Understanding the nuances of Google Analytics 4 (GA4) become even more important with Google’s announcement that Universal Analytics (UA, or GA3) will stop processing new hits on July 1, 2023 (UA 360 property hit processing ends on October 1, 2023). With that timing, to allow for year-over-year comparisons, you’ll want to be recording all of your valuable user actions in a GA4 property by July 1, 2022.

We already wrote about “Attribution of imported GA4 conversions in Google Ads“. To further help you prepare to make GA4 your primary digital property analytics source of truth, below are 10 things we think you should know about how GA4 is different from Universal Analytics. There are other nuances for sure, but this is a top 10 list. 🙂

Dave Letterman
Late Show Top 10 Lists
  1. Event data model. GA4 employs an event-driven data model measuring both app + web with three components:  (i) events (what happened), (ii) event parameters (more info about what happened) and (iii) user properties (who did it).
  2. User engagement. For a mobile app this is the time that the app was in the foreground; for a website this is the time that the browser tab was active.
  3. Engaged sessions. Defined as > 10 seconds of engagement time, or having one or more conversion events, or having two or more views.  Replaces bounce rate from UA.
  4. An active user is someone who has had at least 1 engaged session during the date range that you’ve selected.
  5. In UA, a new campaign will start a new session regardless of activity. In GA4, a new campaign does not begin a new session. 
  6. GA4 processes events which arrive up to 72 hours late. GA4 batches event delivery client-side. Exceptions where events are sent right away include: conversion events, if a user leaves a page, if the the sendBeacon API is not supported. Make sure your reporting date ranges consider this timing.
  7. Anomaly detection. GA4 predicts what should happen on your digital property, and when there is a meaningful variance to that forecast, it surfaces notifications describing related anomalies for a given metric or segment (i.e., a subset of your users) to drive actionable analysis.
  8. Modeled conversions. GA4 uses data modeling to estimate conversions that can not be observed (e.g., where Safari ITP has truncated a conversion window by auto-deleting a first-party cookie). If there isn’t enough traffic to inform the model, modeled conversions are attributed to the “Direct” channel. Note that conversions imported into Google Ads from linked GA4 properties can include modeled conversions.
  9. Predictive metrics. GA4 leverages machine learning in analyzing your recorded data to predict the future behavior of your users. Predictive metrics include: purchase probability, churn probability and predicted revenue. There are prerequisites for these metrics to be populated (e.g., sending purchase events, and enough data consistently available to train the models).
  10. Predictive audiences. A predictive audience is based on at least one predictive metric (e.g., ‘likely 7-day purchasers’). You can use these audiences for targeting in any Google Ads account linked to a GA4 property that has met the requirements and is populating the related audience. Targeting use cases include excluding users that are not likely to engage or purchase in a given timeframe, or including users in targeting that are likely to do so.
Other helpful support articles:
Related: (i) License renewal considerations when transitioning from GA360 UA to GA360 GA4, (ii) Why is Google sunsetting GA Universal Analytics and promoting GA4?
Is your GA4 property ready to replace Universal Analytics?