The Analytics Beta That's Really a Roadmap: Reading App + Web

Google Analytics shipped a new property type at the end of July with a name calibrated to sound like a niche feature: ‘App + Web,’ nominally for teams that need apps and websites measured in one place. We read the documentation the way we read all documentation, for the architecture underneath the framing, and found something considerably larger under the modest name.

Measurement rebuilt on an event model: every interaction an event with parameters, the pageview demoted from atomic unit to just another event. Cross-platform identity baked in as a starting assumption. Machine-learned analysis built in as a native layer rather than a bolt-on.

Hold that design against Universal Analytics, the session-and-pageview model most readers have run since the 2012-era migration. The conclusion is hard to avoid, and worth stating before the trade coverage catches up. This is not a feature. This is Universal Analytics’ successor, in beta, being tested in public under a name that defers the announcement.

No sunset has been declared, no timeline exists, and this post predicts none. What it does instead: read the roadmap honestly, name the tradeoffs early, and separate the homework worth doing now from the migration worth refusing until the software earns it.

Why measurement is being rebuilt, and why now

The session model’s obituary has been drafting itself for two years, and the beta is best understood as Google reading the same evidence. Universal Analytics assumes a world that is ending. Durable cookies: the ITP era and the consent gap mean identification decays on a growing share of traffic. Browser-centric sessions: the session idea strains against apps, single-page sites, and people who switch devices. Complete data: the model reports what it saw. It sees less each year, which turns ‘the numbers’ into ‘the numbers among the seen.’

An event model flexes where the hit-type model breaks. Events with parameters describe app and web interactions in one grammar. Identity resolution can work across platforms and degrade gracefully. Most telling for the roadmap reading: a partially observed event stream is exactly the input that machine-learned modeling is designed to complete. The missing-data future is the design assumption, not the failure case.

The tradeoffs are equally legible in the beta and deserve early naming. You buy modeling power by giving up familiar reports. The session-based reports a decade of stakeholders learned to read do not map cleanly onto events. And metrics across the two models will never reconcile perfectly, because they measure with different rulers. Teams that absorb that now will spend the transition explaining their method. Teams that don’t will spend it defending gaps.

The advice, sorted by timeline

• Now, and free: design your event taxonomy. The homework that carries over is about ideas, not code. What actions actually matter on your sites? The applications started, the appointments requested, the donations completed. And the content habits that predict them. Name them the same way each time. Structure them with parameters. Write them down as the measurement vocabulary your team means. This taxonomy outlives any platform timeline. It improves your current setup right away. It becomes the base of the migration whenever the migration comes.

• Now, if curious: dual-tag a property. Run the beta alongside production, a parallel App + Web property fed by the same tag layer. It costs little. It builds team fluency with the event grammar on live data. It starts the history baseline in the new ruler’s units. Your future self will be glad during the metrics-never-match talks.

• Not now, on principle: production migrations from a beta. Production measurement moves when the successor is stable, complete for your needs, and announced as the successor. Betas are for evaluation. Measurement continuity is precisely the asset one does not gamble on a roadmap reading, including this one.

• Throughout: mind the consent layer. Whatever the platform timeline, your measurement now lives inside the consent plumbing, the GDPR and CCPA layer. Design the taxonomy to assume consent-scoped collection from the start rather than bolting it on later.

Field notes from our own dual-tagging

Disclosure by demonstration. We have been running App + Web properties alongside production analytics on our own site and two volunteer client sites since September. The early field notes are worth a paragraph for teams considering the same.

The tagging layer cooperates easily. The same tag-manager container feeds both, which is the whole point of having governed your tags. The taxonomy exercise proved immediately valuable in exactly the predicted way. Writing down what an ‘engagement’ actually means for an admissions funnel exposed three different informal definitions across the marketing team. Reconciling them improved the current reporting before the beta contributed anything.

The new interface’s learning curve is real. The session-report muscle memory reaches for screens that are not there. The debug tooling is decent, and the modeled-insight features are too young to grade. Nothing in the experience changes the advice: dual-tag if curious. Migrate nothing. Treat the exercise as taxonomy practice with a free history baseline attached. The teams that find this boring are correct. Boring is what preparation is supposed to feel like.

Why measurement changes arrive as migrations

The genre observation that earns the title: measurement changes are announced as features and arrive as migrations. The pattern’s record runs long. The Universal Analytics transition itself arrived as an upgrade and functioned as a re-instrumentation. The protocol shifts. The times sessions got redefined.

The mechanism is structural. Measurement platforms cannot announce successors early without freezing their user base. So succession happens in public under feature names, legible to anyone reading architecture instead of headlines, deniable until the timeline is ready.

The reading this post commits to: App + Web is the successor. The event model is the future of this platform. The sunset conversation arrives on some future date already implicit in this beta. The teams that will experience that date as a settings change rather than a crisis are the ones doing the free homework now. Taxonomy first, fluency second, migration on the software’s earned schedule, skepticism throughout. The beta’s name says App + Web. The architecture says ‘next.’ Read the design. Calendar nothing. Prepare everything. Let the announcement find you annotated, dual-tagged, and unhurried, whenever it comes.