How to Set Up Google Analytics 4 From Scratch
The setup itself takes about ten minutes. Several of the choices made in those ten minutes cause problems eighteen months later. None of them looks important at the time. This guide is about those choices rather than about where the buttons are.
Ten Minutes That Decide A Lot
Creating an analytics property is genuinely quick and that is exactly the problem. It gets done in a hurry during a website build, several defaults get accepted without anybody reading them, then the consequences appear much later when somebody finally wants an answer the setup cannot give.
Why the mistakes hide. Nothing breaks.
A badly configured property still produces reports. They look complete, the numbers move about convincingly and nothing announces that the important half was never switched on.
When they surface. At the worst moment.
Usually when a business asks how many enquiries came from search last year. Or wants to compare against the same period. The data to answer it does not exist.
Why that cannot be fixed later. No retrospective collection.
Analytics records forward only. Anything not being collected today is simply absent from history, permanently. No supplier can recover it.
What this page covers. Decisions, not buttons.
The interface changes regularly, so a walkthrough would date within months. Everything here is what to decide and what each choice costs you.
Who normally does this. Whoever is nearest.
A web designer during a build, an agency during an onboarding or a business owner on a wet Sunday. None of those people is necessarily thinking about what the business will want to know in two years, which is how the defaults survive.
What it does not cover. Putting it on the site.
That is a separate job. the installation guide covers it including the ways sites end up measuring everything twice.
The Structure, As A Decision
Three things nest inside each other and the naming is not intuitive. Understanding what each one is for takes a minute and prevents the arrangement most businesses regret, which is one they set up without realising there was a choice.
The account. The business.
The outer container. It should represent your company rather than a website. One business, one account, whatever it owns underneath.
The property. The website.
Where the data actually lives and where reporting happens. A property covers one website. It is the level at which almost everything is configured.
The stream. The connection.
How the site talks to the property. It produces the identifier that gets installed, which block three covers.
The decision to make before creating anything. Several sites.
A business with more than one website has to decide whether they share a property or get one each. Separate properties keep the reporting clean and mean no combined view. One property with everything in it produces a combined view and blurs the sites together.
What we recommend for most businesses. Separate.
Distinct websites usually serve distinct purposes. Mixing them makes every figure harder to read. Combining later is easier than untangling.
The mistake worth naming here. A property per page.
We have seen businesses create a property for each section of one site, which fragments the data completely and cannot be merged afterwards.
Data Streams
A stream is the link between one website and one property. It exists so the property knows where its data is coming from. It produces the identifier you install. For a single website business that is the entire subject.
What it gives you. An identifier.
A short reference that goes on the website. That is the thing your developer or platform is asking for when it requests a tracking code.
How many you need. One per website.
Not one per page, not one per section and not one for the mobile version. A website is a website as far as this is concerned.
The common error. Several for one site.
Businesses create additional streams thinking it separates something usefully. It does not. It splits one site's data across parallel routes and makes the reporting harder to read for no gain.
The other common error. The wrong address.
A stream pointing at a different version of the address than the site actually serves. That produces a property that looks configured and receives nothing, which is a leading cause of the empty reports our troubleshooting guide covers.
What a stream is not. A filter.
Some businesses expect streams to separate sections of a site or types of visitor. They do neither. Separating anything within one website is a reporting job rather than a structural one. Our custom reports guide covers it.
What to check afterwards. That data arrives.
Never assume installation worked. Our troubleshooting guide covers verifying it and what an empty property usually means.
The Mistakes That Cost You Later
Five mistakes, ordered by what they cost rather than by how often they happen. Every one is made during setup, every one is invisible at the time and the first two are the reason most inherited properties cannot answer a commercial question.
Nothing configured as a result. The worst.
A property recording visits and nothing else. It will produce reports for years and none of them will tell you whether the website generated any work. Block eight covers the fix.
Internal traffic not excluded. Self inflicted inflation.
The business, its staff and its developer all visiting regularly, counted as customers. On a small site that can be a substantial share of everything reported.
Test and staging traffic included. The same problem, worse.
A development copy of the site sharing the same identifier means test visits and real ones are pooled. It is common after a rebuild and nobody notices.
Retention left at the default. Silent data loss.
The shorter option is the default, so detailed history is discarded on a rolling basis. Block five gives this its own space because of when it gets discovered.
Nobody but the agency holding access. The commercial one.
An account created under a supplier's own login, which becomes a problem at exactly the moment the relationship ends. Block seven covers it.
The one that is easiest to fix. Also the last noticed.
Excluding internal traffic takes moments and is almost never done at setup. It is worth doing today even on a property that has been running for years, since the distortion continues until somebody stops it.
What they have in common. None of them errors.
Not one produces a warning or a broken report. That is why they survive for years in properties that look perfectly healthy.
Retention Is The One Nobody Notices
The property discards detailed data after a period. The shorter option is what you get unless somebody changes it. Nothing warns you. The loss is discovered a year later by somebody trying to compare against last year and finding there is no last year to compare against.
What it affects. The detailed data.
Headline figures are generally retained longer. The ability to interrogate them, segment them and answer specific questions about an old period is what disappears.
Why the default is short. Storage and privacy.
Holding detailed behavioural data indefinitely is neither cheap nor uncontroversial, so a conservative default is reasonable. It is simply not the right setting for most businesses.
What changing it costs. Nothing.
It is a setting. There is a longer option available on the free version and switching to it takes moments, which is what makes overlooking it so annoying.
When to change it. Immediately.
On the day the property is created, because the setting is not retrospective. Extending it today does not bring back what has already been discarded.
What to do regardless. Export.
Even at the longer setting, data eventually goes. A short monthly export of the figures that matter is the only history nobody can take from you.
Where this bites hardest. Seasonal businesses.
A trade wanting to compare this winter against last winter needs more than a rolling few months. That is precisely the comparison the default prevents.
Exclude Yourself
A small business visiting its own website every day is counted as an audience. On a site with modest traffic that can distort every figure on the page. The business ends up reading its own behaviour back as market demand.
Who counts as internal. More people than you think.
You, your staff, your developer, your designer and anybody checking a page after an update. All of them look identical to customers.
What it distorts. Nearly everything.
Visit counts, which pages appear popular, engagement figures and any conversion rate calculated from them. Internal visits behave nothing like customer visits.
The version that misleads most. A new page.
Everybody in the business looks at it on the day it goes live. The page appears to have launched successfully and the traffic was entirely your own team.
How it is done. A setting, not a technique.
The property can be told to disregard visits from your own connections. It is a configuration matter rather than anything clever. It belongs in the first ten minutes.
The related case worth handling. A staging site.
A development copy carrying the same identifier pours test visits into the same reports. If your site has a staging version, that needs separating before anybody trusts a figure from either.
The limitation to know. It is not perfect.
Staff working from home or on mobile connections will not be caught. It removes most of the distortion rather than all of it.
Ownership And Access
The analytics account should be created in the business name, on an address the business controls, with agencies and developers added as users. That sentence prevents a specific and common problem. It costs nothing to get right at the start.
What goes wrong otherwise. The history leaves.
An account created under a supplier's login belongs to that supplier. When the relationship ends, so does access to years of your own data.
How often we see it. Regularly.
It is among the commonest problems on inherited accounts, alongside the same arrangement on advertising accounts and business listings.
Whether it is deliberate. Usually not.
Mostly it happens because somebody set it up quickly while logged into their own account. The effect is the same regardless of intent.
What to do if it has already happened. Ask, early.
Access can be transferred while the relationship is good. Requesting it during a handover is considerably harder than requesting it now.
Who should hold the top level. An owner.
Somebody who will still be at the business in five years, on an address that survives staff changes. Not a personal account belonging to whoever built the site.
Where the wider position sits. Our hiring material.
The ownership commitment covers listings, advertising and analytics together. It is one of the questions worth asking any agency before signing.
Set Up Conversions On Day One
A property recording nothing but visits is answering the wrong question from the moment it starts. Deciding what counts as a result, then configuring it, is the single most valuable thing on this page and it is almost never done at setup.
Why it belongs at the start. No history otherwise.
Configure it in a year and you have a year of data about visits and nothing about outcomes. The comparison you eventually want will not exist.
What counts as a result. Your answer, not a template.
A form sent, a number tapped, a booking made, a quote requested. It should come from how the business actually gets work rather than from what is easy to measure.
What most businesses configure. Whatever is easiest.
Which is usually a page view on a thank you page, if anything at all. That is better than nothing and it is rarely the full picture.
What this changes in the reporting. The question.
Instead of how many people came, you can ask how many enquired and from where. Every other page in this cluster becomes more useful once that exists.
What to decide before configuring anything. The list.
Write down every way a customer currently contacts you, in order of value to the business. That list is the specification. It takes ten minutes with a pen rather than any time in the tool.
How long it takes. Half an hour.
For most small businesses, less. Setting up conversions covers doing it properly, including the telephone problem. Everything sits on the Google Analytics guide.
A badly set up
property still
sends you
confident reports.
Nothing warns you that results were never configured, that your own visits are in the figures or that your history is being discarded on a rolling basis. The reports keep arriving and they look complete. The gap appears a year later when somebody asks a question the setup was never able to answer.
What we check on any property:
If your account was created under a previous agency's login, that is worth resolving while the relationship is still cordial.
Every guide.
One tool.
Getting it installed and configured properly, reading the reports without being misled, measuring outcomes rather than activity, the advanced reporting and what to do when the numbers look wrong.
