The last post in this cluster is about the part of GA4 that goes beyond “looking at reports”: defining which event counts as success, splitting credit between channels, building audiences and passing them to the ad account, and exporting raw data to your own warehouse. In other words, the stage where GA4 turns from a tool into a system.
Advanced GA4 rests on five pillars: (1) key events — what counts as success, (2) attribution — which channel gets the credit, (3) audiences — who to show ads to again, (4) consent mode — how data is modelled when there’s no cookie permission, (5) the BigQuery export — unlimited analysis of raw data.
This post builds on the previous four. If you don’t have an event setup yet, go back to the events and dataLayer post first; for reading reports, see the GA4 reports post.
Key events and Google Ads conversions
GA4 doesn’t know which event matters to your business — you tell it. You mark the event you consider important as a key event in the Admin → Events list. This name changed in 2024: before that they were simply called “conversions”.
Now the word “conversion” is used in GA4 only for actions passed to Google Ads. In practice the difference is:
Key event
- GA4’s own concept
- Appears as a column in all standard reports
- Attribution reports are built around it
- Sends nothing to the ad account
Google Ads conversion
- Created from a key event
- Visible only in GA4’s Advertising section
- The bidding strategy in Ads optimises for it
- Needed for automated campaign management
So marking an event as a key event doesn’t automatically send it to Google Ads — that’s a separate step.
Limit key events to 3–5. When you declare everything “important”, nothing is important: attribution reports lose their meaning, and Google Ads starts optimising for the wrong signal.
Attribution: who gets the credit?
Someone first sees an Instagram ad, two days later searches for your name on Google and finds you, and a week later visits directly and fills in a form. Who brought this lead? An attribution model answers exactly this question.
GA4 currently has three models:
| Model | How it works | When |
|---|---|---|
| Data-driven | Uses machine learning to calculate the real effect of each touchpoint — comparing paths with and without a conversion | The default; the most accurate for most cases |
| Paid and organic last click | Gives 100% of the credit to the last click (excluding direct visits) | For simple reporting and matching with ad accounts |
| Google paid channels last click | Gives credit only to the last Google Ads click | For reconciling with Google Ads reports |
In November 2023 Google completely removed four old models — first click, linear, time decay and position-based. If you come across old articles online about those models, they’re no longer relevant.
The lookback window
This setting decides “how many days before the conversion should touchpoints be taken into account”. It’s under Admin → Attribution settings:
- Acquisition events (
first_visit,first_open): default 30 days, alternative 7 days. - All other key events: default 90 days, alternatives 30 or 60 days.
For services with a long sales cycle (for example B2B), keeping 90 days is right — otherwise the first touchpoints drop out of the count. A change only affects the future.
Audiences and remarketing
An audience is a group of users who meet certain conditions. It’s built in GA4, passed to Google Ads and becomes an ad target there.
| Parameter | Free GA4 | Analytics 360 |
|---|---|---|
| Audiences per property | 100 | 400 |
| Conditions per audience | 10 | 10 |
| Maximum membership duration | 540 days (18 months) | 540 days |
| Backfill on creation | Up to 30 days (for Google Ads only) | Up to 30 days |
Three useful audiences for a service website:
- People who viewed the pricing page but didn’t get in touch. The hottest remarketing group.
- People who touched the form but didn’t submit it (
form_startyes,generate_leadno). They’re one step away. - Blog readers. They’re not ready to buy yet — show them content, not sales.
Audiences don’t fully work retroactively. When created, GA4 backfills the last 30 days of data (for Google Ads only), and history before that isn’t included. So audiences need to be built before the campaign starts.
Consent Mode v2
When a visitor chose “decline” in the cookie banner, under the old logic measurement stopped completely. Consent mode fills this gap: the tags still run, but don’t write cookies — they only send a cookieless signal. Google runs statistical modelling on those signals.
There are four parameters:
| Parameter | What it controls |
|---|---|
analytics_storage | Analytics cookies — GA4 measurement |
ad_storage | Advertising cookies |
ad_user_data | Sending user data to Google for advertising purposes |
ad_personalization | Personalised advertising (remarketing) |
The visitor declines
“I don’t accept” in the cookie banner
A cookieless signal
Minimal data that doesn’t identify the person is sent
Modelling
Google statistically recovers behaviour and conversions
Report
The gap is partly filled
The key difference: in the advanced setup, tags load before the banner and send a cookieless signal on decline — modelling works. In the basic setup, tags don’t load at all and no data is collected.
The BigQuery export
The GA4 interface is limited to ready-made reports. The BigQuery export moves raw event data — every event, every parameter, a separate row for every user — into Google’s data warehouse.
| Export type | How it works | What it’s for |
|---|---|---|
| Daily | All of the previous day’s data is exported once a day | Complete, unsampled analysis |
| Streaming | Data arrives within minutes | Near-real-time monitoring; no completeness guarantee |
In free GA4 the daily export limit is 1 million events. For a small or medium website that’s a distant number.
What does the export give you? Three things: no sampling — all data is raw; report limits don’t apply — the “(other)” row disappears; you can join it with your own data — you see real sales from your CRM in the same table as website behaviour. The last is the most valuable: you measure not the leads an ad brought but the money.
Data quality: three hidden traps
- Cardinality. When a dimension has very many different values (for example thousands of different page URLs), GA4 exceeds a daily limit and groups the remaining values into an
(other)row. The fix: group dimensions or move to BigQuery. - Thresholding. If a segment is too small, GA4 hides data to prevent individuals from being identified. You literally see an empty row in the report — it’s not an error but a protection mechanism.
- Data retention. Explorations are limited to 2 or 14 months. This limit doesn’t apply to the BigQuery export — data stays there as long as you want. That’s the real reason to use it for long-term analysis.
The order for moving to the advanced level
Clean up key events
Keep 3–5; leave the rest as regular events.
Check the attribution settings
Data-driven model, lookback window matched to your sales cycle.
Build audiences
Before the campaign starts — only 30 days are backfilled.
Set up consent mode correctly
The advanced variant — modelling only works there.
Turn on the BigQuery export
Turn it on even if you don’t need it yet — the export doesn’t work retroactively, so today’s decision protects your future history.
The essence of advanced GA4 isn’t more reports — it’s fewer reports that drive decisions. Reducing the number of key events is often more useful than connecting a new tool.
What’s the difference between a “conversion” and a “key event”?
In 2024 Google renamed the “conversion” concept in GA4 to “key event”. Now the word “conversion” is used only for actions passed to Google Ads that drive the bidding strategy there. Key events appear in GA4 reports; conversions appear in Ads and in GA4’s Advertising section.
Which attribution model should I choose?
For most cases, the data-driven model — it’s the default and gives the fairest split when you use several channels. Use the last-click model only when you need to reconcile numbers with ad accounts. First-click, linear and time-decay models no longer exist in GA4 since 2023.
When I create an audience, do past users get into it?
Partly. When an audience is created, GA4 backfills the last 30 days of data, and this works only for passing to Google Ads. Older history isn’t included. So remarketing audiences need to be built well before the campaign, so the audience is already full when the campaign starts.
What do I lose without consent mode?
Every visitor who declines in the cookie banner drops out of measurement completely — neither their session nor their conversion is counted. In the advanced variant of consent mode, a cookieless signal is sent and Google partly recovers the missing part with statistical modelling. As a result, the gap in reports shrinks.
Is the BigQuery export paid?
The export itself is free on the GA4 side (up to 1 million events a day for a free property). On the BigQuery side, Google Cloud pricing applies for storage and queries. For small websites this amount is usually very low, because there are free usage tiers.
What is server-side tagging and do I need it?
In a normal setup, data goes from the visitor’s browser directly to Google. With server-side tagging, it first goes to a server you control and is passed to Google from there. The advantages: ad blockers have less impact, the page gets a little faster and you have more control over the data. In return it requires extra server costs and technical support — small websites usually don’t need it.
How can I get rid of the “(other)” row in reports?
This row appears when the cardinality limit is exceeded: a dimension has too many different values and GA4 doesn’t store them all separately. Practical fixes: group page URLs, don’t write unnecessary unique values (for example order numbers) into parameters, and use the BigQuery export for deep analysis — this limit doesn’t exist there.
Conclusion
The advanced part of GA4 isn’t about learning new buttons — it’s about tying measurement to business decisions. Key events show what success is, attribution shows who gets credit, audiences show who to reach again, and BigQuery shows the money an ad brought in.
The start of this cluster: what GA4 is and how it works. If you’d like me to build the whole system for your business — see the analytics and tracking page or fill in the get a quote form.
Source: Google official documentation — attribution, audiences, consent mode, BigQuery export, conversions vs key events.
