Paid UA for subscription apps is judged on payback, not installs, because a campaign can bring 10,000 cheap installs and zero paying subscribers. I judge campaigns on blended ROAS and payback against real revenue, checked against both the ad platform and your MMP, because the two rarely agree.
This page is for a subscription app whose paid channel has stalled, or has never been built properly. It explains what makes subscription different from every other kind of app, what has to be fixed upstream of the campaign, and what happened on two accounts where that was done: Videa’s D0 ROAS from 20% to a 43% record week and EatBetter’s $10K to $180K MRR in two months.
What I judge a subscription campaign on
What CPA ceiling should a subscription app set before raising its ad budget? I set it from your unit economics, not a benchmark: proceeds per paying subscriber after store commission, tax and refunds, less variable costs. On trial optimization I divide cost per trial by the matured trial to paid rate before comparing. I raise spend only while cost per paying subscriber at the same cohort age stays under the ceiling. The full method, including what Meta’s Subscribe event counts is written out.
Each number answers a different question, and only the last one is the verdict.
| Number | What it tells me | How I use it |
|---|---|---|
| CPI and installs | What an install cost | Never the verdict. CPI on its own tells you almost nothing. |
| Cost per trial | What a trial start cost | Never the verdict either. When it falls, the platform may have found cheaper trial starters, and cheaper trial starters convert to paid at a lower rate. |
| CPA ceiling | What a payer may cost | I set it from unit economics first: proceeds after commission and tax, less refunds and variable costs. |
| Cost per paying subscriber | What a payer did cost | The number to watch, compared at the same cohort age. |
| D0 ROAS | Whether a cohort makes sense, fast | Only where most initial revenue lands on install day, not days later behind a trial. On one account, 38 Meta campaigns and $606,462 of spend, day 0 value picked the better of any two campaigns on D28 ROAS 92% of the time, and cost per purchase 80%. Treat it as a direction, not a rule. |
| Blended ROAS and payback | Whether the spend pays back | The verdict, against real revenue. Subscription revenue matures, and a campaign that looks breakeven on day 7 can be clearly profitable by day 28. |
What is different about subscription
The table lists each event in a subscription funnel, whether it reaches the ad platform on its own, and what changes that.
| Event | Comes back on its own? | What changes that |
|---|---|---|
| Trial start | Yes, it comes back | Nothing is needed, and that is the problem. |
| First payment | Only a slice of it after ATT, and days after the install, usually behind a trial | A purchase signal through CAPI, Meta's Conversions API, so Meta optimizes toward payers instead of trial starts. It does not restore attribution for users who declined tracking. |
| Renewal | Only if someone sends it from your server | Server side delivery wired to your subscription backend, so renewals reach the platforms even when the app is closed. |
| Cancellation | Usually not at all | The event layer, where cancellation is one of the canonical events, with a value, a currency and an ID. |
Where the work starts
Left alone, Meta optimizes on half the data and finds you the users most likely to start a trial, which is a different population from the users most likely to pay. That gap is why D0 ROAS can fall while CPI stays flat.
So the work starts upstream of the campaign: a purchase signal the platform can actually optimize on, predicted value in the bid where the payback window is long, and a ROAS framework set by country so budget follows payback, not cheap installs. That layer is its own discipline, and at Videa it ran country by country and took D0 ROAS from 20% to a 43% record week while spend scaled past $200K a month.
Where creative comes in
Once the signal is clean, creative is the lever. EatBetter went from $10K to $180K in monthly recurring revenue in two months by turning influencer content into paid acquisition on Meta, with the paywall tuned alongside it. Blended ROAS went from 70% to 230% in the same window, and a $2M user acquisition financing round from Leus followed in July 2026, underwritten on the account. How I make UGC ads for apps is the production side of that.
The questions that come up
Each of these is a post, because each one deserved the numbers. The short answer is here, and the full one is in the post.
| Question | What I found | Source |
|---|---|---|
| Should I remove the free trial? | Keep it when trial to paid conversion is high; I have seen apps convert 50% from trial to purchase. On one app where trials barely converted, removing the trial for paid Meta traffic took conversion from 7.1% to 12% and cost per result from $58 to $40. A result was a trial or a purchase before the change and only a purchase after it, so $58 and $40 are not the same unit. One app, five weeks, no holdout. | source |
| Meta, RevenueCat or Adjust: which number do I trust? | Each pair answers a different question. Meta against RevenueCat: what paid revenue can be directly proven. Meta against Adjust: how much credit paid should receive. Adjust against RevenueCat: whether the revenue actually arrived. None of them tells you whether paid was incremental. | source |
| Is my iOS organic actually paid Meta traffic? | On one account over 103 days, every 100 extra paid iOS installs came with about 28 extra organic ones, against about 6 on Android. Most of that gap is likely paid installs landing in organic. | source |
Who it is for
- Subscription apps with live revenue and a paid budget around $100K a month, or funded to reach it
- Teams whose Meta spend has plateaued on campaigns optimized for trials
- Apps where the ad platform, RevenueCat and the MMP each report a different number
- Founders who want the person reading the data to be the person running the campaigns
What you get
- Meta, Google App Campaigns, Apple Search Ads and TikTok, run daily
- A purchase signal, and predicted value in the bid, that the platforms can optimize on
- A ROAS framework set by country, so budget follows payback
- Creative testing with a steady pipeline of new concepts
- Weekly revenue, ROAS and payback against target
Questions people ask
Which channel should a subscription app start on?
Meta, almost always, because creative testing moves fastest there. I open Google, Apple Search Ads and TikTok once the funnel holds and the numbers are stable. The channel comparison has real D28 ROAS from one app across all four.
What ROAS should I expect?
The numbers from my own accounts are on the benchmarks page, each linked to its case. I do not quote industry averages because I cannot verify them.
Should I remove the free trial?
It is a test, not a rule, and the table above has what happened on an account that removed it.
How do you handle the gap between Meta, RevenueCat and Adjust?
By reconciling them instead of picking one, because each answers a different question; the table above says which number to trust for which decision.
Sources
- Meta Ads RevenueCat Docs. By default Trial Started goes to Meta as StartTrial, and Trial Converted, Initial Purchase and Renewal all go as Subscribe. RevenueCat does not configure SKAN or AEM. No page date shown. Checked 29 September 2026.
- Reconciling with App Store Financial Reports RevenueCat Docs. Revenue is net of refunds, and proceeds are revenue less RevenueCat’s estimate of store tax and commission. Checked 29 September 2026.
- About Meta’s Aggregated Event Measurement Meta Business Help Center. The protocol Meta uses to measure web and app events from people on iOS 14 and later, with identifiers removed, differential privacy added and data aggregated across users; events sent through the Conversions API may also be processed within its limits. Checked 29 September 2026.
- AdAttributionKit Apple Developer Documentation. Apple’s framework for measuring ads for apps in the App Store and alternative app marketplaces while helping maintain user privacy. Checked 29 September 2026.