Predictive analytics in GA4
One of the major selling points for Google Analytics 4, when it was released in late 2020, was the inclusion of predictive analytics metrics, a Big Data feature that until then had largely been the domain of enterprise analytics packages.
Now that Google has announced that GA4 will replace Universal Analytics for all users by July 1, 2023, it’s worth a second look at how GA4’s predictive analytics can impact your business.
What is "Predictive Analytics", anyway?
For business users, all analytics are more or less "predictive," at least in practice. Suppose you see that an audience segment responded well to a discount offer for sci-fi movie posters. In that case, you're going to re-target that audience with offers for sci-fi-themed coffee mugs, sweatshirts, baseball caps – any inventory item that you can predict (infer, really) the audience will like, based on explicit past behavior.
It's just common sense.
Predictive analytics radically expands on this principle by using machine learning and other Big Data technologies to incorporate thousands of direct and indirect behaviors into the math that determines who is likely to engage in a desired activity.
For example, if users who exit your site through a specific path and who have also read two blog posts tend to respond to mid-funnel CTAs, you'll skip the high funnel in your email campaigns and get right to business with that audience.
And the best part is that you don't have to find the predictive patterns yourself. With machine learning, analytics systems can now scan the entirety of your users' behavioral data and map them to key events – purchases, whitepaper downloads, subscriptions – you identify as the endpoints of valuable customer engagement. Based on these multi-factor patterns, you can build audience segments and marketing campaigns that anticipate users' next steps in their customer journeys.
Enterprise marketers build massive databases of first-party and second-party user data as the basis for predictive analytics. They hone their campaigns to an incredibly fine point – think the color of a car depicted in a display banner ad. PA can also be extremely useful in predicting broad business trends, like quarterly revenue or newsletter subscription growth.
What does this mean for my business?
Honestly? Maybe nothing at the moment. When we talk about these enterprise marketers, we're talking about a selective group with a fairly high threshold of sales required to use this new feature of Google Analytics. It's possible that your business doesn't qualify for this function of GA4... yet. But it's still an interesting exploration into where Google believes the future of analytics is heading.
As this technology continues to be developed and adopted, you are likely to see more applications in businesses of all sizes. It's also worth understanding more of this process because chances are if you are using Google Analytics, your website's anonymized data is part of the massive dataset Google is using to build this technology.
Even if you don't reach the threshold for predictive analytics, the basic events they are built on can be incredibly useful and things you should be tracking for your website. If you do meet the threshold, then this is a powerful, free tool to take advantage of.
The basics of Predictive Analytics in GA4
GA4 includes three core predictive analytics metrics built largely around eCommerce behaviors.
These are, per Google's documentation:
The probability that a user who was active in the last 28 days will log a specific conversion event within the next 7 days.
The probability that a user who was active on your app or site within the last 7 days will not be active within the next 7 days.
The revenue expected from all purchase conversions within the next 28 days from a user who was active in the last 28 days.
These metrics can be used to build audiences or create custom reports in GA4s Explorations toolset.
GA4's three predictive models are based on two standard events: purchase and in_app_purchase. To be eligible for the predictive analytics, your property must send these two events, as well as collect the value and currency parameters for the purchase event.
It's always advisable to implement numerous events across your site that reflect user behaviors you want to quantify in your reporting. Of course, Google is never entirely transparent about its algorithms, but it's a safe bet that events weigh heavily in its predictive models. See our overview of GA4 events for more info about how to gather this data.
Other prerequisites for GA4 predictive analytics
In addition to having these events implemented, your site must generate enough data to credibly support an analysis model. Google has set that threshold at 1,000 users in the last 28 days to support both a positive and negative test for the core criteria. In other words, to support the purchase predictive metric, 1,000 returning visitors must make a purchase in the last four weeks, and 1,000 returning visitors must not buy anything.
Google also has issued general guidance about sustaining "model quality" over time, which loosely translates to "follow our prescribed best practices or we will cut you off." These include collecting as many recommended events as possible on your site and, most notably, turning on the benchmarking feature in your data-sharing settings.
Two key points here:
- Google wants all the data. Data is the lifeblood of Google's advertising business. Gathering data is why it gives away a robust platform like Google Analytics in the first place.
- In this case, you get a pretty good return for sharing your data. As I've said earlier, predictive analytics is based on massive volumes of data. By sharing your first-party data with Google through benchmarking, your analytics are dramatically enhanced by including behaviors of anonymized users on millions of other sites who "look like" your first-party visitors. This is effectively the same tech that Google uses to build audiences for its own advertising businesses.
I can't stress this enough – enterprise marketers pay huge dollars for this kind of data augmentation, and Google is offering it in trade to GA4 users, if only for three core metrics.
Unless you are subject to serious privacy regulations:
- Go to the GA4 Admin center
- Click the account you want to edit
- Select Benchmarking
And save your changes.
How to use Predictive Analytics to grow revenue
Now that you have a way to credibly predict future user behavior, how do you use these insights to fuel your business?
Within GA4, predictive is used for two primary purposes.
Building audiences for Google Ads campaigns
Core predictive metrics can be used as a filter on audience segments such as likely to abandon the site or likely to purchase within a week. Thresholds can be set on the predictive filters to create more broad or precise audiences, depending on your needs.
You can modify a suggested predictive audience by following these steps:
- Navigate to Configure > Audience > New Audience
- Click Predictive under Suggested audiences
- Choose an audience that is "ready" – i.e., meets Google's data volume and quality criteria
- Modify the template using GA4s Audience Builder (see our primer on these features here.)
You can also use predictive as a condition in custom audiences.
To edit a predictive condition, select one of the options next to it: Most likely, Least likely, or a custom range. The summary chart to the right will show you the size of the audience that meets your predictive criteria.
This walkthrough at Optimize Smart gives several examples of how predictive can be used to create audiences for retargeting. For example, you can pair Churn probability with Lifetime Customer Value (LTV) to find users that have bought from you but exhibit behaviors that indicate they are likely to drop from your site. These customers should be front-of-queue for special promotions to keep and build customer loyalty or even for customer satisfaction surveys.
Push events, including messages from Firebase, can be triggered when GA4 adds a user to one of your predictive audiences.
Predictive can also be added to your LTV report and other Explorations for detailed analysis of user behavior, including the value of audiences by acquisition source or other user attributes.
By default, predictive is not one of the available metrics families, so you will need to add it under Metrics in the Variable tab. Each of the available predictions will ask you to set percentage thresholds at which users are added to the prime metric category, similar to the Audience Builder.
Understanding the attributes of audiences who are likely to visit or leave your site over a period of time can be enormously helpful in planning promotions, sales, and even content refresh cycles.
A glimpse of the future with GA4 predictive analytics
Big Data technology has opened the door to incredible levels of insight into patterns that were previously invisible to human site managers looking at simple reports. Google Analytics 4's built-in predictive filters help increase ROI on your marketing dollars by letting you target users who are likely to buy or respond. They also help you manage your site based on probable user behavior, not only for current customers but those you have not yet engaged.
Predictive, when coupled with expanded Events and Exploration reporting, is a key way to get the most from your transition to GA4.