At Google IO 2018, Google introduced a tool to make machine learning with your app analytics easier. The product, Firebase Predictions, uses your data in Firebase Analytics to perform classification of users. Out of the box, Predictions will classify users into groups based on their likelihood to stop using the app (termed churn) or to make a purchase. Using these groups you may send push notifications, adjust the app using remote configuration, or perform A/B testing via pre-existing Firebase services. Predictions also provides a way for you to define custom ways to classify your users. Predictions is currently a free bundled component of Firebase which makes the product extremely compelling to app developers.
CapTech offers the following recommendations or considerations when thinking about leveraging Firebase Predictions.
Good Entry Level Tools That Can Reduce Time to Value
While somewhat restricted, Firebase Predictions may be a good tool to use to get your first exposure to machine learning. With machine learning, it can seem like a very steep adoption curve with the first value coming long into the project. Predictions will reduce that time if you're already using Firebase, and it may shorten the time to value even if you're not already a Firebase customer. A tool like this can show early value and provide the evidence to management to fund further, more bespoke, machine learning projects.
Beware of Lock-in
Firebase Predictions uses data within Firebase Analytics. Firebase Analytics is targeted more at independent developers rather than enterprise organizations. Therefore, the reporting and export features of Firebase are more limited than standalone analytics packages. Once your data goes into Firebase Analytics you can extract it to Google's BigQuery service for more in-depth reporting. You can increase your flexibility by inserting an abstraction layer between your view tier code and analytics calls. This abstraction layer should route analytics events to multiple analytics providers. For example it would route them to Firebase and to your internal Site Catalyst services. Having an abstraction layer between your business logic and your analytics provider is a good idea anyway; it gives you the flexibility to change providers with little impact on the app code.
Requires Use Of Other Firebase tools
Firebase Predictions only apparent means of output is creating user groups in Firebase. Those user groups are useful for sending notifications or remotely controlling the app configuration. If your app is not using Firebase remote config or notifications the groups created by Predictions will not be very useful.
Everybody Needs to Understand ML
One clarion call from the Google IO keynotes is that everybody needs to understand machine learning to some extent. It is starting to touch many aspect of web and mobile apps. You don't need to be a Ph.D. in statistics and multivariant calculus (unless you're a data scientist doing machine learning algorithm research) to leverage machine learning. But you do need to understand the capabilities of machine learning types. You also need to develop methods to manage the probabilistic results from machine learning algorithms.
Anyone Can Do This With Their Analytics Data
With the proper tools almost any company can begin to leverage machine learning. You can build world class machine learning pipelines with the same open source software used by the silicon valley giants. Data governance is key to exploiting machine learning. You'll need to have a grasp on the data you have, and don't have, and the characteristics of that data.
It seemed like machine learning was the major theme of Google IO this year. Everything from new hardware especially designed for machine learning to Firebase Predictions and ML Kit all provide the capability for you to exploit machine learning. CapTech believes you can get the greatest value from machine learning by tuning machine learning solutions to your specific business problems.