Google Launches Firebase Predictions for User Segmentation Using Machine Learning

Google Launches Firebase Predictions for User Segmentation Using Machine Learning

Today in Amsterdam at the 2017 Firebase Dev Summit, Google launched Firebase Predictions, their attempt at helping you predict what your users are going to do, before your users actually do it.

Trying to figure out how to target promotions within your app has historically been an absolute pain, up until now requiring manual segmentation of your userbase in Firebase in order to decide who to send each promotion to. It required you to attempt to figure out what your users were likely to do with the limited information that you had available to you. With Firebase Predictions, Google is aiming to permanently change the way this works on all platforms that Firebase is with.

As with many of Google’s products that have been integrated with machine learning recently, Firebase Predictions uses Google’s custom built hardware and cutting edge software to attempt to take the raw data that we receive, and turn it into something useful and useable. Firebase Predictions currently works by looking at all the small things that users do to attempt to sort them into one of four categories:

  • Users who are predicted to churn in the next 7 days
  • Users who are predicted to stay engaged with your app
  • Users who are predicted to spend money
  • Users who are predicted to not spend money in the next 7 days

Using that segmentation, you can use Remote Config to target promotions or ads to specific subsets of your userbase that you are interested in. This will reduce the number of people who see your promotions that you are not interested in showing the promotions to (avoiding any annoyance that might come with showing an ad to the wrong person), and help ensure that the promotions get to the right users.

Firebase Predictions Dashboard

Firebase Predictions will also allow you to target for any custom Analytics conversion event you want, allowing you to segment based on users that seem likely or unlikely to hit that target, and send different in-app promotions with Remote Config to each group. One example that Google gave for a game would be to target it based on reaching a specific level in your app if you find that people who reach that level are likely to stay with your game long term. Once you define that custom Analytics conversion event, Google will use machine learning to try to predict which users are likely to hit that target, and which ones are likely to fall short, and allow you to aim your in-app promotions accordingly.

This can have a major impact on user retention, as a well timed promotion to the users that are likely to churn in the next 7 days can help turn them into users who are likely to stay engaged with your app. One of Google’s launch partners for this feature, Halfbrick Studios, found that by targeting their promotions, they saw a 20% increase in their 7 day retention rate. 20% may not sound like much out of context, but that is a substantial uptick in long term users, and is great to see. It should be interesting to see the affect it will have in the future as more developers get their hands on this tool, and have a chance to test it out for themselves.

For more information about Firebase Predictions, check out the Firebase Predictions webpage later today when it goes live, or check out the Youtube Video down below which talks a bit more about it.

Will you be using Firebase Predictions in your app? What do you think of the application of machine learning to user segmentation? Have you been successful with user segmentation in the past? Let us know in the comments!

Discuss This Story

READ THIS NEXT