The Universal Analytics sunset created a generation of broken GA4 implementations. Most dashboards people rely on are showing sampled, incomplete, or just wrong data — and nobody noticed because the numbers still look plausible.
Start with questions, not events
The most common GA4 mistake is event-naming as performance theater: tracking every click, every scroll depth, every accordion open. You end up with 400 events and no idea what any of them mean.
Instead, write down the actual business questions you need to answer:
- Where does revenue come from?
- Which content drives leads?
- Where does the funnel leak?
- Which campaigns actually pay back?
Build four or five core events that answer those. Mark the right ones as conversions (in GA4, conversions are events you've flagged — not a separate concept). Ignore the rest until you genuinely need them.
Key takeaway
The events we recommend for most B2B / lead-gen sites
form_submission— with a parameter for which form. Marked as conversion.phone_click— everytel:link. Marked as conversion on lead-gen sites.email_click— everymailto:link. Same.whatsapp_click— the floating FAB. Same.scroll_90— fires once at 90% page depth. A weak but real engagement signal.outbound_click— clicks to external domains. Useful for portfolio sites and partnership pages.
Server-side tagging is no longer optional for paid media
Safari, Firefox, and Chrome's ongoing privacy changes have made client-side tracking unreliable for attribution. If your ROAS looks 40% worse than it did two years ago and nothing else changed, this is probably why.
Server-side GTM routes tracking through a domain you control, which sidesteps most browser-level tracking restrictions. It's not a silver bullet — it doesn't recover data the browsers block, it routes what's left through a domain you control. Set expectations accordingly.
What to do next
Anakali Pawan Kalyan
Founder · Global Marketing Zone
Founded Global Marketing Zone in June 2026 after nearly a decade in performance marketing. Writes about the parts of marketing that actually compound — and the ones that don't.