Most posts about google analytics 4 benefits read like a feature inventory: 22 capabilities listed flat, ranked by no one. The real question analytics owners have is different. Which features change the way a team measures, and which ones look impressive in a deck and then sit untouched for a year? This piece sorts the 10 that pull their weight, with the limits, eligibility thresholds, and quota math that most lists quietly skip.
Universal Analytics stopped processing data on July 1, 2023 (Google Help). Standard GA4 properties have been the only Google Analytics for nearly three years, so the operational questions have shifted: less “should we migrate,” more “are we using the engagement model, the BigQuery export, and the predictive audiences in a way that improves a single decision.” scandiweb runs GA4 as a Google Premier Partner across 2,100+ eCommerce projects delivered, including a Looker Studio + BigQuery setup that saved one merchant β¬135,000 annually versus a GA360 contract. The benefits below are framed in that operational light.
Overview
- GA4’s event-based model, engagement metric, and BigQuery export are the three features that genuinely change a measurement workflow. The rest are useful but secondary.
- Predictive metrics need 1,000+ qualifying users in the last 28 days to activate, and standard data retention defaults to 2 months unless BigQuery is set up.
- Honest limits matter. Explorations can sample over 10M events, default retention is short, and the API quotas bite fast when reporting scales.
π Quick takeaway
The features worth knowing are the ones that change a workflow. Three do, the engagement model, the event-based schema, and BigQuery export. The other seven are useful in specific cases, and we will say where.
1. Engaged sessions and the engagement rate (the metric that replaced bounce rate)
The engaged session is GA4’s headline behavioral change. A session counts as “engaged” if it lasts at least 10 seconds, fires at least one conversion event, or includes at least two pageviews or screenviews (Google Help). Engagement rate is the percentage of sessions that meet one of those three. Bounce rate, in GA4, is its inverse.
Why this matters: in Universal Analytics, a “bounce” was any single-page session, no matter how long the visitor stayed. An eight-minute read that ended in a tab close was indistinguishable from a misclick. GA4’s engaged session reads that visit correctly. Teams comparing GA4 against archived UA numbers should expect engagement rate to land roughly where 1 minus bounce rate did, but the meaning is closer to what they actually wanted to measure.
For our own blog reporting, the engagement-rate switch cut “false bounces” on long-form articles by roughly a third versus the legacy UA metric, which mostly meant the SEO team stopped chasing a number that was never the right one.
π Quick takeaway
Engaged sessions count visits with at least 10 seconds, one conversion event, or two pageviews. Engagement rate is its percentage. Bounce rate is its inverse, same data, more useful framing.
2. The event-based data model (everything is an event)
GA4 replaced UA’s session-and-pageview model with a flat event model. A pageview is page_view. A purchase is purchase. A button click is an event you name, with the parameters you choose. There is no hit limit per session because there are no sessions in the schema, just session_start events the system reconstructs at query time.
The practical payoff: any interaction the website or app can fire (a 25%-watch video, a quote-form submit, an add-to-cart variant) becomes a first-class entity that can be turned into a conversion, a custom dimension, or a BigQuery row, without coding around a hit-count cap. The old UA 10M-hits-per-property ceiling is gone.
The friction is at the implementation end. A well-instrumented GA4 needs an event-naming convention that survives a year of marketing experiments. Inconsistent event names (“Add to Cart” vs add_to_cart vs AddToCart) are the single most common reason a GA4 audit finds the data unusable.
3. Cross-device and cross-platform tracking (web + app in one property)
GA4 properties unify web (gtag) and app (Firebase) data inside one property. A user who reads on desktop, browses on mobile web, and buys in the iOS app shows up as one journey, provided GA4 has a User-ID or Google Signals to stitch them. Before GA4, this needed manual data joins or a paid GA360 license.
The honest part: identity stitching is only as good as the signals it has. A strong logged-in customer base resolves most journeys via User-ID. If most visitors are anonymous and Consent Mode v2 is denying analytics_storage, GA4 falls back to modeled data, useful, but a model, not a measurement. Treat cross-device numbers as directional unless you can show what fraction of sessions carried a User-ID.
4. BigQuery export (the feature that decides whether a GA4 is mature)
The BigQuery export was the headline payoff of GA4 from day one. Every standard property can stream raw, event-level, unsampled data into a Google Cloud project. Under GA360 this was a five- or six-figure annual line item. In GA4 it sits inside the free tier, subject to BigQuery’s own pricing: 10 GB of active storage and 1 TB of on-demand query data per month at no charge (Google Cloud BigQuery pricing).
The operational truth: for a high-traffic eCommerce property, the 1 TB query allowance fills quickly once a few dashboards run scheduled queries. The merchant case study linked in the intro came in around β¬4,000 a year all-in for the full GA4 + BigQuery + Looker Studio stack, with cost dominated by storage, not query.
For teams hitting GA4’s API rate limits in Looker Studio, our GA4 API quota breakdown covers moving the heavy lifting into BigQuery.
π Quick takeaway
BigQuery’s free tier handles modest reporting needs. For high-traffic merchants, the costs are real but small versus GA360, typically four figures a year, not five or six.
5. Predictive metrics and predictive audiences
GA4 ships three predictive metrics from Google’s machine-learning models: purchase probability (chance a user active in the last 28 days buys in the next 7), churn probability (chance a user active in the last 7 days goes inactive in the next 7), and predicted revenue (expected 28-day revenue from an active user). Each can be turned into a predictive audience and pushed into Google Ads (Google Help).
The eligibility thresholds are what most listicles skip. A predictive metric only activates if, over the prior 28 days, at least 1,000 returning users triggered the relevant predictive event (purchase or in_app_purchase for purchase models, any user-engagement signal for churn), with at least 1,000 users not triggering it. Properties under that traffic threshold see “insufficient data” and nothing else. Most failed predictive audiences are not a setup bug, just a volume floor.
For merchants over the threshold, predictive audiences run hot. Our PPC team has used purchase-probability audiences to weight Google Ads bidding on a fashion merchant’s account and recorded a roughly 30% lift in conversion rate on the predictive segment, with spend reallocated rather than added.
6. Native integrations with Google Ads, Search Console, BigQuery, and Looker Studio
GA4’s native integration set is more usable than UA’s was. Google Ads audiences sync both ways. Search Console surfaces organic queries inside GA4’s Acquisition reports. BigQuery is the export hub. Looker Studio reads GA4 through the dedicated connector or, for any non-trivial dashboard, through BigQuery for speed and quota relief.
The integration most under-used in practice is the Search Console one. Two clicks to enable, and query-level organic data appears alongside the rest of acquisition: useful for any content team historically living in a separate tab for organic. The dashboard we built around this for an apparel merchant is documented in the Looker Studio migration case, which also covers the BigQuery-routed pattern that keeps the GA4 API quota off the critical path.
7. Intelligent tracking with Consent Mode and behavioral modeling
GDPR, ePrivacy, and parallel rules in California, Brazil, and the UK have made cookie-only tracking a dead-end. GA4’s answer is Consent Mode v2 plus behavioral modeling. When a visitor denies analytics_storage, GA4 still records anonymous, modeled signals (no client_id, no cross-session join) and uses machine-learning to estimate the behavior of the consented population on top.
What this means in practice: a GA4 property with Consent Mode v2 set up and modeling enabled does not see the hard drop in measured conversions that UA properties hit the day a cookie banner went live. The modeled data is not perfect, but it closes the difference between a 20-30% measurement gap and an indefensible one. For merchants in the EU, getting Consent Mode v2 right is now table stakes. The CMP and signals need correct implementation before modeling can do its job.
8. Free advanced analysis β explorations
The exploration suite (path, funnel, segment overlap, cohort, user lifetime, free-form) is GA4’s answer to what was previously GA360-only Analysis Hub. Every standard property can build ad-hoc funnels, see the most common paths between any two events, or compare segment overlaps without a paid tier.
The exploration that pays for itself first is the funnel. A 4- or 5-step funnel from product view to purchase, with drop-off rates per step, tells a merchant where the leak is in 30 seconds. Path exploration, in contrast, is interesting to look at and rarely changes a decision unless used surgically.
The quirk worth flagging: explorations sample for very high event counts. A property with more than ~10M events in the chosen date range will show a yellow shield icon and a sampled result. Standard reports, by contrast, are unsampled. If the number matters for a budget call, run it in standard reports or BigQuery.
9. Retention reports and user lifetime
GA4 ships first-class retention and cohort reports out of the box. The retention report shows how returning users behave over the 42 days following acquisition. The user-lifetime exploration shows LTV and engagement curves by acquisition source, campaign, or audience.
These are the reports most marketing teams forget exist. A weekly look at retention by acquisition channel often separates the channels that buy revenue from the channels that buy customers, and the gap is often counter-intuitive. Paid social often ranks high on first purchase and low on 30-day retention. Organic search and email tend to do the opposite.
π Quick takeaway
Look at the retention report by acquisition channel monthly. It will rank your channels differently than your last-click report does, and the gap is often where the budget reallocation lives.
10. Data-retention controls plus a privacy-first default
GA4 lets the admin set event data retention to 2 months or 14 months in the property settings (Google Help). The default is 2 months. After the window expires, event-level data is purged from the interface (aggregated counts remain, but no event-level explorations against expired data).
The 2-month default is the single thing that catches most teams off-guard in year one. A team running a quarterly review in month four cannot pull event-level audiences against month one’s data unless retention was raised to 14 months on day one, and 14 months is the maximum GA4 offers. For anything beyond that, BigQuery export is the only durable answer, which is why BigQuery sits at #4 on this list rather than as a nice-to-have.
For a deeper read on the pattern, see how to overcome GA4 API limits.
Where GA4 still bites: the honest limitations
GA4 is the only Google Analytics now, but it is not free of friction. The five things teams should plan around:
- API quotas hit faster than expected. The GA4 Data API’s per-property and per-project quotas (concurrent requests, tokens per hour, tokens per day) can be exhausted by a moderately complex Looker Studio dashboard. Standard fix: push heavy queries into BigQuery.
- Explorations sample on large date ranges. Over ~10M events in the window, the result is sampled. Standard reports and BigQuery are not. For budget decisions, do not pull from a sampled exploration.
- Data retention is short by default. Two months standard, 14 months maximum in the UI. Anything beyond that needs BigQuery export from day one.
- Identity stitching is signal-dependent. Without User-ID or Google Signals, cross-device journeys are modeled. The model is reasonable but it is a model.
- The interface still has rough edges. Some reports load slower than UA equivalents, the custom-report builder is more capable but less intuitive, and teams new to GA4 typically need 4-6 weeks before the muscle memory catches up.
π Quick takeaway
The two friction points that catch most teams, 2-month default retention and the API quota, have the same fix. Set up BigQuery export early.
What is the biggest benefit of Google Analytics 4 versus Universal Analytics?
The single biggest benefit is the unsampled, event-level data warehouse GA4 makes accessible through BigQuery on the free tier. Under UA, raw event-level data was a GA360 feature with a five- or six-figure annual price tag. In GA4, every standard property streams it for free, subject to BigQuery’s own usage pricing. Engagement rate, predictive audiences, and cross-platform tracking all flow from that change.
How long does GA4 store data, and why does it matter?
GA4 stores event-level data for 2 months by default and up to 14 months if the admin raises the retention setting in property administration. After the window, data is purged from the GA4 interface, though aggregated counts in standard reports remain. For year-over-year, cohort, or customer-lifetime analysis past 14 months, BigQuery export from day one is the only durable answer.
Is Google Analytics 4 free, and where do the costs actually come from?
GA4 itself is free under the standard usage limits. Costs appear in two places: BigQuery (storage and query, beyond the 10 GB and 1 TB monthly free tiers), and the human time to instrument events well. Most eCommerce merchants land between β¬1,500 and β¬6,000 a year all-in for the GA4 + BigQuery + Looker Studio stack, a small fraction of what GA360 cost for the same data depth.
FAQ
Is GA4 replacing Universal Analytics?
GA4 replaced Universal Analytics on July 1, 2023, when standard UA properties stopped processing data. GA360 UA properties followed on July 1, 2024. Existing UA reports are no longer accessible in the Google Analytics interface, so GA4 is the only Google Analytics for any merchant running it today.
What is the difference between Google Analytics 4 and Universal Analytics?
GA4 uses an event-based data model and unifies web and app data in one property, where UA used a session-and-pageview model and tracked them separately. GA4 also ships free BigQuery export, predictive metrics, and a privacy-first identity model, none of which UA offered on standard properties.
How many users does a property need to activate predictive audiences in GA4?
Predictive metrics activate once a property has at least 1,000 returning users who triggered the relevant predictive event (and 1,000 who did not) in the prior 28 days. Properties under that volume see “insufficient data.” This is the most common reason small-merchant predictive audiences fail to populate.
Can GA4 be used without cookies?
GA4 with Consent Mode v2 records modeled, anonymous signals when a visitor denies analytics_storage. The data is modeled rather than measured, but the modeling closes most of the measurement gap that cookie-banner deployments otherwise create. The CMP setup needs to be correct before modeling can do its job.
What is an engaged session in GA4?
An engaged session lasts at least 10 seconds, fires at least one conversion event, or includes at least two pageviews or screenviews. Engagement rate is the share of sessions meeting one of those three, and is GA4’s replacement for UA’s bounce rate.
Does GA4 sample data?
Standard reports are not sampled. Explorations are sampled once a date range covers more than ~10 million events. BigQuery export is never sampled. For a budget or campaign decision, run the number from standard reports or BigQuery, not from a sampled exploration.
How do I avoid hitting GA4 API quota limits in Looker Studio?
Route heavy reporting through BigQuery rather than the GA4 Data API directly. Looker Studio reads from the BigQuery export, which has no per-property GA4 quota and runs faster.
Sorting which GA4 features actually move a measurement workflow is a different exercise from listing them. If you want a clear-eyed read on which features are pulling weight in your property today, and which ones look impressive in a dashboard and then sit untouched, book a GA4 audit with the scandiweb analytics team.
Share on: