Google’s Universal App Campaigns (UAC) Uses Machine Learning to Drive Quality Users to Your App

You’ve built a great app… but how do you drive quality adopters to your download page?

Google may have the answer in their new Machine Learning powered Universal App Campaign (UAC). Read on to learn more…

High-Quality Users

A high-quality user is someone that interacts regularly with your app. A high-quality user is someone who uses your app to drive revenue via subscriptions, in-app purchases, ad revenue, or any other way your app generates revenue. A low-quality user is someone who installs your app and either never uses your app—or worse—uninstalls it.

What about someone that users your app regularly but does not drive any revenue? For example, someone who regularly uses your app but never subscribes, purchases anything, or clicks on any ads? This person would be described as a “potential high-quality” user if any of his interactions, or “proxy actions,” lead to a “success action” such as subscribing, purchasing something, or clicking on an ad. This person may also be a potential high-quality user if he regularly interacts with your app and your revenue model is predicated on an acquisition for users (think about free apps such as WhatsApp, Instagram, Twitter, etc.)

Let’s turn to the brick-and-mortar world for a minute to make a comparison of high-quality users. Suppose Betty Goodeal’s Auto Dealership attracts 100 people on any given day. Of those 100, five people purchase cars that day. Clearly, they are high-quality users because they have completed a success action. Also, of the 100 people, 20 test drive cars. The test drive would be considered a proxy action with the potential to lead to a success action (purchasing a car). Hence, these 20 users are considered potentially high-quality users.


Machine Learning and Customer Acquisition

Machine learning is the branch of artificial intelligence concerned with finding patterns in data. Often referred to as “applied statistics,” machine learning uses large amounts of data to find predictable patterns. Machine learning is the mainstay of banks like Chase, financial institutions like Goldman, retailers like Walmart, entertainment companies like Netflix, and juggernauts like Amazon. Machine learning has exploded in use because of one simple fact: It works.

For a review on machine learning, check out some of my other blogs, including: Machine Learning 101

One of the most common and easy to understand applications of machine learning is recommendation engines. A recommendation engine seeks to recommend products or services based on commonalities in other products or services, commonalities in other people who have used the same product or service, or a hybrid approach, which is the most common because its ensemble of several approaches is far more powerful and accurate than any single approach.

For example, if I like Star Wars, and someone else who also liked Star Wars also likes Star Trek, then I may also like Star Trek. Similarly, the revers is also often true: if I like Star Wars, and a movie like Star Trek is similar to Star Wars, I may also like Star Trek. This is an oversimplification, but the general theme maintains; the actual application is performed on dozens of dimensions using advanced mathematics and statistics.

For a review on recommendation engines in machine learning, check out: Recommendation Engines

Employing to this same method of recommendation, Google’s Universal App Campaigns (UAC) drives users to your app based on similarities in user demographics and behavior. For example, Google UAC would drive a fitness enthusiast to your fitness app. Google knows which of their users are fitness enthusiasts based on the massive amounts of personal data collected daily including (but certainly not limited to) the videos they watch, the articles they read, the sites they search for, the sites they visit, and even the physical locations they visit.

Let’s turn back to the brick-and-mortar example and revisit Betty Goodeal’s Auto Dealership. Betty knows that her customers are fitness enthusiasts. So, she hands out fliers at Jim’s Gym to drive traffic to her dealership. Her daily traffic increases from an average of 100 customers per day to 200 customers per day—that’s a 100% increase. Great! But her high-quality users that perform a success action (purchase a vehicle) only increases 20% to 6 people per day and her potential high-quality users that perform a proxy action (test drive a car) also only increases 20% to 24 per day. Why?

Using high-level data correlation, like identifying fitness enthusiasts, is certainly a better approach than throwing mud at the wall, but it’s not granular enough. What if Betty could identify the driving behavior and habits of the customers who purchased vehicles from her? For example, what if she knew that there was a commonality among the people who actually purchased her vehicles that included: going on a road-trip two times per year in July and November; driving to outdoor activity venues on the weekends; driving to work in the early morning via surface streets more than highways; watching Netflix at night; listening to Spotify during their workouts; and reading books on the topic of finance. Imagine how much more targeted her advertising would be. Imagine how much more her return on ad spend (ROAS) would be? A lot.


Success Actions, Proxy Actions and Machine Learning, and Customer Acquisition

With the earlier examples in mind, let’s see how Google’s UAC works.  While Google UAC can certainly drive people to download your app who are similar to people who have already downloaded your app, there is not guarantee that the new users will be high-quality users. In order to truly determine a high-quality user, we need to identify users beyond simple the install criteria (or simply the “buys a car” criteria) to the actions the user performs while using the app. In other words, by drilling down in the granular actions and identifying and recommending like-minded people based on specific actions your users perform in the app versus simply installing the app, Google’s UAC can drive high-quality users to your app.

For example, suppose we identify a success action in your app as a purchase of $10 or more. Then, Google UAC will identify and drive new users based on the demographics and behaviors of the users who completed the success action of a purchase of $10 or more. You can also drive users based on proxy actions. Example, if a user adds more than three items to a cart, even if they do not actually check out, that could be considered a proxy action to a purchase. Google’s UAC would then drive users based on the commonalities of the users who successfully completed a proxy action. The beauty is, you don’t have to do a thing. This is where the magic of machine learning comes into play, driving quality users to your app.


Getting Started

Here are a few links to get you started.

First, check out Google UAC here:

Tracking success actions and proxy actions in your app can be accomplished with Google Analytics for Firebase, which can be downloaded and installed here:

SerpicoDEV can help you install Google Analytics for Firebase and work together with you to identify success actions and proxy action in your app. We can further work with your success-team to set the right actions to drive quality users to download your app and be active daily users.  Contact SerpicoDEV