KE Lee · UX Leader

Led UX design for Firebase’s first ML-powered product, enabling developers to predict user behavior and trigger personalized, timely engagement

Background

In 2015, Firebase was evolving beyond backend infrastructure into a growth platform for mobile developers. At the time, there were no accessible tools to help developers apply machine learning to user behavior without deep data science knowledge.

I led UX design for Firebase Predictions, the platform’s first machine learning–based product. Our goal was to help developers anticipate user behavior—like churn, engagement, or purchases—and automate responses using Firebase tools such as Remote Config, Notifications, and In-App Messaging.

Because this was Firebase’s first ML-powered feature, I also defined a set of new UX principles and patterns to address emerging challenges in explainability, trust, and confidence—long before these ideas were standardized. This work laid the foundation for making complex ML capabilities intuitive and actionable for everyday developers.

This case study highlights how I translated early ML capabilities into a practical, intuitive experience that shipped at Google I/O and laid the foundation for ML adoption in Firebase.

Problem

Developers faced multiple barriers to leveraging ML in their apps:

Lack of expertise: Most developers lacked the knowledge or resources to build and train ML models.
Limited tools: Firebase did not support behavioral predictions or dynamic segmentation.
Low trust: Developers were skeptical of “black box” predictions they couldn’t interpret or control.

We needed to bridge this gap with a product that was powerful, transparent, and easy to use—without requiring ML expertise.

Approach

As UX lead, I partnered with PM and engineering from concept to launch. My focus was on simplifying the ML mental model while making the experience intuitive, explainable, and actionable.

Product Framing

Positioned Predictions as “intelligent user segments” developers could apply directly to existing Firebase tools
Simplified the core promise: “Let Firebase help you act on what users are likely to do”
Aligned feature scope around the most valuable prediction types: churn, engagement, and in-app purchase likelihood

UX Design & Flows

Designed end-to-end flows: selecting a prediction type, previewing the segment, and applying actions (e.g., send message, update UI)
Used progressive disclosure and tooltips to explain prediction confidence and segment size
Developed UI that linked predictions directly to developer outcomes (e.g., increased retention)

UX Principles & Pattern Creation

Created foundational UX patterns for ML explainability, confidence indicators, and user control
Established early principles for how to communicate ML logic without overwhelming the user
Informed how the Firebase team (and broader Google teams) would build future ML experiences

Testing & Iteration

Collaborated to run developer usability studies to refine mental models and reduce friction
Iterated on ML explanation language, confidence labels, and UX copy to improve understanding and trust
Collaborated with Firebase Growth and Analytics teams to ensure cohesive experiences across tools

Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.

Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.
Create a Prediction
Developers could easily choose from pre-built behavior types—like churn, conversion, or in-app purchase—and create predictive user segments without writing ML models.
Use a prediction
The dashboard displayed real-time segment size and prediction confidence, helping developers validate impact before applying actions.
Target predicted users
Developers could link predictions directly to Firebase tools like Remote Config or Notifications to automate targeted messages and in-app changes.

Impact

Firebase Predictions launched at Google I/O 2017, bringing ML-driven capabilities to developers without requiring any model training or ML expertise.

Developer Empowerment

Introduced ML-powered segmentation in a visual, transparent format
Enabled developers to act on predicted behavior without writing code
Built trust through clear segment logic, confidence levels, and real-time metrics

Product & Growth Impact

Helped apps reduce churn, increase retention, and personalize experiences at scale
Became a key part of Firebase’s Growth toolkit alongside A/B Testing and Notifications
Drove adoption of Firebase in data- and outcome-driven product teams

Strategic Foundation

Set UX standards for explainability and ML accessibility across Firebase
Influenced future ML and personalization efforts within Firebase and beyond
Demonstrated how thoughtful UX can make ML approachable, trustworthy, and developer-friendly

Led UX design for Firebase’s first ML-powered product, enabling developers to predict user behavior and trigger personalized, timely engagement

Create a Prediction

Use a prediction

Target predicted users

Create a Prediction

Use a prediction

Target predicted users