeds at a 2.5% fee, and Adapty Finance advances up to 85% from 2.5%. Both have conditions, and an advance helps most once the numbers already pay back.