GA4 Is Here, and the Deadline Is Unwritten but Real

The beta we profiled last November under its deliberately modest name has graduated. Google announced this month that App + Web is now Google Analytics 4, a proper name with a version number. Attached is the detail that does the real talking: GA4 is now the default property type for new Google Analytics setups.

Universal Analytics, the platform this readership has run for eight years, whose reports every stakeholder in your building reads by muscle memory, continues operating. Not deprecated, not sunset, officially unbothered.

Last November we staked our reading in writing. The beta was the successor in public testing. The event model was the future of the platform. The sunset conversation would arrive on a date already implicit in the architecture. This month graduated the reading along with the product, and the tea leaves now require no special skill. Platforms do not demote their flagship to legacy-in-waiting without a sunset in the drawer. No date is announced. One is coming. This post is the strategy Google’s announcement omits, the mapping work that separates useful migrations from flattering ones, and the calendar argument for doing the cheap part now, while it is cheap.

What graduated, in one honest paragraph

For readers joining at the name change: GA4 is not Universal Analytics with a new interface. It is a different measurement instrument, and the differences are architectural, not cosmetic. The event model replaces sessions-and-pageviews as the atomic grammar. Every interaction is an event with parameters, and the session is demoted to a derived construct. Identity is designed cross-platform and consent-aware from the start. The reports reshuffle around free-form digging rather than the familiar report tree, and they will throw off every UA muscle memory in your building. That is a training line item, not a flaw. And machine-learned modeling is native, built for the partially observed data streams the privacy era guarantees.

The consequence that governs all migration planning: the two models do not reconcile, by design. Sessions differ by definition. Users differ. Conversions differ. A GA4 number and a UA number for the same day are answers to different questions, and no configuration effort will make them match. That fact is not a defect to engineer around. It is the reason the strategy section says what it says.

The strategy the announcement omits

• Dual-tag immediately, and mean this month: a GA4 property alongside production UA costs nothing. The tag layer feeds both. It starts accruing history in the new model’s units. Because the models never reconcile, the parallel runway is the only year-over-year comparison you will ever have when the transition completes. Every month of delay is a month of comparative history that simply never exists. This is the whole argument, and it is math, not opinion.

• Map UA goals to GA4 events deliberately: the auto-migration tooling flatters nobody. The event taxonomy homework is the real substance of the move. Decide what actions matter. Name them the same way each time. Give them useful parameters. The organizations that did it are configuring. The ones that didn’t are re-deriving their measurement vocabulary under deadline pressure, later, at consultancy rates.

• Train the readers before the reports move: the stakeholders who read dashboards deserve the one-page brief now. New model, numbers will differ, here’s why, here’s what we’ll compare instead. The discrepancy explained in advance is methodology. Explained afterward, it is an argument.

• Keep production on UA until the software earns the move: GA4 is young, and feature gaps against mature UA setups are real. The migration will be readier next year than this one. The advice is parallel running, not cutover. Dual-tag is the whole assignment. It is also, conveniently, the cheap part.

The calendar argument, with evidence

The pattern we keep writing down because it keeps costing readers money: measurement transitions are announced later than they begin and executed later than they should be. The Universal Analytics migration of 2012 ran the same course. Feature-framed launch, quiet default status, eventual sunset, deadline scramble. The organizations that ran parallel early spent the transition comparing clean histories while the rest rebuilt baselines from incompatible exports.

The November reading is now confirmed through its second checkpoint. Beta, then default. The drawer’s remaining contents are a sunset date and a migration deadline.

The economics of acting now versus later are asymmetric in the usual way. Dual-tagging this quarter costs an afternoon. It buys a side-by-side history you can never get later. Deferring costs nothing visible now. On the announcement day the tea leaves guarantee, it becomes a scramble with a consulting invoice attached. It runs against whatever deadline the platform sets, with a gap in the history that no budget can buy back. Few will act on this paragraph. The replatforming record says so with statistical confidence, and the consultancy market thanks them in advance. But readers here have a decade of evidence about what believing the roadmap early is worth, and this is the same trade in analytics clothing. The announcements are calendars. The default status is an announcement. The afternoon is available this week.

Our expectations, staked plainly

To check later: we expect a formal Universal Analytics sunset announcement within roughly two years. We expect the announced runway to feel generous and prove tight. They always do, because measurement migrations expand to fill their timelines with taxonomy debates. And we expect the organizations dual-tagged from this season to experience the eventual cutover as configuration while the rest experience it as loss.

The deeper continuity, for readers tracking measurement across its decade. GA4 is the session model’s obituary formally filed. It is the modeled-measurement era’s official opening. It is the privacy-era design arriving as the default tool. Consent-scoped collection. Seeing only part of the traffic. Inference in place of counting. Exactly as predicted the tools eventually must.

The era’s habits transfer intact. Naming before tools. Annotate always. Parallel history as insurance. Skepticism toward any number whose derivation you cannot explain to a stakeholder. New instrument, old virtues. Dual-tag this month, and let the drawer open on Google’s schedule while yours is already kept.