How to configure GA4 to decide, not to watch
Start with the events not to install, the four that actually decide something, why value beats count, and what consent does to your numbers.
Almost every company has GA4 installed. Almost none opens it. And when they do, they look at sessions and users — exactly the part that decides nothing.
The problem is not the tool: it is that it was installed to have data rather than to answer questions. That difference shows up in the configuration from day one.
Why the account nobody opens
Because it does not answer the question the person actually has. Whoever opens GA4 wants to know whether a channel is worth more money, and what they find is a table of page views.
The automatic setup measures generic behaviour: entrances, exits, scroll, outbound clicks. All real. None of it knows which action on your site is worth money, because that cannot be guessed — it has to be declared.
Start with what not to install
This is the part no tutorial covers, and where most time gets wasted. The rule is uncomfortably simple: if nobody can say what decision an event would change, that event does not go in.
What is almost always surplus:
- Menu and footer clicks. Nobody ever changed a strategy because someone clicked “About”.
- Scroll at every percentage. Lots of volume, zero decisions.
- Every form interaction. Every focus and keystroke on every field, when what matters is whether it was submitted and from where.
- Personal data inside parameters. This is not just noise: sending emails, phone numbers or names inside an event breaches the tool’s own policies and the data minimisation that regulation requires. Never.
That last point makes this more than hygiene. It is set out in how we handle data: collect what the goal of the service requires, and nothing beyond it.
The four events that actually matter
For a service business, almost everything gets decided with these four:
| Event | What it decides | Parameter that cannot be missing |
|---|---|---|
| Qualified contact | Whether the channel brings useful people | Where it came from and which service was requested |
| Checkout or quote started | Where the process breaks down | Estimated value |
| Sale or close | Which channel brings money | Real value and your own identifier |
| Contact via direct channel | How much traffic leaves to WhatsApp or phone | Which page it left from |
The third column is the one most often skipped. An event without parameters tells you how many times it happened, which is the useless half of the information. With origin and value inside, the same event tells you which channel is worth it.
And be careful with “qualified contact”: if the event fires on any form submission, you will be optimising toward spam. Better to have the form fire it only once a minimum validation passes.
Send the value, not just the count
This is the cheapest change and the one that most changes the reading. If every event is worth 1, GA4 tells you where the volume is. If each carries its estimated value, it tells you where the money is — and those two answers rarely match.
You do not need the exact amount at the moment of the event. An estimated value per service type already reorders the whole report. And if you later send the real value back from your system, you close the loop — which is what the first-party data article is about.
What consent changes in the numbers
If your site asks for real consent — with the option to refuse — a share of visits will not be measured with cookies. That is not a configuration error: it is the correct design.
What is worth settling before somebody misreads it in a meeting:
- Measured traffic drops the day it ships. Expected, and not a drop in traffic.
- Annotate the change date inside the tool, so that three months later nobody mistakes it for a problem.
- Part of what you see becomes estimated rather than observed. Knowing which part is what stops you deciding on a projection while believing it is a count.
It happened to us here: this site starts with analytics denied and only enables it if the visitor accepts. The effect on the numbers is exactly the one described.
A checklist you can run in an hour
- List the three decisions you want to be able to make with data. If there are not three, there are fewer events to install.
- Define the minimum event that answers each one.
- Add origin and value to each event.
- Mark as conversions only those that represent money, not the intermediate steps.
- Check that no parameter carries personal data.
- Turn off whatever is left without an owner and without an associated decision.
Step 6 is the one nobody does and the one that cleans the most. An account with six events somebody reads is worth more than one with forty nobody understands.
FAQ
- Is the automatic setup enough? It gives you traffic, sources and basic behaviour, which is real but generic. What it cannot know is which action on your site is worth money. Until you tell it, every report describes visits rather than outcomes.
- How many events should I have? Fewer than you think. If nobody can say what decision an event would change, it is not measurement — it is collection. Four well-defined events beat thirty nobody reads.
- Do I need Tag Manager? Not necessarily. It helps when many tags change often or several people touch them. For a site with four events and one owner, it can be more moving parts than value.
- Why do GA4 and the ad platform disagree? Because they attribute differently, over different windows, and one of them models what it cannot observe. That gap is expected. Treat GA4 as your site’s account of events and the platform as its own claim of credit.