GA4: the future is modelled (and what that means for you)

Today, Google announced a major milestone in the evolution of Google Analytics. Universal Analytics will be sunset from 1st July 2023.

GA4 has already been around for some time, and we already know it as the future of Google’s measurement offering. But from mid-2023, it will be official. The next generation of Google Analytics will bring with it many new features and has been the focus of Google’s product development for several years. You can read all about what the product offers in Google’s announcement.

The focus of this article will be the role that data modelling will play in GA4 – and what it means for you. Our aim is to create a series of blog posts that will take a deeper look at some of the areas that will be most interesting for advertisers as they consider how GA4 fits into their business.

The development of machine learning in Google Analytics

Google Analytics 4 promises to make extensive use of machine learning. There are two primary reasons for this. Firstly, machine learning technology has advanced immeasurably since the release of Universal Analytics in 2012, allowing Google to roll it out efficiently at scale. Secondly, Google sees it as a necessity. As privacy technology, regulations, and user expectations have dramatically changed in recent years, data collection is rightfully more limited because of dependencies on consent and privacy features introduced by browsers such as Safari.

There are many benefits to expanded modelling in reports. Thanks to the vast amount of data that Google sits on, the talented engineers building the product, and the computational power at their disposal, Machine Learning in GA4 will be able to generate insights and models that would be extremely difficult for any individual advertiser to reach on their own. The reports will be a useful tool in the arsenal of advertisers who want to benefit from automatically generated insights and a solution to fill the gaps left by the effects of privacy on data collection.

A paradigm shift (the future is modelled, or the future is a black box?)

The introduction of these models signals a fundamental change in the nature of Google Analytics as a product. Google Analytics has always presented observed data, which means that the data you see is a true reflection of the actual data that was collected. The use of modelling was mostly limited to attribution reports, where the models were largely deterministic, which means that they are either based on understandable rules, or that when we run data through the model we would always come out with the same answer.

Now, machine learning will be used to fill in the gaps in observed data and – in Google’s words – enhance reports. This means that the data in many more parts of the product will be based on Google models.

These models present some challenges for advertisers to be aware of as they build their strategy.

Firstly, models in GA4 will be black box. This means that the algorithms and underlying assumptions of the model aren’t available to advertisers. This raises some valid questions. For example, how does Google achieve a high quality model for behaviour on iOS devices, when iOS devices have had extensive privacy features for many years which reduce the quality and quantity of data that can be collected?

Secondly, the models are owned and operated by Google. This means that many core reports, including attribution, will be controlled by the same company who sells online advertising in huge volumes. This has always been a concern with Google Analytics – some advertisers express unease that the measurement of their media buying is performed by the same entity who is selling the media, and a similar debate flared up when Google launched data-driven attribution modelling in Universal Analytics. Now, GA4 represents a big development in this area, where modelled data is both far more extensive and more opaque than ever before.

As an example, Modelled Conversions in Google Analytics 4 will result in fewer conversions attributed to the direct channel thanks to machine learning models which assign conversions to other channels based on Google’s predictions and data about what really happened. This can be useful as a picture of how channels perform according to Google’s best efforts. But advertisers must be aware that, even with the best intentions, all models come with underlying assumptions and biases that affect the shape of the data we consume. This consideration must be part of the equation when building an analytics and attribution strategy for your business.

What this means for advertisers

Advertisers who are either already using, or planning to use GA4, should take the time to thoroughly understand the product and roadmap. GA4 comes with many powerful features, and, when used correctly, the ML and modelling features will generate useful insights that would be hard to achieve elsewhere.

Companies using GA4 as a primary measurement tool and for cross-channel attribution must be mindful of the role of models within the product, and we strongly recommend implementing checks and balances to understand what the unmodeled data looks like so that the effects of the models themselves can be better understood.

One approach is using the BigQuery export, which Google now provides for free for all users of Google Analytics (up to a limited amount), where advertisers can access a raw export of non-modelled data. The BigQuery export is a powerful feature and we’re delighted that it will now be accessible to a larger number of users, where it was previously limited only to users of GA360 (Google’s enterprise offering for Analytics). Another possible approach is to use other analytical or BI tools to validate Google Analytics data. Deciding on the sources of truth, both for modelled and non-modelled data, will be a critical part of every company’s data strategy for the coming years.

Next steps

We at Precis are currently on this journey with many companies in different verticals and stages of maturity. Analytics has always been at the core of what we do, and as we enter this new generation we’re excited to help advertisers to navigate both the benefits and the complexities.

If your company doesn’t yet have a plan for GA4, now is the time to build one. Follow us on socials as explore further considerations for GA4 migration or get in touch if you would like to discuss your plans with the specialists at Precis.