Facebook’s Q1 2021 earnings, reported earlier this month, defied the worst-case predictions that followed iOS 14’s App Tracking Transparency rollout. Revenue came in at $26.2 billion, up 48 percent year-over-year. The company added 200 million monthly active users globally. The stock rose on the news.
The performance prompted a round of “Facebook is fine” analysis that missed the more interesting and operationally important story: Facebook survived iOS 14 by changing its attribution model in ways that give advertisers less insight into what their campaigns are actually doing, while presenting performance numbers that look normal on the surface. Understanding the difference between what Facebook’s dashboard shows and what is actually measurable underneath is now an essential skill for any media buyer running performance campaigns on the platform.
The Revenue Paradox
Facebook’s strong Q1 performance despite ATT seems counterintuitive until you understand how ATT’s impact distributes across Facebook’s business.
Facebook’s core advertising product — ads served within the Facebook and Instagram apps to Facebook and Instagram users — is a first-party advertising environment. Facebook knows who its users are from their Facebook account login. ATT’s opt-in prompt has required users to explicitly consent to tracking “across other companies’ apps and websites” — but within Facebook’s own app ecosystem, Facebook’s first-party tracking continues. The ATT framework does not prohibit a publisher from tracking user behavior within its own app.
Where ATT dramatically affected Facebook is in measurement: the ability to attribute off-Facebook conversions (website purchases, app installs in third-party apps) to Facebook ad exposures. Without IDFA consent from users who see a Facebook ad and then convert elsewhere, Facebook cannot deterministically close the attribution loop from exposure to conversion.
Facebook’s solution is Aggregated Event Measurement (AEM), and the numbers on your Facebook Ads Manager dashboard since iOS 14.5 have been partially produced by AEM’s modeling, not by deterministic measurement.
How Aggregated Event Measurement Works
Aggregated Event Measurement is Facebook’s post-ATT attribution system for iOS campaigns. It operates in two modes: for users who have consented to ATT tracking, deterministic attribution continues as before. For users who have not consented — currently the majority of iOS users — AEM uses statistical modeling to estimate conversions.
The modeling uses several inputs: aggregate patterns from users who have consented, device characteristics (device type, operating system version, carrier), time-series data, and signals from Facebook’s on-platform measurement of user behavior. From these inputs, Facebook estimates how many conversions the opted-out users likely generated.
The practical result: the conversion numbers you see in Ads Manager for iOS campaigns are a mix of real measured conversions and modeled conversions. Facebook does not currently give advertisers a way to see these two components separately in the standard reporting interface.
This matters because modeled conversions are not measurements — they are estimates with confidence intervals that Facebook does not publish. An advertiser optimizing a campaign based on 1,000 reported iOS conversions might be optimizing on 300 real conversions and 700 modeled estimates. The optimization signal is noisier than it appears, and the direction of any modeling bias is not transparent to advertisers.
SKAdNetwork vs. AEM: What You Should Know
Facebook implemented SKAdNetwork — Apple’s own privacy-preserving attribution framework — for App Install campaigns as required. SKAdNetwork and AEM are not the same system, and they are not additive.
SKAdNetwork provides the deterministic iOS attribution data that Apple permits without user consent: an aggregate notification of installs attributed to a given Facebook campaign, with a conversion value representing post-install events, delivered with a delay. SKAdNetwork data can be seen in Facebook’s reporting at the campaign level.
AEM is the additional layer of statistical modeling that Facebook applies to reconstruct the conversion picture that SKAdNetwork’s limited signal set does not fully capture. The two together produce the numbers Facebook reports as campaign performance.
Advertisers who want to understand their actual measured iOS performance — separate from Facebook’s modeling — need to work with the SKAdNetwork data directly, through their mobile measurement partner’s SKAdNetwork reporting. AppsFlyer, Adjust, and Branch have all published detailed guides to SKAdNetwork reporting that provide a way to see the raw Apple-certified data beneath Facebook’s modeled layer.
What the 48% Growth Doesn’t Show You
Facebook’s revenue growth in Q1 was real, but it was driven significantly by factors that predated ATT’s impact: CPM recovery from COVID-depressed 2020 baselines, vaccine reopening category resurgence (travel, restaurants, retail), and the continuation of DTC brand advertising that had shifted to digital throughout 2020.
The more telling metric for understanding ATT’s impact on Facebook’s measurement quality is not revenue — it is advertiser return on ad spend as measured outside of Facebook’s own reporting. Independent media mix models and incrementality studies from agencies and brands who ran those analyses before and after iOS 14.5 are showing larger discrepancies between Facebook-reported performance and externally measured incrementality than were present in pre-ATT periods.
Put differently: Facebook’s dashboard numbers are higher relative to the actual incremental revenue being driven than they were before ATT changed the measurement baseline. The model is filling in the measurement gap in a way that is optimistic about Facebook’s contribution.
This is not an accusation that Facebook is deliberately inflating numbers. Modeling uncertainty produces numbers that cluster around a mean estimate, and the incentive structure for any advertising platform’s internal measurement model to be optimistic about that platform’s performance is real even without deliberate manipulation. Independent measurement is not optional anymore — it is the only way to calibrate what the modeled numbers actually represent.
What Advertisers Should Actually Do
The practical response to Facebook’s post-ATT measurement situation involves three priorities.
Build external measurement frameworks. Media mix modeling and geo-based incrementality tests provide measurement of Facebook’s actual contribution to business outcomes that does not depend on Facebook’s attribution infrastructure. These are time-consuming and expensive to build but are now essential for any advertiser spending meaningfully on Facebook.
Use the conversion API. Facebook’s Conversions API (CAPI) is a server-side signal that sends conversion data directly from your website or app backend to Facebook’s servers, bypassing browser-level tracking restrictions. CAPI does not solve the fundamental limitation of consent-based attribution, but it significantly reduces measurement loss from ad blockers, browser restrictions, and page-load timing issues. It is required infrastructure for any Facebook direct response program.
Separate iOS and Android campaign reporting. Creating separate ad sets or campaigns by operating system gives you the ability to see the divergence between iOS (modeled) and Android (still deterministic) performance without the aggregation hiding it. If your iOS and Android ROAS numbers diverge significantly after controlling for other variables, that is information about measurement quality, not just platform performance.
FAQ
Is Facebook’s Q1 2021 performance a sign that ATT doesn’t actually matter? No. Facebook’s Q1 performance reflected recovery from COVID-depressed 2020 baselines and strong DTC ad demand rather than ATT immunity. The measurement impact of ATT on Facebook is real and reflected in the modeling that now underlies iOS conversion reporting. Whether ATT ultimately reduces Facebook’s long-term revenue depends on how well its Advantage+ and AI-driven optimization products can compensate for reduced signal — a question that will take multiple quarters to answer definitively.
What is the Conversions API and is it required for Facebook campaigns? The Conversions API (CAPI) is a server-to-server integration that sends conversion signal from your backend systems directly to Facebook, complementing or replacing the browser-based Facebook Pixel. It is not technically required, but in the current measurement environment it is effectively required for any performance campaign that depends on conversion optimization. Without CAPI, you are depending on browser-side Pixel data that is degraded by ITP, ad blockers, and ATT consent loss. Facebook provides implementation documentation and several tag management and e-commerce platform integrations that simplify CAPI deployment.
How do I know how much of my Facebook conversion reporting is modeled vs. measured? Facebook does not currently expose this distinction in standard Ads Manager reporting. The nearest proxy is to run the same campaign periods through your mobile measurement partner’s SKAdNetwork reporting (for app campaigns) or through an independent web analytics comparison (for web conversion campaigns) and compare the totals. The gap between independently measured conversions and Facebook-reported conversions represents the modeled component. Some agencies with advanced measurement practices are using this methodology to calibrate Facebook’s reporting systematically.
Should we reduce Facebook spend because of measurement uncertainty? Not necessarily, but spending should be calibrated to incremental impact rather than Facebook-attributed ROAS. If external measurement — incrementality tests, media mix modeling — demonstrates that Facebook spend is driving incremental business outcomes at acceptable efficiency, maintain or grow it regardless of dashboard ROAS. If external measurement shows that Facebook-attributed performance significantly overstates incremental impact, that is a signal to reallocate. The answer should come from measurement, not from reacting to uncertainty.