AG4: Comparisons, segments

If you have tried working with GA4, you have probably noticed that there are different techniques available for "segmenting" data for analysis or creating subsets of data based on specific conditions.

At GA4 we are fortunate to have three different techniques to create these subsets of data when for example at Universal we only had two ways: Advanced Segments and Audiences.

We are fortunate to have three different techniques to create these subsets of data when for example at Universal we only had two ways: Advanced Segments and Audiences.

We have noticed that the GA4 user has a hard time understanding the concept of each and the main differences. But what is more difficult sometimes is deciding which to use to get the data you need. Well, we have tried to simplify all this to make it easier for you.

Well, we have tried to simplify all this to make it easier for you.

First we explain briefly what each one of them is with its features and limitations: 

Segments

Our main knowledge about segments comes from Universal Analytics in which we had the advanced segments and in which we could make subsets of data of any type with any dimension and with any metric in even make sequential segments: step 1, step 2, step 3

Features


  • You can use segments in explorations, such as Free Form or Funnels, you can apply up to 4 segments at the same time.

  • Segments are retroactive.

  • You can create them from scratch in scans or export them from the comparison in standard reports.

  • You can create sequential segments but only in the User Scope.

Limitations:


  • Segments cannot be used in standard reports. 

  • Segments are not shared between different scans. 

  • You cannot use segments in Looker Studio.

Audiences

Audiences are another type of user groups/subgroups used in analytics. We were familiar with audiences already in Universal Analytics and used to use them to import from Google Ads and do retargeting campaigns.

Features


  • You can use audiences as a dimension of a comparison.

  • You can either create an audience from scratch or create it from a segment.

  • You can import audiences and use them in Looker Studio.

  • You can import the audiences from Google Ads and create audiences for display campaigns.

Limitations:


  • Audiences are not retroactive. They start accumulating data from the moment you create them.

  • Audiences cannot be edited (except for the name and description and activating the audience trigger). But their settings cannot be changed.

Comparisons

Comparisons in Google Analytics allow you to quickly compare different subsets of data in standard reports. They are the alternative to segments that were used in previous versions of GA.

Features


  • Persist when you navigate from one standard report to another. 

  • Comparisons are retroactive unless you use an audience as the dimension.

  • Comparisons can be imported from Looker Studio.

  • Only up to 5 conditions can be added in a single comparison.

  • You can use up to 4 comparisons at the same time.

  • They can be exported to scans and converted to segments.

Limitations:


  • Can only be used in standard reports.

  • .
  • No OR conditions can be used only AND conditions.

  • Does not accept regular expressions.

  • Some useful dimensions are missing.

These are the main differences between them:


  • Retroactive: Only the Hearings are not retroactive.

  • Explorations: Only segments can be used in Explorations, such as Free Form or Funnels. Audiences cannot be used in Explorations (but you can build audiences from segments within Explorations). And Comparisons cannot be used in explorations directly. But if you click the Explore button in the comparison's navigation sidebar, those comparisons will become segments in that particular scan.

  • Standard Reports: only segments can't be used in Standard Reports.

  • Export to Looker Studio: both audiences and comparisons can be exported

  • Sequences or steps: you can use sequence or steps in Audiences and also in segments but only in User Scope user scope.

To make all this easier for you you can refer to this diagram:


GA4 - Segmentation Diagram

It's still a little bit of a diagram.

It's still a bit of a mess so when you have to create a subset of data you should start by asking yourself the table questions, from most important for your task to least.

Example 1:

Am I going to use the data subset in Looker Studio? Well, I choose Audience or Comparison.

Well, I choose Audience or Comparison.

Do I need the data subset to be retroactive? Then it has to be a comparison

It's just that I could do with it being sequential, i.e. step 1, step 2... so you have to choose Audience and forgo retroactivity.

Is it imperative that it be sequential and also retroactive? Then I have to forgo importing the dataset from Looker and make a segment that I will use in an Exploration.

 

Example 2:

Do I need the subset of data for Google Ads? I have to create an audience.

I hope you found it helpful. If so, please share with anyone who might be interested.

Thanks!

ANTERIOR
SIGUIENTE

TIPS DE EXPERTOS

Suscríbete para impulsar tu negocio.

ÚLTIMOS ARTÍCULOS

Group your data like a pro: clustering with K-Means and BigQuery ML

Working with large volumes of marketing data—whether it’s web traffic, keywords, users, or campaigns—can feel overwhelming. These data sets often aren’t organized or categorized in a useful way, and facing them can feel like trying to understand a conversation in an unfamiliar language.

But what if you could automatically discover patterns and create data groups—without manual rules, endless scripts, or leaving your BigQuery analysis environment?

That’s exactly what K-Means with BigQuery ML allows you to do.

What is K-Means and why should you care?

K-Means is a clustering algorithm—a technique for grouping similar items. Imagine you have a table with thousands of URLs, users, or products. Instead of going through each one manually, K-Means can automatically find groups with common patterns: pages with similar performance, campaigns with similar outcomes, or users with shared behaviors.

And the best part? With BigQuery ML, you can apply K-Means using plain SQL—no need for Python scripts or external tools.

How does it actually work?

The process behind K-Means is surprisingly simple:

  1. You choose how many groups you want (the well-known “K”).

  2. The algorithm picks initial points called centroids.

  3. Each row in your data is assigned to the nearest centroid.

  4. The centroids are recalculated using the assigned data.

  5. This process repeats until the groups stabilize.

The result? Every row in your table is tagged with the cluster it belongs to. Now you can analyze the patterns of each group and make better-informed decisions.

How to apply it in BigQuery ML

BigQuery ML simplifies the entire process. With just a few lines of SQL, you can:

  • Train a K-Means model on your data

  • Retrieve the generated centroids

  • Classify each row with its corresponding cluster

This opens up a wide range of possibilities to enrich your dashboards and marketing analysis:

  • Group pages by performance (visits, conversions, revenue)

  • Detect behaviors of returning, new, or inactive users

  • Identify products often bought together or with similar buyer profiles

  • Spot keywords with unusual performance

How many clusters do I need?

Choosing the right number of clusters (“K”) is critical. Here are a few strategies:

  • Business knowledge: If you already know you have 3 customer types or 4 product categories, start there.

  • Elbow Method: Run models with different K values and watch for the point where segmentation no longer improves significantly.

  • Iterate thoughtfully: Test, review, and adjust based on how your data behaves.

Real-world examples

With K-Means in BigQuery, you can answer questions like:

  • What types of users visit my site, and how do they differ?

  • Which pages show similar performance trends?

  • Which campaigns are generating outlier results?

Grouping data this way not only saves time—it reveals opportunities and issues that might otherwise go unnoticed.

Conclusion

If you're handling large data sets and need to identify patterns fast, clustering with K-Means and BigQuery ML can be a game-changer. You don’t need to be a data scientist or build complex solutions from scratch. You just need to understand your business and ask the right questions—BigQuery can handle the rest.

Start simple: take your top-performing pages, group them by sessions and conversions, and see what patterns emerge. You might uncover insights that completely shift how you approach your digital strategy.

 

Claude 4.0: Advances and Challenges in Conversational AI

Artificial Intelligence (AI) continues to progress at an accelerated pace, and Claude 4.0, developed by Anthropic, marks a major milestone in this journey. This next-generation language model stands out for its ability to comprehend complex contexts, deliver accurate responses, and adapt to a wide range of business needs.

AlphaEvolve: The new coding agent powered by Gemini

In a world where technology advances at unprecedented speed, artificial intelligence has emerged as a key driver of transformation. Among the most promising innovations today is AlphaEvolve, an evolutionary coding agent that combines the creative power of large language models (LLMs) with automated evaluators, opening new frontiers in software development, algorithm optimization, and solving complex problems in mathematics and computing.

How AI Is Revolutionizing Design and Development

At its Config 2025 event, Figma made it clear: the future of digital design will be deeply shaped by artificial intelligence. Beyond announcing new features, the company highlighted a paradigm shift — design is no longer a standalone process, but the core that connects creativity, technology, and product development.

data
Mallorca 184, 08036
Barcelona, Spain