Leggi in italiano →

Strategy & Data ·

GA4 three years later: the 5 mistakes I keep seeing

TL;DR

After three years of GA4 on real projects, the recurring mistakes are not bugs but choices: 1) measuring everything without knowing which decision the data will support; 2) trusting conversions without checking how they are defined; 3) ignoring consent and then being surprised by gaps in the data; 4) comparing GA4 and Universal as if they were the same tool; 5) building reports nobody uses. GA4 is not hard, it is honest, and it shows how little we knew before.

GA4 has earned a bad reputation, partly because of the interface and partly unfairly. After three years using it on real client projects, I have noticed that the most expensive mistakes are not technical. They are setup mistakes. Here are five.

1. Measuring everything, deciding nothing

The first instinct with GA4 is to track every possible event. Then you end up with two hundred events and no answers. A number that does not change a decision is furniture. First the question, such as whether a layout change helped, then the measurement. Never the other way round.

2. Trusting conversions without checking their definition

“Conversions are down” means nothing until you know what counts as a conversion. Often it is a badly defined event: a click on a button that leads nowhere, or a form that fires the event even when submission fails. Before panicking, open the definition.

With Consent Mode and many users refusing tracking, GA4 estimates and models part of the data. If you read those numbers as absolutes, you make bad decisions. The data is directional, not cadastral. Knowing how consent is configured is half the interpretation.

4. Comparing GA4 and Universal as if they were twins

Different data model, different session definition, different attribution. Comparing GA4 numbers from 2024 with Universal numbers from 2022 and calling it a trend is a method error. They are two tools, not two versions of the same ruler.

5. Building reports nobody uses

The best report is the one someone reads and uses to decide. Many dashboards exist to reassure, not to inform. AI helps here, but not in the way many think: it is not for writing more slides, it is for asking better questions of the data. A good report ends with “so we do X”, not with one more chart.

The uncomfortable truth

GA4 is not hard: it is honest. It shows how little we knew before, when Universal told us convenient stories with inflated sessions and comfortable attribution. Annoying? Yes. But an uncomfortable true number is worth more than a comfortable false one.

If you are interested in how I use data to decide, that is my Strategy & Data pillar.