Most subscription app accounts on Meta are not failing on creative or targeting. They are failing on the event. The campaign optimizes toward whatever you tell it counts as success, and for an app that choice is harder than it is for a store, because the money arrives days or weeks after the ad did its job.
This is the Meta-only version. Channel choice and budget split across platforms are in paid acquisition for subscription apps. Here we cover the setup that decides whether Meta can learn anything at all.
The short rule
Send trial starts and paid subscriptions to Meta as standard events, from one source or deduplicated across two. Optimize toward the deepest of those events that each ad set can reach about 50 times a week. Run iOS 14 and later campaigns on their own, because they are measured differently. Then check every number against your subscription records before you move budget.
Step 1: get the events to Meta
Meta receives app events from the Facebook SDK inside the app, from a mobile measurement partner, or through the Conversions API for app events, which sends them from your server. The Conversions API uses one endpoint for web, app, and other events, all sent into a dataset connected to your app. For app events you set the action source to app and include the advertiser tracking field, which tells Meta whether the user allowed tracking. Meta lists it as required.
If you send the same event from more than one place, for example the SDK fires StartTrial on the device and your server fires it again when the trial is confirmed, include the same event ID on both so Meta counts it once. Installs work differently: Meta already deduplicates them and attributes only one install in a 90-day window.
Use the standard event names. Meta’s app events reference defines StartTrial as the start of a free trial of a product or service, such as a trial subscription, and Subscribe as the start of a paid subscription. Standard names keep setup and reporting consistent across tools. Send a value and currency with Subscribe. Send an expected value with StartTrial only if you have a defensible number from your own trial-to-paid history.
The server route is worth the effort for one reason. The subscription is decided on your server or by the app store, not on the device. A server event can fire when the payment actually clears, which is the fact you care about.
Step 2: choose the optimization event
You have three realistic choices: install, trial start, or subscribe. Deeper is better in principle, because it shows Meta what a valuable user looks like. The limit is volume. Meta says an ad set usually exits the learning phase after about 50 results in the week after its last significant edit, and that performance during learning is less stable. An event you can only buy 20 times a week keeps the ad set learning.
The table runs one budget through all three choices.
| Optimization event | Cost per event | Events per week | Reaches about 50 a week | Paid subscribers per week | Cost per paid subscriber |
|---|---|---|---|---|---|
| Install | $3.50 | 600 | Yes | 36 | $58.33 |
| Trial start | $14.00 | 150 | Yes | 54 | $38.89 |
| Subscribe | $52.50 | 40 | No | 40 | $52.50 |
At this budget, trial start wins. Install optimization buys the cheapest event and the most expensive subscriber, because Meta finds people who install and then do nothing. Subscribe optimization is the right goal but cannot collect enough results to settle. Triple the budget and Subscribe might reach 120 events a week, at which point it could be the better choice. You only find out by testing it, one change at a time.
Trial design moves this table. A card-required trial and a no-card trial produce very different trial-to-paid rates, which changes which event is worth optimizing. That tradeoff is covered in free trial or paid.
Step 3: know what iOS will and will not tell you
On iOS, an app that shares data with other companies to track people across apps and websites has to ask permission under App Tracking Transparency. Some people decline. For them, Meta offers Aggregated Event Measurement for iOS apps, which measures app events from iOS 14.5 and later users who opted out of tracking, through the Facebook SDK. Meta also lets iOS 14 and later app promotion campaigns use Apple’s SKAdNetwork as the attribution method instead, configured in Events Manager.
SKAdNetwork is where subscription apps feel it. Apple’s documentation sets three conversion windows after first launch: days 0 to 2, days 3 to 7, and days 8 to 35. Postbacks arrive after a random delay, 24 to 48 hours for the first and 24 to 144 hours for the second and third. Only the first postback can carry a fine-grained conversion value, and only when the data tier allows it. Later postbacks carry a coarse value.
Put a trial on that timeline. A user installs and starts a 7-day trial on day 0, so the trial start lands in the first window. The first payment lands on day 7, in the second window, where it can only be reported as a coarse value, a day or more later. A 14-day trial pushes the first payment into the third window. The event you most want Meta to see is the one SKAdNetwork reports least precisely.
Meta’s reporting reflects the split. Its attribution settings page says iOS 14 and later app install campaigns using Meta’s attribution report on 1-day or 7-day click, and campaigns using SKAdNetwork report as “From Apple’s SKAdNetwork.” Meta also notes that app activity custom audiences may get smaller, and suggests broad audiences or targeting expansion. Apple’s SKAdNetwork documentation now points developers to AdAttributionKit as well, so check which framework your measurement setup supports before you build around either.
Step 4: structure campaigns around measurement
Structure follows from the measurement split. Keep it simple enough that each ad set can reach its results.
- Separate iOS 14 and later campaigns from Android. They use different attribution methods and report differently, so mixing them turns every number into an average of two measurement systems.
- One optimization event per campaign. Do not run install, trial, and subscribe ad sets side by side and compare them in Ads Manager. Compare them in your subscription records.
- Few ad sets, broad audiences. Meta says that when you create many ads and ad sets, the system learns less about each one, and it points iOS advertisers toward broad targeting as custom audiences shrink.
- Change one thing at a time and leave it for a week. Significant edits restart learning.
Step 5: creative that sells the product, not the download
App creative on Meta has one job: show what the product does fast enough that the person who installs is the person who will pay.
- Show the app working in the first seconds. A screen recording of the core action tells people more than a logo and a tagline.
- State the price and trial terms in the ad or on the first screen after install. Surprising people at the paywall produces installs and trials that never pay.
- Build vertical versions for Reels and Stories rather than cropping a landscape video.
- Make variations that differ in the problem they open with, not only in color or copy length. Near-duplicates rarely teach you anything new. Creative diversity on Meta goes into why.
Step 6: read results against your subscription records
Ads Manager is one view of what happened. Your subscription records, from App Store Connect, Google Play, or your own billing system, are the account of who paid. Reconcile the two every week, by operating system.
- Pull trial starts and first payments from your subscription records by install week and platform.
- Pull the same weeks from Ads Manager for the campaigns that drove installs.
- Compute the ratio of Meta-reported events to recorded events for each platform. On iOS expect it to move around, because of delays, opt-outs, and coarse values.
- Judge campaigns on cost per paid subscriber from your records, divided across campaigns by Meta’s reported share. Say that assumption out loud when you report it.
- Watch that cost for four to six weeks before calling a winner. A single week on iOS is mostly postback timing.
Once cohorts are a few months old, the question moves from cost per paying subscriber to cost per subscriber who stayed. The same churn arithmetic applies to apps and to physical products. Subscribe-and-save math works it through for a store, and the principle carries over: what a subscriber is worth depends on the churn you actually see.
What most app advice gets wrong
The usual advice is to optimize for the deepest event from day one, because that is what you want. With enough volume, it can be right. On iOS, the deepest event for a trial-based subscription is exactly the one SKAdNetwork reports late and coarsely, so the campaign gets thin, delayed feedback on the thing it is meant to optimize.
The opposite advice, buy cheap installs and let the app convert them, is worse. An install is the event most loosely connected to payment. Meta will find people who install. That is all it was asked to do.
The practical answer sits between them: the deepest event that arrives in time and in volume, usually trial start, checked against records that show whether those trials pay.
When this is wrong
- Hard paywall at install. If people pay on day 0, Subscribe lands in the first SKAdNetwork window and can carry a fine-grained value, which makes it a much better optimization event.
- Large budgets. If an ad set can buy well over 50 subscriptions a week, test Subscribe optimization directly.
- Very short trials. A 3-day trial still puts the first payment on day 3, which is the start of the second window, so the gain is smaller than it looks.
- Web checkout. If you sell the subscription on the web and send people to the app afterward, the purchase is a web event and the iOS app limits do not apply to it. That is a different funnel with its own rules.
What to do this month
Late November is a poor time to rebuild an account and a good time to check the plumbing.
- Confirm StartTrial and Subscribe fire once per user, with matching event IDs if they come from two sources.
- Check that the advertiser tracking field is set on every server event. Log what you find, and fix anything non-urgent after Cyber Monday.
- If a campaign mixes iOS 14 and later with Android, note it and split it in January, unless you can make the change now and let it settle before peak.
- Do not change the optimization event during peak weeks. Write down what you want to test and start it in January, when a restarted learning phase will not collide with your busiest weeks.
- Set up the weekly reconciliation sheet now, so holiday cohorts are readable when they renew.
Common questions
Should a subscription app optimize Meta ads for installs or trials?
Do I need the Conversions API if I already have the Facebook SDK?
What is the difference between AEM and SKAdNetwork for iOS app campaigns?
Why do Meta’s numbers not match my App Store subscriptions?
Should iOS and Android be in the same Meta campaign?
If you want the event setup and campaign structure checked against your subscription records, Meta ads for SaaS covers how we run Meta for subscription products. The CAC payback calculator turns cost per paid subscriber into months to recover it.