Every app that scales profitably takes tracking seriously. And in 2026 the hard part isn’t installing an SDK. It’s knowing which of your three conflicting numbers to believe.
We run paid campaigns for subscription apps across Meta, Google and other channels, and measurement setup is where we start every engagement, before touching a single campaign. So here is the full picture: what to install, what has changed since the last time you read a tracking guide, and how to actually know if your ads work.
The 2026 stack at a glance
For a subscription app advertising on Meta and Google:
- Firebase SDK for Google campaigns (the default route; a supported MMP integration also works, and Android installs track through Google Play), plus Consent Mode v2 signals if you have EEA users.
- An MMP, once you’re on 2+ ad networks (more below on whether you need one at all).
- Meta SDK + Conversions API feeding one Dataset. Pixel-only setups are below baseline now.
- Your subscription platform (RevenueCat or similar) as the source of truth, sending events out to everything else.
- SKAdNetwork and AdAttributionKit both configured on iOS, since we’re mid-transition.
- Periodic incrementality checks to keep all of the above honest.
If that list looks long: the good news is most of it is set-once. The expensive mistakes are almost always in what you optimize toward, not in which tools you bought.
Do you even need an MMP?
Honest answer: not on day one. If you’re running Google only, Firebase covers you. Meta only? The SDK plus CAPI does.
The moment you run two or more networks, an MMP earns its cost. Otherwise every platform claims the same conversions and your blended numbers are fiction.
Who’s who in 2026:
- AppsFlyer: the enterprise default with the widest integrations. Fun fact: Google, Meta, Moloco and Unity all bought minority stakes in it in 2026. So the “independent” measurement layer is now part-owned by the platforms it measures. Still the safest big-company choice.
- Adjust: automation and CTV strength (still owned by AppLovin).
- Branch: deep linking and web-to-app journeys. A natural fit if your funnel starts on the web.
- Singular: cost aggregation and analytics, typically the cheapest per conversion.
- Airbridge: positioning itself hard as the subscription-app MMP (install, trial, subscription and renewal in one view). Worth a look if that’s you.
Most MMPs have free tiers (AppsFlyer’s covers roughly your first 10-12k conversions in year one, Adjust around 1,500 attributions per month). Check the current terms, they shift.
iOS: SKAN is legacy, and no, you can’t fingerprint your way out
Three things every app marketer should know about iOS measurement in 2026:
1. SKAdNetwork is the past, AdAttributionKit is the future, and you need both today. Apple never shipped SKAN 5. Its successor is AdAttributionKit (AAK), which adds re-engagement attribution and configurable windows. But network adoption is staggered, so the practical setup is both schemas configured in your MMP. A remarkable stat from a mid-2026 survey: a third of iOS marketers never implemented SKAN at all, and only around 11% call themselves fully implemented. The bar for doing this properly is low. Clearing it is an edge.
2. ATT opt-ins plateaued. Depending on whose data you trust, global opt-in sits somewhere between 25% and 38%, flat or slightly rising, with big category spreads (gaming lowest, finance and utility highest). A well-designed pre-prompt priming screen is still the one lever that reliably moves it. The prompt itself is also changing in the EU, after Apple agreed to neutral wording and parity rules in a 2026 settlement with German regulators.
3. Don’t build your own stack on fingerprinting. Apple’s rules prohibit user-level fingerprinting for attribution, and the privacy manifest requirements (enforced since May 2024) make it hard for SDKs to do quietly. The honest nuance: the platforms’ own modeled systems (Meta’s AEM, Google’s newer modeled attribution) do some probabilistic-style matching in a gray zone that Apple has so far tolerated. Big platforms can carry that risk. Your own stack shouldn’t depend on it.
Android: the plot twist is that nothing changed
For years the industry braced for Android’s own privacy reckoning: Privacy Sandbox, the death of the GAID, an “Android SKAN moment”. It never came. Google killed the entire Privacy Sandbox program in October 2025, and GAID deprecation is off the table.
So Android attribution in 2026 works like it always did: device ID plus Play Install Referrer, essentially deterministic. This makes the iOS/Android measurement asymmetry permanent. Android is your high-signal platform, your testing ground, and usually your more honest dataset. Plan around that instead of waiting for it to change.
The subscription-app pattern: your backend is the truth
Here is the setup we consider the 2026 standard for subscription apps, and it flips the old thinking. Instead of asking ad platforms what happened, your subscription platform (RevenueCat or similar) is the source of truth. It’s the only system that actually knows about trials, conversions, renewals and refunds. From there, events fan out via webhooks: to Meta’s Conversions API, to Google, to your MMP and to your warehouse, all under shared event IDs so deduplication works.
One thing to be clear about: the backend is the source of truth for what happened (trials, conversions, renewals, refunds), not for which channel caused it. It gives you blended results without attribution. That’s exactly what makes it useful for marketing mix modeling and geo-lift studies later. For channel-level attribution you still rely on the platforms and your MMP.
On the Meta side specifically: Pixels became “Datasets” (one ID aggregating browser, server and app events), install dedup is automatic, and setup is genuinely easy now. There is no excuse for a Pixel-only or SDK-only setup anymore.
Two rules that decide more performance than any tool choice:
- Optimize toward an event that fires on day 0 and at least 10 times per day per campaign. If your true north star doesn’t clear that bar, engineer a proxy that does.
- Event mapping is a performance lever, not plumbing. A trial start mapped as a “purchase” competes in the auction against $200 e-commerce orders and loses. Map events to what they actually are, so the algorithm competes in the right auctions.
Consent: the silent killer of your conversion data
If you have EEA users, this section alone is worth the read. Google’s Consent Mode v2 has been enforced for apps since July 2025. Without proper consent signals flowing through the Firebase SDK plus a consent platform, Google silently drops personalization, remarketing and conversion tracking. There are documented cases of apps losing 90% of reported conversions overnight, with campaigns still spending. Nothing errors. The data just disappears.
Also worth knowing: the DMA’s first fines landed on Apple and Meta in 2025. And in the US, twelve states now require honoring universal opt-out signals, with Washington treating health-adjacent app data (fitness, wellness, anything symptom-shaped) as opt-in only. If you run a health or wellness app, check with legal before sending user events to ad platforms. The rules there are stricter than most marketers assume.
Web2app: the measurement cheat code
One reason web-first funnels took over subscription-app acquisition: in some cases, web measurement is simply better. A pixel plus server-side purchase events gives ad platforms deterministic, full-value signal. No ATT prompt, no SKAN thresholds, no modeled delays. A meaningful share of web2app’s famous 2x conversion advantage is really an attribution advantage.
Related and newer: since 2025, US-storefront apps can link out to external web checkout. Apple’s commission on those purchases is still being settled in court, so check the current state before building on it. Web checkout also means web attribution on your purchase events, on top of the fee math.
Stop looking for the one true number
The biggest mindset shift in measurement isn’t a tool. It’s accepting that no single report is the truth. Platform dashboards are incomplete in different ways: on iOS, SKAN numbers are the worst-case scenario, while Meta’s AEM usually gets quite close to reality (its blind spot is view-through, which it can’t attribute at all). MMP last-click over-credits clicks. Your backend knows revenue but not causes.
The 2026 answer is triangulation:
- Platform-reported numbers for in-flight optimization (note: Meta’s newer incrementality-based attribution covers web campaigns, not app campaigns, for now).
- MMP plus warehouse blended view for cross-channel comparison.
- Geo-lift tests periodically, to calibrate both. This is the most signal-loss-resistant method available, and open-source tooling made it cheap.
- Lightweight MMM once you’re spending across 3+ channels. Before that it’s overkill.
- App Store Connect as a cheap extra check we use often: the app referrer and web referrer breakdowns show where installs actually come from, straight from Apple. No SDK required.
And if you’re not ready for any of that, the scrappy version still works: baseline your installs and trials per geo for a week, make one significant budget change, watch the trend for several days, conclude, repeat. It has obvious limitations. But it catches the big lies, and the big lies are the expensive ones.
FAQ
Do I need an MMP in 2026?
Not until you’re on 2+ ad networks. After that, usually yes. Pick by fit: enterprise (AppsFlyer), subscription focus (Airbridge), web-to-app (Branch), cost analytics (Singular).
Firebase or an MMP?
Not either/or. Firebase is required for Google campaigns regardless, and the MMP sits alongside it for cross-network attribution.
Is SKAdNetwork dead?
Legacy, not dead. AdAttributionKit is Apple’s successor, but adoption is staggered. Configure both and let your MMP handle the transition.
What should a subscription app optimize campaigns for?
An event that correlates with revenue, fires on day 0, and reaches 10+ events per day per campaign. Trial starts early on, paid conversions once volume and trial-to-paid rates support it. Never optimize toward a number your backend can’t verify.
What happened to Privacy Sandbox on Android?
Google killed it (October 2025). GAID lives on, and Android attribution is unchanged. It remains the most reliable data you have.
Measurement setup is step zero of every engagement we run, before campaigns and before creative. If you suspect your numbers are lying to you, we do a small number of free tracking audits: markus@ppc.ee.