Segment analysis in order to improve CR

If you have ever created an online ad you will have come across the concept of segmentation. On social sites, we have the ability to select opportunities for different demographics, gender, interests. However, Facebook or Google will not know how visitors behave on our site. Ad sites know their own users very well but they don’t know how they behave outside of their own platform.

Project Overview

With Google Tag Manager, you have the possibility to extend the basic tag. This way, we can provide additional relevant information about visitors, which can help targeting of ad pages and thus reduce ad costs. However, we can add further options to our re marketing toolkit by looking at the behavior of our website viewers and creating groups. This is called segmentation.

The more visitors, the more segments. But resources are finite, so we need to create larger segments. When creating segments, we can focus on the main and micro conversion goals. In this case, we created segments focusing on the main conversion goal.

Here is the list about the used technology during the project.

Toolkit:

• BigQuery
• SQL
• Python
• Gephi

If you have ever created an online ad you will have come across the concept of segmentation. On social sites, we have the ability to select opportunities for different demographics, gender, interests, etc… However, Facebook or Google will not know how visitors behave on our site. Ad sites know their own users very well but they don’t know how they behave outside of their own site

Data Collection

BigQuery

In order to start the analysis, it is necessary to connect BigQuery powered by Google to our Google Analytics account.
What is BigQuery? BigQuery is a big data tool that gives you free access to the data sources collected by Google Analytics. This allows us to customize them to create a report that supports our assumptions.

BigQuery is free to use (subject to certain limits) and provides access to raw data, but it also has its drawbacks. As we wrote, the data is raw, which means that to make it understandable you need to format it. This requires some data perspective and knowledge of SQL programming language.

But if all these are available, then there is no choice but to move forward.

The process starts from the website tracking code. We check the analytics settings and then establish the connection with BigQuery. Then the data collection begins. At the start of the measurement, it is worth defining what events we want to see in our data source. Once we have a sufficient amount of data, we can start to capture it. At this point, the use of SQL programming language is necessary. The filtered data list can be processed in several different ways. We have used the Python programming language and have chosen machine learning classification. The steps of the process in bullet points.

Above you can see the result of the process discussed earlier. But what is this big blob?
Different colors belong to different segments. The following events were processed during the analysis:

• Number of sessions
• First visit
• Number of page views
• Scroll deep
• Number of user engagement
• Order control started
• Search button clicked
• Begin checkout
• Add payment info
• Number of item views
• Number of item list views
• Add to cart
• Select promotion
• Number of clicks
• Purchase event occurred

As you can see, the number of inputs is far more than the human brain can process and perceive at one time. But we don’t need to, because machine learning methods allow us to do a much wider range of research, and the result is that our results will be much more accurate. Without machine learning, we would not be able to analyze data in such depth.

What is the next step? Test all 10 segments. From the further conversion data we will see how we should prioritize the 10 groups. As soon as we can do further segmentation, it is worth repeating the segment analysis to see if we can see any additional patterns in behavior.

This method can be used to increase conversion rates more dramatically, resulting in 2-5x revenue without increasing advertising costs.