Most GA4 problems do not come from the platform being complicated. They come from GA4 being installed — often in a rush, during the Universal Analytics migration — without anyone seriously checking what data comes in, how events are defined and whether the reports actually answer the business's questions. The setup "works", numbers keep flowing, and only months later someone notices the numbers do not add up.

The good news: you do not need a specialist to figure out whether you have a problem. You only need one to fix it. Here are the signs.

Clear signs the implementation has problems

GA4 conversions do not match reality. The simplest test: count your real leads or orders for a month — emails received, orders in your system — and compare with what GA4 shows for the same period. A few percent of difference is normal. A 30–50% gap, in either direction, means you are either losing events or counting them twice.

Some forms send events, others do not. Typical for sites built in stages: the contact form is tracked, but the quote form — added later — sends nothing. The result is that part of your enquiries are invisible in reports, and the campaign that brought them looks like it "does not work".

Duplicated events and inconsistent naming. If your reports show form_submit, formSubmit and form_submission side by side, tracking was added at three different moments without anyone checking what already existed. Duplicates inflate the numbers; chaotic naming makes reports unreadable six months later.

Reports cannot answer simple questions. "How many enquiries came in this month, and from where?" — if answering that takes ten minutes of digging through explorations, the problem is not the reporting. It is the data model underneath.

Why implementations go wrong

Almost always for the same three reasons. First: the automatic migration from Universal Analytics, which created GA4 properties with default settings and mechanically translated events. Second: tracking added in layers, by different people, at different times — each adding their own tag without checking what was already there. Third: the absence of a model — nobody ever decided what counts as a conversion for this business, so every tool counts something different.

What you can check yourself, in 30 minutes

  • The reality test — the real-leads vs. GA4 comparison above. It is the most valuable check and needs no tools.
  • Submit a test form and watch GA4's Realtime report for the event — appearing once, not twice. Repeat for every form on the site, not just the main one.
  • Open the events list in GA4 (Admin → Events) and read the names. If you cannot tell from the name what each one measures, neither can your reports.
  • Check what is marked as a conversion. A typical sign of a neglected setup: page_view or session_start marked as key events. Conversions should be only the actions that bring money.

What a serious audit checks

A tracking audit goes beyond the list above: it verifies GTM tags and triggers one by one, hunts for duplicates between hardcoded tracking and GTM, validates that consent data (Consent Mode) is not blocking legitimate measurement, and — the step most often skipped — connects every event to the business question it must answer. The result is not a document of observations, but a repair list in the order the fixes are worth making.

Repair or rebuild from scratch?

The question comes up after almost every audit. The practical rule: if the problems are isolated — one untracked form, two duplicated events — you repair on top of the existing setup, in hours, not days. But if the data model is chaotic at its foundation — meaningless names, conversions defined differently in every tool, hardcoded tags mixed with GTM — a clean rebuild is usually cheaper than patching, because every repair on a bad foundation adds another layer of complexity someone will have to understand a year from now.

A simple decision criterion: if nobody in the company can still explain what each event measures and why it exists, the model — not the implementation — is the problem. And the model gets rebuilt once, properly, with names you will still understand in two years.

What you gain

Tracking you can trust changes how you decide: you know which campaigns bring enquiries and which only burn budget, you see where visitors drop off, and you can answer "what worked this month?" in two minutes instead of a meeting. And once the data is clean, every next step — optimisation, automated reporting, automation — is built on solid ground instead of sand.

If you want to know exactly what is right and wrong in your setup, here is how an audit works — or write to me directly and we will find the starting point together.