Facebook ads and machine learning

In this post you'll learn how to work with machine learning in advertising. For the best performance you should focus on your creative, testing and budget. Your competitive advantage (aside from your product and landing page) is your ability to test and learn.

What does a machine learning algorithm need? It needs data. The more data it ingests, the better it can do its job. Who provides the algorithm with this data? That’s you as an advertiser, the people scrolling on Facebook, and the people buying your product/service. Getting user data has become increasingly challenging with the latest Apple update (iOS 14.5). Apple now limits apps in the App Store to retrieve user data. This also applies to the Facebook app, so  it’s harder to add real-time data to the algorithm. More on how to mitigate this in the Advertising in a New Era chapter.

Facebook’s machine learning is dynamic; it constantly improves based on new signals and increases its performance based on the rules it has received. These rules could be a bid or a demographic that you want to target. The “rules” you set determine ad delivery. No personas or ideal customers are used in this process. The system only looks at data that will help it improve performance. The more positive signals a specific group of users show, the more the algorithm will be skewed toward this group. For example, you might think your audience for your cool mobile app game should be young people in big cities. But when you let the algorithm go its way, you find out the highest conversion, lowest CPA and biggest scale is with women over 60 in suburbs. Thank you algorithm for scaling my product...

This concept requires a different way of looking at and setting up your advertising campaigns. You have to understand the concept of letting the algorithm do the hard work of distributing your ad for you. This means you should not interfere too much with distribution yourself. You should only be thinking about making this automatic distribution work as well as possible. Facebook has almost three billion active users. Let the platform distribute your ad based on its sophisticated models and data. You just worry about delivering a great product, a compelling ad creative, a landing page that converts, and setting an appropriate advertising budget.

To succeed is to provide as much data as possible, to have the least constraints on your ad campaign, to have a high enough budget, and to have a testing roadmap. Testing is key when advertising with Facebook as it is a very dynamic environment. It changes every time a new data point is added, which affects the algorithm. There is no world where you effortlessly “find” the ideal setup, targeting and creative, then just keep pouring money into the Ads Manager.  You have to be constantly changing it up. Sounds like a lot of work, but with a clear testing and learning roadmap, it’s really not that difficult. 


  • Facebook automatically distributes your ad to the right people. 
  • You should focus on your creative, testing and your budget.
  • Your competitive advantage (aside from your product and landing page) is your ability to test. 


Get Exclusive Social Ads Tips

Receive the same tips we used to double our ROAS in just two weeks!