Why does Meta show fewer iOS trials than my billing data? Some were never sent, some never credited.

I am Samet Durgun, a fractional Head of UA. I run paid UA for subscription apps and mobile games and write up what I find in the accounts I manage. This piece sits under what signal engineering fixes; more about me.

Meta will never show you every iOS trial, and for users who declined tracking, it never tells you which ad a trial came from. Some missing trials were never sent to Meta, and that can change. The rest arrived and were not credited, or fell outside the attribution window. A stable gap is normal and workable. Sometimes Meta counts too many. I judge cost on billing counts. Once the count is settled and cost is the problem, the first thing I check is whether each ad set gets enough events. Too few events is not always the cause. A web step, a web page between the ad and the App Store, is supposed to narrow the gap. In my accounts the record is mixed.

AEO is app event optimization. Ads Manager called the goal “maximize number of app events” when I last checked, on 29 September 2026.

When someone tells me trials are missing, I ask first: how many are missing, out of how many, over which dates, and how many trials came before the ads. Then I ask where: in Events Manager, or only on the campaign in Ads Manager.

Scope: Meta app promotion campaigns for iOS that optimize for a trial start, with a subscription platform such as Adapty or RevenueCat, which I call the billing tool below. This is from my own accounts, and the sample is small. Two accounts had exactly this problem with trials. Several more had it with installs, on apps without a trial and on games. Neither of the two was on Adapty. I am an Adapty Berlin ambassador and I cohosted a public webinar with Adapty in August 2026, so weigh every mention of Adapty accordingly.

Is a gap between Meta and my billing tool normal, and can I get the trials back?

Yes, it is normal since App Tracking Transparency (ATT), the iOS prompt that asks users whether an app may track them. I put it to one team this way: you will never get 100% coverage from iOS, and there will always be a miss. My checklist has the other side of that rule: “If you see 0% discrepancy, something is broken.”

The general case, any two tools that disagree, is in which number to trust for which decision.

I have no benchmark for the size of the gap, and I have quoted a different number almost every time I was asked. In a guest post on Adapty’s blog I wrote “Sometimes 50% gaps on certain days.” That was an estimate.

On one subscription app, I once put Meta’s new subscribers next to the billing tool’s. For one week the two counts matched. Over the three weeks around it, they did not match, though at the time I said the gap was not that bad. The app’s offer changed in those weeks, so I cannot say why they differed. How close the counts looked depended on the dates I picked.

I treat a stable gap as a calibration factor you can correct for, and a gap that jumps on a certain date as a sign of something that changed around then. Recent days are still filling in, so I ask for them to be grayed out in reports, not hidden. What paid UA can still measure on iOS after ATT covers Apple’s reporting windows and delays.

In the EU, Apple offers an alternative ATT prompt from iOS 27.2. I have not found a measurement of what it does to opt in rates.

I once told a team not to make closing the gap an obsession, because on iOS it will never close. On that call, Meta counted more than the mobile measurement partner (MMP) on web campaigns. The team agreed but pressed one point, which I accepted: you still need to know which source is off and by roughly how much, or you correct in the wrong direction.

Where can one trial go missing between the paywall and Ads Manager?

I count seven places, and only some have to do with ATT. AEM is Meta’s Aggregated Event Measurement, its way of counting app events from users who declined tracking.

Where it drops What happened Where to look Can it change
1. Never sent RevenueCat sends App Store events through the Conversions API without ATT consent only when “Send events when ATT consent is not authorized” is on. On the older App Events API, its reference page says iOS events “also require ATT authorization”, and Meta no longer recommends that API for new integrations. Adapty’s page says its Facebook Ads integration “works on iOS 14.5+ only for users with ATT consent”, and that with facebookAnonymousId set, “if the user’s IDFA is unavailable, the integration will continue to function.” It does not say what is sent for users who declined Billing trials against Events Manager They can be sent. No page I know says Meta then credits them. The app owner’s privacy decision, with counsel
2. Sent without what AEM needs The event lacks identifiers or the ATT answer, or comes from a different integration than the one that sent the install When Events Manager marks the trial ineligible Yes, a setup fault
3. Received, not credited The user declined tracking, and AEM did not match the trial to an ad Present in Events Manager, absent on the campaign Not for individual users
4. Never clicked The user saw the ad and searched for the app later App Store Search in App Store Connect Only a view window could credit it
5. Outside the attribution window, or not reported yet The trial started after the window closed, or is not in yet. If you optimize for the first payment, a 7 day trial pays on day 7 or 8, about where a 7 day click window stops “Compare attribution settings” in Ads Manager Late events still arrive
6. Not a trial from the campaign The billing tool counts organic trials too Weekly trials from before the campaign Nothing to fix
7. Counted twice or mixed Meta’s number is too high The section on overcounting below Yes

How do you tell which kind of missing you have?

First I look at the blended numbers and the store, before any attribution tool. Blended means all trials and all proceeds, paid and organic together. I read them for one platform in one country, never for the whole account. If iOS billing trials and proceeds held while the attributed numbers fell, the measurement changed and demand did not. App Store Connect is my source of truth for proceeds and for where installs came from.

Second, I check App Store Connect downloads by source. App Referrer includes clicks from Meta ads and links from any other app, so it cannot tell a paid Instagram click from a click on an influencer’s post. App Store Search also includes the people who saw the ad and searched the app name later, next to ordinary search and Apple Ads. My rule for what the reports call organic: if the app had no organic traction before the ads, what shows as organic after them is mostly paid.

Third, I compare three counts for the same dates, platform and country: trial starts in the billing tool, StartTrial events in Events Manager, and trials credited to the campaign in Ads Manager. Events Manager shows organic and paid together, but only for events that are really sent. If Events Manager is short, the trials were not sent, which can change. Present there but missing on the campaign means the trial is organic or not credited.

Invented numbers to show the method:

Line, one week, iOS, one country Trials
Trial starts in the billing tool 150
Organic baseline: weekly trials before the campaign 30
Trials from the ads (150 minus 30) 120
StartTrial events in Events Manager, organic and paid 120
Organic among those 120, if both are sent at the same rate (120 of 150, so 80% of 30) 24
Trials from the ads that reached Events Manager (120 minus 24) 96
Trials credited to the campaign in Ads Manager 72
Trials from the ads, received and not credited (96 minus 72) 24
Trials from the ads, never sent (120 minus 96) 24

The naive reading is that 78 trials are missing, 150 minus 72. Line by line, 30 never belonged to the campaign, 24 were never sent, and 24 arrived and were not credited. I invented the credited share, 72 of 120, and it is not a benchmark.

Fourth, I use Android as the control. If an Android campaign matches the store and the billing tool, the fault is on the iOS side. If Android does not match either, the problem is in how trials are tracked.

Can Meta’s count be too high?

Yes. On one iOS subscription app with a trial, the team wanted to return to a low cost per trial from an earlier period. The app had changed its tracking setup more than once since. In that cheap period, Meta’s trial count was almost equal to all new trials in the billing tool, organic included, and on some days above it. Afterwards the gap went back to normal. I became almost sure it was double counting. My guess was that every billing trial had been forwarded to Meta in those weeks. So the cheap cost per trial was a counting artifact. Meta had never credited trials at that cost on its own.

Many store downloads in that period came from sources other than Meta, and they moved with the campaigns. My read at the time was that the inflated count was close to what the campaigns really brought in from all sources.

I took from it that Meta should normally see fewer trials than the billing tool, because the billing tool includes organic trials. If the two are nearly equal, suspect double counting.

The count can also run too high when two senders send one event. RevenueCat tells you to remove purchase tracking in the app for the events it sends, because the Meta SDK and RevenueCat share no event ID, so Meta may count one purchase twice. Meta’s Conversions API page for app events says it deduplicates server events against the Meta SDK by event ID and event name, and warns that “Inaccurate event IDs may cause your conversion to be wrongly deduplicated.” So two senders count a trial once only when both send the same ID and name. If the IDs cannot match, keep one sender per event. My post on hiring a Conversions API expert covers when two senders can work.

On RevenueCat’s default mapping, Trial Converted, Initial Purchase and Renewal all go to Subscribe, so Subscribe can mix first payments and renewals. Installs can run high too: on one subscription account, campaigns with 1 day view counted 1.20 times the MMP’s installs, on an OS that post does not name.

How do you judge and lower cost when an AEO campaign shows only part of the trials?

The count is a reporting problem. The cost problem, as I told one team, is often that the campaign does not get enough events to optimize itself. A tool that measures better does not by itself buy cheaper trials.

Add $2,400 of spend to the invented example. Ads Manager shows 2,400 divided by 72, or $33.33 per trial. Spend divided by all 150 billing trials is $16.00. Spend divided by the 120 trials above the baseline is $20.00. The blended $16.00 is the floor, and the real cost is never lower. The Ads Manager figure is the high end, as long as Meta is not overcounting. A reliable baseline tells you where between the two the real cost sits. That is my read.

Cost per trial in Ads Manager is spend divided by counted trials, so it moves with the counting, even when nothing real has changed. Beyond cost per trial, I judge on cost per paying subscriber at the same cohort age.

To lower real cost I first look at how many events each ad set gets. An ad set usually exits the learning phase after about 50 results in a week, as I read Meta’s learning phase page on 14 November 2026. My post on how many ads a budget can judge has the full rule. Fewer campaigns help an ad set reach that number. To get there, I would not move to an event that is not tied to paying, such as paywall view, although I have no test to back that up.

For paid traffic I believe selling without a trial often converts better and tracks better. My published case is a before and after, with caveats, and my own app has no trial on iOS.

A qualified trial event counts a trial only if the user does not cancel early. Build one only when your billing export shows that early cancels are substantial. It must still reach about 50 events a week. Check in Events Manager whether it reaches an AEM ad set.

With a weekly trial and a direct subscription side by side, I still do not know a clean way to optimize for both on Meta.

Which setup faults widen the gap, and what narrows it without a web step?

Most of this section comes from install campaigns. On one iOS install campaign, I checked in this order:

  1. Events Manager: app installs arrive, split by iOS and Android.
  2. The ad set’s attribution choice between SKAdNetwork (SKAN) and AEM, as my signal engineering page describes it.
  3. The MMP’s Meta integration settings.
  4. The event mapping.
  5. The SKAN configuration in Events Manager.
  6. The ad set’s eligibility check.
  7. The “actions” column in Ads Manager.
  8. App Store Connect downloads by source type.

The install event must also carry the user’s ATT answer. How you do that depends on your SDK, so ask whoever maintains it to confirm. A high opt in rate is not the goal, but a rate in the low single digits is a symptom to investigate.

Apple’s privacy manifest page says network requests to domains an app declares for tracking “fail” when the user has not granted tracking permission. Trial events sent to such a domain fail for users who declined, and Events Manager comes up short. That is Apple’s documentation, not a case from my accounts.

I do not recommend a default attribution setting here. Open “Compare attribution settings” in Ads Manager, look at view and click conversions separately, and check what your ad set offers.

Between AEM and SKAN, my default is AEM with the SKAN configuration left in place. My understanding is that SKAN helps AEM, and that this comes from Meta.

As I understand it, four things raise the share Meta credits, in this order: correct tracking, volume, identifiers sent after install such as a hashed email, and then a web step. I have no Meta page for this order or for the SKAN point. Apple’s privacy page asks whether an app may use a hashed email or phone number in place of the IDFA to track a user who has not given permission. Its answer is “No”: you need “the user’s permission through the App Tracking Transparency framework to track that user.” So for users who declined tracking, sending a hashed email is a privacy decision for the app owner with counsel.

I took one case to the platform and the MMP, and as far as I know it was never solved.

Does a web step before the store, or Adapty’s attribution, fix it?

People mean two things by web to app: a redirect, where the ad click passes through a web page straight on to the App Store, and a full web funnel with web payment. If you try the web route, start with the redirect. My line on the full funnel is “It’s a cherry on top, not a shortcut.”

I changed my mind on this during 2026. At the start of the year I called web to app the real solution. By August I was saying “you get more accuracy than Meta’s AEM. But it will never get much better than that. There will still be missing parts.” What I saw in between was mixed, and each case used a redirect or a dedicated web to app attribution tool:

  • On one iOS subscription app, Meta’s trial count recovered after the team moved to such a tool. That shows counting, not real cost.
  • On a second subscription app, on iOS, the team found a large share of one day’s proceeds not credited to any campaign. That total also included organic buyers and renewals, so I read it as a warning, not a rate.
  • On a third account, on iOS, I paused two Meta web campaigns with a redirect before I could read them. I had three reasons: no time to test them properly, the click only campaign needed longer, and on the click plus view campaign most revenue came from view attribution, which felt off.
  • On my own app I installed such a tool. I have no result from it to report.

“Adapty attribution” can mean the Facebook Ads integration that sends subscription events to Meta, or Adapty’s own attribution product with its web pixel and link.

Apple’s rule is that “you may not derive data from a device for the purpose of uniquely identifying it.” Adapty’s attribution page says it matches installs to campaigns “using probabilistic signals”. Neither that page nor its web pixel page says how a click is matched to an install. That question applies to every tool of this kind, and I do not assess any of them against Apple’s rule. The app owner decides with privacy counsel.

I have no result of my own from Adapty’s web to app link, have not set up Adapty’s Meta integration and have not built a full web funnel. So I do not recommend or rank any vendor. If you are on AppsFlyer, the debugging steps are in why Meta web to app shows clicks but no installs.

What should you check this week?

  1. Billing trials, StartTrial events and trials on the campaign, for the same dates, platform and country.
  2. Weekly trials from before the ads, as your organic baseline.
  3. Meta’s trial count against all billing trials. If they are close, look for double counting.
  4. The sender of each event, which Meta API it uses, and what your billing tool sends for users who declined tracking.
  5. “Compare attribution settings” for the campaign, with view and click conversions shown separately.
  6. Spend divided by billing trials, next to cost per paying subscriber at the same cohort age.
  7. The gap, written down every week.

On 14 November 2026 I read the vendor, Apple and Meta pages linked above. The example numbers are made up, no account is named, and the estimates are mine, not benchmarks.

Questions people ask

Can I recover the trials Meta does not attribute on iOS?

Not all of them, and for users who declined tracking, Meta will never tell you which ad a trial came from. Trials that were never sent to Meta can be sent. For users who declined tracking that is a privacy decision for the app owner with counsel, and I know of no Meta page that says Meta then credits them to a campaign. For trials that arrived and were not credited, use your billing tool and App Store Connect for the count. Meta is still useful for direction.

Is a gap between Meta and my subscription platform normal after ATT?

Yes. I know of no benchmark for the size of the gap, and I have quoted different numbers myself. A stable gap is a factor you can correct for. A gap that jumps on a specific date points to a setting change, an SDK change or an app version.

Can Meta report more trials than the ads really brought?

Yes. Two senders for one event that do not share an event ID, renewals counted with first payments, or every billing trial forwarded to Meta can all inflate the count. My rule is that Meta should normally show fewer trials than your billing tool, because the billing tool includes organic trials. If the two are nearly equal, suspect double counting.

Does a web to app redirect fix iOS attribution for Meta?

A redirect is supposed to give Meta more data to match a click to an install than its Aggregated Event Measurement (AEM) has on its own. In my accounts the results are mixed, and some conversions are still not credited to any campaign. Whether you may use a tool that matches clicks to installs is a decision for the app owner with privacy counsel.

How do I calculate cost per trial when Meta undercounts?

Divide spend by the trials your billing tool recorded for the same dates, platform and country. That blended figure is the floor. The real cost per trial from ads cannot be lower, because the billing count includes organic trials. Read the Ads Manager figure as the high end, and judge the campaign on cost per paying subscriber at the same cohort age.