CRO · GA4
GA4 as thescorecard, not the lift.
Key events worth testing against, a readable exploration, and no credit given to the install.
The short answer
Measurement for a test, not a migration trophy.
GA4 is the right scorecard when the events match the conversion you would actually change a template to improve. It is a weak fit as a project whose only outcome is 'we installed GA4.' The median lift on the primary conversion of templates we tested was 18%. The change was the template. GA4 was how that result was read. Installing GA4 did not produce the lift, and a property full of unused events will not produce it either.
A real event
Use GA4 when the conversion can be named.
This fits a team about to change a template, a form, or a checkout step and willing to call one event the primary conversion. A demo request, a purchase, a qualified call, or a booking can be that event. Explorations and the standard reports should agree on its name. If they do not, you are not ready to judge a test.
It is the wrong project when the hope is that a new property will raise conversion by existing. It is also wrong when consent or a broken tag means the event is missing for half the traffic and nobody wants to fix that first. We will not run a template test we cannot read. The install is not the experiment.
- One primary conversion is named before any template changes.
- The event fires once, on the action, for the traffic you can legally measure.
Events worth keeping
Fewer events, tied to the template.
We mark key events for the actions a test might move, and we archive the clutter that made every click look like success. Parameters carry the template or the form id so a winner can be tied to a page, not to the whole hostname. Ecommerce events are added only on a store, and only if they match the order. A newsletter signup is not labeled as a demo request because both are forms.
Debug views and the live site are checked. Internal traffic is filtered. The Google Ads link uses the same conversion meaning. Consent that blocks tags is documented.
- Key events limited to actions you might change the page to win.
- Form and template identifiers on the event, so results can be split.
Mistakes
A migration celebrated as a conversion win.
The failure is a property cutover, a spike of new events, and a slide that credits the tool for a change in conversion rate. Template tests we have run showed a median lift of 18% on the primary conversion. That lift belonged to the template. Attributing it to installing GA4 is how a measurement project steals credit from the page.
The other failure is every event marked as a key event. The reports shout. The test cannot pick a primary. Or the opposite: a purchase event that never fires because the tag lives on a thank-you URL the checkout no longer uses. We verify the event on the current path. A setup guide from a different checkout is not verification.
- No claim that the GA4 install caused a conversion lift.
- No key event list that includes scrolls, clicks, and the real conversion alike.
The test read
One primary conversion, before and after the template.
A template test is read on the primary key event, with the same definition before and after. 18% is the median lift on that primary conversion across templates we tested. Use it as context for the kind of work, not as a forecast, and not as an install bonus. Secondary events can explain the path. They do not get to replace the primary if the primary missed.
We also watch whether the data is complete: consent rates, tag failures, and a sudden drop that is a release rather than a design. A broken tag is fixed and the clock restarts.
- Primary conversion held to one definition for the whole test.
- Completeness checked before a winner is called.
Example engagement
A B2B software company judging a noisy form event.
B2B software company. GA4 was new. The key event called 'generate lead' fired on the demo form, the newsletter form, and a chatbot greeting. A homepage template test was about to be judged on that event. Any change that increased newsletter signups would have looked like pipeline.
We split demo requests from newsletter signups and watched a demo event fire once on the current thank-you state. The template test, when it ran, was read on the demo event only. Across template tests we have run, the median lift on the primary conversion was 18%. That figure is the template work. It is not what they got for installing GA4, and it was not a forecast for this homepage. The install stayed a measurement task. The page change had to earn its own read.
18%
Median lift on the primary conversion of templates we tested, not from installing GA4
The example uses anonymized results from a Quorum engagement. It is one account, not a benchmark you should budget against.
Questions
GA4 questions.
Will installing GA4 raise conversion?
No. 18% is the median lift on the primary conversion of templates we tested. The template changed. GA4 recorded it. A property migration does not inherit that lift.
How many key events should we have?
As few as you are willing to change a page to improve. One primary for the test, and a short list of supporting events. If everything is a key event, nothing is. We would rather delete markers than add another vanity click.
What if a consent banner blocks the tags?
Then we measure what we are allowed to measure and we do not pretend the rest is zero demand. A test that only sees consented traffic has to be described that way. Fixing the banner to be honest is part of making the scorecard real.
Do you build the experiment tool inside GA4?
We use GA4 to read the outcome. The change still has to ship on the site, in the theme or the app, with a way to tell variants apart. GA4 does not design the variant. If the event cannot distinguish them, the test is not ready.
Can paid media use the same events?
Yes, when the meaning matches. A primary conversion in GA4 and a primary conversion in the ad account should describe the same action. If ads optimize to a newsletter and the test optimizes to a demo, you will 'win' two different arguments. We align them before spend or design moves.
More in this lane
Other CRO platforms.
Talk through ga4.
One conversation. A diagnostic. A plan you can kill if it is not the work.