feature

Every brand wants their customers to remain customers. With Predictive Churn, Braze customers can leverage Machine Learning to identify individuals who are at risk of abandoning their app or website. Let's dive in.

When customers don't feel heard, they leave. But while no brand wants its customers to walk out the door, finding ways to identify which users are at risk and taking steps to rebuild those relationships can be a major challenge. To make it happen, you need both genuine insight into your user base and the tools and tactics it takes to speak to them effectively.

But just because something is difficult doesn’t mean it’s not worthwhile—and, in this case, understanding which users are at risk of churning can be transformational to your business. That’s why Braze is pleased to announce the official release of Predictive Churn, the first feature in the new Braze Predictive Suite.

What Is Predictive Churn?

Braze Predictive Churn leverages machine learning (ML) to allow brands to identify key behaviors correlated to user churn, giving marketing, growth, and engagement teams an unprecedented look at otherwise hidden customer retention warning signs. This feature allows brands to determine for themselves what constitutes churn when it comes to their specific business, then leverage per-user churn propensity scores and actionable churn-related insights to take steps to boost retention and reduce customer disengagement.

With this new feature, brands can input their own definitions of customers churn (e.g. has not used app in more than three days or has not made a purchase in more than 14 days) and define their target group—for instance, first carried out a specific custom event more than 30 days ago or first used app more than seven days go. Once that information is in place, Braze will create a ML model that can detect behavioral patterns correlated with churn and use those correlations to predict the likelihood that targeted users will depart.

Why Braze Decided to Build Predictive Churn

Customer retention is an essential part of modern marketing. If you want to make the most of your customer engagement efforts, you need to ensure as many of your customers stick around as possible. But you need the tools to do that effectively—and those tools can require a lot of time, effort, and knowledge to build.

It’ll probably come as no surprise to you that successfully implementing machine learning to support this kind of project is no easy task. Defining clear modeling objectives and then gathering, preprocessing, flattening, and cleaning the data—not to mention working out the kinks through repeated testing—takes many months of time and work from a team of data scientists. Beyond that, layering in the kinds of sophisticated algorithms (such as decision trees or deep learning) that this kind of feature requires takes even greater expertise. Now imagine that you needed to incorporate all the infrastructure engineering needed to serve up ML-powered predictions for hundreds of millions of users.

But while retention is a concern for all companies, building the kind of tools needed to optimize customer retention isn’t a core competency—or a good use of resources—for most brands. However, this kind of project is a core competency for Braze. That’s why we have spent so much time and effort to bring these capabilities to our customers through the launch of our new Predictive Suite.

And that work has paid off: By streamlining and automating all the complexity that comes with identifying at-risk users, Braze customers can now take all the steps necessary to deploy a Predictive Churn model in as little as 51 seconds. (Yes, a first-time user really did this—and, yes, we timed it!) Then, within twenty minutes, ML predictions will be available, allowing brands to target millions of users based on their churn risk. Even better? These Churn Risk Scores are updated with fresh user data on a schedule that you can set, ensuring that you can keep leveraging these insights going forward.

How Accurate Is Predictive Churn?

No predictive model is perfect; that’s why each Predictive Churn model showcases a Prediction Quality rating, allowing you to properly assess the value and expected ROI associated with usage according to your business rules. That rating looks at how effective each model is when tested on historical data—so, for instance, a Prediction Quality score of 0% means that the model does no better than a random guess when it comes to predicting who will churn, while anything above 0 means the algorithm is correctly identifying some users. While 100% accuracy is not possible in practice, the closer to 100%, the better!

How Brands Can Leverage Predictive Churn Effectively

The purpose of Predictive Churn doesn’t end when you receive your predictions—you have to understand and act on those predictions to see this feature’s full value. To help marketing, growth, and engagement teams use this feature effectively, let’s look at a few common use cases for Predictive Churn:

  • Categorize Users by Churn Risk: With Predictive Churn, brands can assess how many of their users are at high, medium, and low risk of churn according to a given model and investigate which behaviors are associated with increased churn risk, allowing them to optimize their overall messaging strategy to minimize the number of lapsing users.
  • Save a Churn Risk-Based Segment: Marketers can segment their audience using insights from Predictive Churn, allowing them to target “high-risk” users (or other key groups) in an ongoing way directly from the Braze platform’s Churn dashboard or via filters contained within the Braze Audience Builder feature.
  • Build Targeted Campaigns and Canvases: Once a brand has used Predictive Churn, they can use the segments and filters that the feature supports to message effectively to at-risk members of your audience—whether that’s one-off, highly targeted win-back messages or long-term drip campaigns aimed at preventing churn with new benefits or incentives.

Final Thoughts

Every brand wants their customers to remain customers. With Predictive Churn, Braze customers can leverage ML to identify individuals who are at risk of abandoning their app or website, making it possible to hold onto those customers longer, increase lifetime value, and avoid excessive messaging to customers who don’t require that extra nudge—all without requiring any of your own data science resources.

Interested in checking out Predictive Churn? Braze customers who are eligible for the new Predictive Suite can now build a free trial Prediction using Predictive Churn in order to showcase how this new feature works. (However, this “Preview Prediction” cannot be used to target users for messaging and will not be regularly updated; it is intended only to allow Braze customers see the overall accuracy of the predictive model.)

To gain access to Predictive Churn, please contact your Braze Account Manager or Customer Success Manager. You can learn more about the Braze platform’s full suite of AI/ML functionality, and where we’re taking that here.

Boris Revechkis and Ashley Christiano