July 2023 - July 2025: Two Years of GA4
On July 1, 2023, Google shut down Universal Analytics. For many companies, it was a shock: migrations were not ready, teams were not trained, and familiar reports vanished overnight.
Two years later, where do we stand? What has actually changed — for better and for worse? And most importantly, are the mistakes made during the rushed 2023 migration still corrupting your data today?
Spoiler: in 81% of cases, the answer is yes.
What Improved with GA4
Let’s be fair: GA4 brings real advances over Universal Analytics. After two years of intensive use, here is what works.
The Event-Based Model Is More Flexible
Universal Analytics was built around pageviews and sessions. GA4 is built around events. Everything is an event: a pageview, a click, a scroll, a purchase, a download.
This flexibility enables measurement of interactions that UA made artificially complicated: video clicks, multi-step form interactions, progressive scrolls, micro-conversions.
In practice: companies that properly restructured their events have a far more granular view of user journeys than they ever had with UA.
Cross-Platform Tracking
GA4 natively unifies web and mobile apps in a single property. With UA, you needed a web property AND a Firebase property, with manual reconciliation.
For companies with a web + app ecosystem, this is a considerable gain in simplicity and data accuracy.
Free BigQuery Export
This may be the most underestimated advancement. With UA, BigQuery export was reserved for GA360 accounts (starting at $150,000/year). With GA4, everyone gets it for free.
This unlocks:
- SQL analysis on your raw data
- Custom attribution models
- Looker Studio dashboards connected to BigQuery
- Machine learning on your navigation data
- Unlimited data retention (vs 14-month max in the GA4 interface)
In practice: companies leveraging BigQuery have a major analytical advantage. Those who have not activated it are missing GA4’s biggest benefit.
Predictive Audiences
GA4 offers machine learning-based audiences: “Likely 7-day purchasers,” “Users likely to churn.” These audiences, exportable to Google Ads, enable smarter targeting.
Limitation: you need a minimum data volume (1,000 positive and 1,000 negative events over 28 days) for the models to work. Many sites do not reach these thresholds.
Native Consent Mode Integration
GA4 integrates Consent Mode v2, which allows modeling conversions from non-consenting users without violating privacy regulations. This is a major lever for recovering data lost to consent refusals.
What Remains Painful
The Interface Learning Curve
Two years in, the GA4 interface continues to confuse many users. Standard reports are less intuitive than UA’s. Navigation is different. Terminology has changed (“Bounce rate” came back, but redefined. “Active Users” replaced “Users”).
Concrete result: in our audits, we find that 60% of marketing teams use only 10-15% of GA4’s capabilities — often the same basic reports, ignoring Explorations, segments, and funnel analysis.
Sampling on Large Datasets
GA4 (free version) samples data when queries cover large volumes or long time periods. Specifically, if you have more than 500,000 events in the analyzed period, GA4 may work on a statistical subset.
For large e-commerce sites or media publishers, this is a recurring problem that distorts detailed analysis. The solution: use BigQuery for precise analyses.
Loss of Historical Data
This is irreversible: Universal Analytics data was deleted in July 2024. If you did not export your historical data before that date, it is permanently lost.
The consequences:
- Impossible to compare current trends with pre-2023 years
- Internal benchmarks must start from scratch
- Seasonal analyses have lost their historical depth
Some Reports Are Still Less Intuitive
A few examples of UA features that were more accessible:
| Feature | In UA | In GA4 |
|---|---|---|
| Behavior flow | Clear native report | ”Path exploration” (more complex) |
| Exit pages | Dedicated report | Dimension to add manually |
| Internal search | Native report | Manual event configuration |
| Segment comparison | Easy and visual | Explorations required |
| Multi-channel attribution | Dedicated report | ”Advertising” report (less complete) |
The 81% Reality: Persistent Migration Mistakes
Here is the most critical point. The majority of companies performed a “quick” migration in 2023 — often under pressure, often without dedicated expertise. The result: 81% of GA4 configurations we audit contain significant errors.
Mistake 1: Not Restructuring Events
Many simply activated GA4 with default enhanced measurement, without creating custom events adapted to their business. Result: generic tracking that does not measure what matters for the business.
An e-commerce site without properly structured view_item_list, select_item, add_to_cart events following the GA4 spec loses all funnel analysis capability.
Mistake 2: Copying UA Goals As-Is
UA “Goals” and GA4 “Conversions” (now “Key Events”) do not work the same way. Copying a destination URL as a goal without rethinking the event logic creates erroneous conversion data.
Mistake 3: Ignoring Consent Mode
In 2023, many migrated to GA4 without implementing Consent Mode. Two years later, consent refusal rates have increased (30-50% depending on sector), and companies without Consent Mode are losing a growing share of their data.
Mistake 4: Not Training Teams
The technical migration was done, but nobody trained the marketing teams to use GA4. Result: teams continue searching for UA reports that no longer exist, do not use Explorations, and make decisions on misinterpreted data.
Mistake 5: Not Validating Data Accuracy
How many companies verified that their GA4 conversions match actual sales? Very few. We regularly find discrepancies of 15-40% between GA4 conversions and CRM/ERP data — a sign of broken or incomplete tracking.
What to Do Now
If you recognize yourself in one or more of these mistakes, here is the path forward:
1. Conduct a Complete GA4 Audit
Not a quick check — a structured audit covering:
- Collection quality (events, parameters, Data Layer)
- Conversion configuration (Key Events)
- Filters and exclusions
- Consent Mode and compliance
- Data accuracy (GA4 vs source-of-truth comparison)
2. Clean Up Event Structure
Revisit your event naming convention. Follow Google’s naming convention (snake_case, standardized parameters). Remove unnecessary events. Add missing ones.
Essential e-commerce events:
- view_item_list (with item_list_name, items[])
- select_item (with item_list_name, items[])
- view_item (with items[], value, currency)
- add_to_cart (with items[], value, currency)
- begin_checkout (with items[], value, currency)
- add_payment_info (with payment_type)
- purchase (with transaction_id, value, currency, items[])3. Train Your Teams
Invest in GA4 training tailored to your teams. Not generic training — training on your configuration, your reports, your use cases.
Key areas to cover:
- Navigation and standard reports
- Explorations (funnel, path, cohorts)
- Segments and comparisons
- Export and Looker Studio connection
- Interpreting GA4 metrics (engaged sessions, engagement rate, etc.)
4. Activate BigQuery
If you have not done so already, activate BigQuery export now. It is free, and it gives you access to raw data — insurance against the limitations of the GA4 interface.
5. Validate Your Data Regularly
Set up a monthly validation process: compare GA4 conversions with your CRM data. If the gap exceeds 10%, investigate.
The Bigger Picture
GA4 is not going away. It is the reality for every business that relies on web analytics. The question is not whether to use it — it is whether you are using it correctly.
Two years of accumulated misconfigurations do not fix themselves. They compound. Every month with broken tracking is a month of bad data feeding bad decisions.
The good news: most issues are fixable in 2-4 weeks with a proper audit and remediation plan.
Was your GA4 migrated in a rush in 2023, and you have never done a thorough check? We perform a complete audit of your configuration to identify and fix the errors that have been corrupting your data for 2 years.