Understand why GA4 problems must be fixed bottom-up, and how to sequence your audit for maximum impact.
This chapter explains the model and the priority order that makes the biggest difference first.
A GA4 account has four layers, each depending on the one below. Fix layer 3 without checking layers 1 and 2 and you get clean-looking reports that describe the wrong thing.
The four-layer model
Work bottom-up.
Image placeholder
Layer dependency illustration
How a layer 1 problem (such as the tag not firing on checkout pages) cascades up through layers 2, 3 and 4. Arrows connect the root cause to the corrupted metrics at the top: what looks like a reporting problem is a collection problem.
Most analysts start at layer 4: they open the Acquisition report, notice an odd channel split and try to fix it. But a broken channel split is usually a symptom of problems in layers 1, 2 or 3. Fix the foundations first and the reports often correct themselves.
This sequence addresses the highest-impact issues first, before refinements that won't matter if the data underneath is unreliable.
Check real-time reports while browsing the site. If a key page type (product pages, checkout, blog posts) isn't sending events, you have a gap that invalidates everything downstream. This is the most important check and takes under 10 minutes.
Your team's own visits distort every metric (engagement rate, conversion rate, channel mix). On a small site, a five-person agency team can be 10-15% of sessions. Filter internal traffic and activate the filter, not just create it in Testing state.
GA4 defaults to 2 months of retention for Exploration reports, and aged-out data is gone permanently. Changing it to 14 months takes 30 seconds and prevents a problem you can't fix retrospectively.
Check which events are key events: 3-5 should represent real business outcomes. Set counting to "once per session" for lead forms (avoiding duplicates) and "once per event" for purchases. Miscounted conversions are a common inherited problem.
Image placeholder
Conversion counting diagram
Before/after of how counting method affects reported lead form conversions. Once per event: a user submitting 3 forms in one session generates 3 conversions. Once per session: 1 conversion. Monthly numbers show the scale of the discrepancy.
Payment gateways (Stripe, PayPal, Afterpay) often send a referral back after the transaction, creating a phantom "referral" session that breaks the original source. Add these domains to the referral exclusion list so attribution is preserved through to the purchase event.
Once data quality is solid, fix how traffic and pages are grouped. Unassigned above 5% usually means UTM gaps in paid campaigns. No brand vs non-brand split in organic means SEO wins and brand-driven sessions are counted together.
Custom event parameters (a logged-in user ID, a content type, a product category) must be registered as custom dimensions in GA4 Admin before you can report on them. Unregistered parameters sit in the data but can't be queried or filtered.