How to Create Custom Reports in GA4
The standard reports answer general questions and every business has specific ones. Closing that gap is what custom reporting is for. The discipline that makes it work is deciding the question before touching anything, because most custom reports answer nothing at all.
Decide The Question First
Most custom reports are assembled from whatever fields looked interesting at the time. They contain a great deal of data, they get looked at twice and they answer nothing, because nobody decided what they were meant to answer before building them.
What a good question looks like. Specific and answerable.
Which of our service pages produce enquiries. Whether visitors from search behave differently from the rest. Both have a shape, so the report builds itself.
What a bad one looks like. A subject.
Wanting a report about mobile users. Or about content. Those are topics rather than questions, so they produce a table nobody can act on.
Where to write it down. Before you open anything.
One sentence, on paper. If you cannot phrase it in a sentence, the report will not answer it either.
What this saves. The usual afternoon.
Building four elaborate reports, feeling productive, then never opening them again. That afternoon happens in most businesses exactly once.
Which tool for which question. A quick division.
Questions you will ask every month belong in a custom report. Questions you are asking once belong in an exploration, which is a different tool with its own guide.
What A Report Is Made Of
Two ingredients. What you are counting, plus how you are cutting it up. Every report in the tool, standard or custom, is some arrangement of those two things. Understanding that is worth more than any list of available fields.
The things you count. Numbers.
Visits, visitors, enquiries, revenue. Anything that can be added up across a period.
The ways you cut them. Characteristics.
By page, by source, by device, by country. Each one splits the numbers into rows so you can compare parts against each other.
How that becomes a report. Pick both.
Enquiries cut by landing page. Visitors cut by source. That combination is the whole exercise. The question from block one tells you which pair to choose.
Why we will not list the fields. They change.
The available set is revised regularly, so a reference published here would mislead within a year. The model does not change and the field names do.
The mistake this prevents. Adding everything.
A report with nine columns is harder to read than one with three and contains no more insight. Choose the pair that answers your question and leave the rest out.
Where to start. The standard set.
Check whether the standard reports already answer it. Most businesses never exhaust what exists by default.
Filters
There are two kinds of filtering and they are different in kind rather than in degree. One hides data you can bring back. The other throws it away permanently. Knowing which you are using before you save anything matters more than everything else on this page.
Filtering a view. Reversible.
Narrowing a report to show only part of the data. The rest is untouched and reappears when you change your mind. This is the ordinary, safe kind.
Filtering at collection. Permanent.
A rule applied as data arrives that excludes whatever it catches. That data is never recorded, so it cannot be recovered by removing the rule later.
What that means if you get it wrong. A hole.
A rule written too broadly quietly discards traffic you wanted. You discover it weeks later and the months in between are simply gone.
Why the distinction gets missed. Both are called filters.
The word covers a harmless view setting and an irreversible collection rule. Nothing in the language warns you which you are configuring.
What to do before saving one. Ask the question.
Is this changing what gets recorded or only what I am looking at. If the answer is the first, treat it with the caution you would give deleting a folder.
The one exception worth having. Internal traffic.
Which is block four. It is the case where permanent exclusion is exactly what you want.
Excluding Internal Traffic
Your own visits are in your figures unless somebody excluded them. For a small business that can be a substantial share of everything reported. It is the single most worthwhile filter available.
Who is included. Everybody in the business.
You, your staff, your developer and anybody checking a page after a change. All of them look identical to customers in the reporting.
What it distorts. More than visit counts.
Which pages appear popular, how long people seem to stay and any conversion rate calculated from those figures. Internal behaviour looks nothing like customer behaviour.
The clearest example. A new page.
Everybody in the business looks at it on launch day. The page appears to have landed well and the audience was your own team.
How it is done. A setting.
The property can be told to disregard visits from your own connections. Our setup guide covers where it belongs in the first ten minutes.
Its limitation. Not complete.
Staff working from home or checking on mobile connections will still be counted. It removes most of the distortion rather than all of it.
When to add it. Today, whatever the age of the property.
The distortion continues until somebody stops it, so there is no advantage in waiting for a tidier moment.
Custom Dimensions
A custom dimension records a characteristic the tool does not collect by default, so you can cut your reporting by something specific to your business rather than only by the standard categories.
What it is for. Your own structure.
Which category of service a page belongs to, which branch it relates to, which type of customer a page was written for. The tool has no way of knowing any of that.
What it enables. Better questions.
Comparing all your commercial pages against all your domestic ones, rather than reading fifty individual pages and trying to group them mentally.
The limitation that catches people. Not retrospective.
It applies only from the moment it is created. Nothing before that carries the characteristic, so there is no history and no year on year comparison until a year has passed.
What follows from that. Create early.
If you can see a grouping you will eventually want, set it up now rather than when you need the report. The cost of being early is nothing.
Who sets them up. Usually a developer.
Something has to tell the tool what value applies to each page, which is implementation work rather than a setting.
When to skip it. Small sites.
A business with a handful of pages can group them by eye. This earns its keep once the site is large enough that counting by hand stops being practical.
Audience Segments
A segment defines a group of people so you can look at that group specifically. It is not a report in itself. It is a way of narrowing whatever report you are already reading down to the visitors you care about.
What defines a group. Behaviour or characteristics.
People who arrived from search. People who visited a particular page. People who enquired. Any of those can become a group you examine separately.
Why it beats an average. Averages describe nobody.
A figure combining browsers, prospects and existing customers describes none of them. Segmenting is how you find out what the useful visitors have in common.
The most useful one for a service business. People who enquired.
Look at that group and ask what they did beforehand. Which pages, which sources, how many visits. That is where the actionable finding usually sits.
The second most useful. Your own area.
For a local business, isolating visitors from the places you serve stops the rest of the country diluting every figure.
What to be careful of. Small groups.
Cut the data finely enough and you reach a handful of people, where ordinary variation looks like a pattern. If a segment covers very few visitors, treat what it shows as a hint rather than evidence.
Where the limit comes from. Privacy behaviour.
Very small groups can also be withheld entirely, which our troubleshooting guide covers.
Keep It To A Few
The usual outcome of learning this is a workspace full of reports nobody opens. Three that get looked at every month beat twenty that were built in one enthusiastic afternoon and never revisited.
Why it happens. Building is satisfying.
Configuring a report feels like progress in a way that reading one does not. So the building continues past the point of usefulness.
What too many produces. Paralysis.
Nobody knows which report to trust or which was superseded. Faced with twenty, most people open none.
The test for keeping one. Did anybody open it.
If a report has not been looked at in three months it is not serving anybody. Delete it rather than leaving it to confuse whoever inherits the property.
How to build them. One at a time.
Create one, use it for a month and see whether it changed a decision. Then decide whether a second is needed. That pace produces reports that survive.
What to name them. The question.
Name each report after the question it answers rather than after its contents. In six months that is the only thing anybody will remember about it.
Who should own them. Somebody named.
A report belonging to nobody stops being maintained. One person should be responsible for each, which also settles who deletes it.
Who It Is For Changes What Goes In It
A report for the person doing the work and a report for the business owner are different documents about the same data. Trying to serve both with one produces something that fails both, which is how most monthly reporting ends up ignored.
For the person doing the work. Detail and diagnosis.
Page level figures, source breakdowns and enough granularity to find out why something moved. This is a working document rather than a summary.
For the business owner. Outcomes and direction.
How many enquiries, from where, against last period, plus what is being done about it. Short enough to read between jobs.
What owners do not need. Everything you have.
Sending a specialist's working view to somebody who wanted one number is how reporting gets ignored. It reads as evidence of activity rather than as information.
What both should contain. A comparison.
A figure without a comparison period is not information at either level of detail.
What neither should contain. Unlabelled changes.
If the measurement changed, say so on the report. Otherwise somebody reads a definition change as performance.
Where the wider argument sits. Our measuring performance material.
That covers what belongs in a monthly report and why activity gets mistaken for progress. Everything sits on the Google Analytics guide. a dashboard is how the owner version usually gets delivered.
Some filters
hide data.
Others throw
it away.
A rule applied as data arrives permanently excludes whatever it catches. Removing the rule later brings none of it back. A rule applied to a view only narrows what you are looking at. Both are called filters. Nothing in the wording warns you which one you are about to save.
How we build reporting:
If a report has not been opened in three months, it is not serving anybody and it should go.
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.
