Creating User Personas

The first step was to leverage our rich dataset to develop comprehensive user personas. I conducted interviews and began consolidating and analyzing data from various sources, including qualitative insights from interviews and surveys, as well as quantitative data from in-app analytics. This process involved collaborating closely with our business team to identify key behavioral patterns and user characteristics.

We employed clustering techniques to group users with similar attributes and behaviors (collaborative filtering). This data-driven approach allowed us to move beyond traditional demographic segmentation and uncover deeper insights into our users' needs, motivations, and pain points.

The result was a set of distinct, well-defined user personas that accurately represented our diverse user base. Each persona included: Demographic information, Behavioral traits, Goals and motivations, Pain points and challenges, Preferred features and interaction patterns. These personas became invaluable tools for our design and product teams, informing decision-making throughout the product development process.

Developing the AI/ML-based Recommendation and Personalization System

With our user personas in hand, we moved on to creating a sophisticated recommendation and personalization system. Our goal was to enhance the user experience by providing tailored content, features, and interactions based on each user's persona and individual behavior.

Working closely with our business and engineering teams, we designed a system that could:

• Classify new users into the appropriate persona category based on their initial interactions

• Continuously learn and adapt to individual user preferences over time

• Provide personalized recommendations for content, features, and products

• Customize the UI/UX based on persona preferences and individual user behavior
  
  

Designing the Recommendation Carousel

While integrating AI into the carousel UI, we needed to address potential ethical concerns, particularly around user autonomy, privacy, and diversity of recommendations. Recommendation systems, like those used by Netflix and Amazon, utilize predictive analytics to suggest content or products based on user data. While enhancing user experience, these systems raise ethical concerns:

• User Autonomy: By consistently presenting personalized recommendations, users might be nudged towards specific choices, potentially limiting their exposure to diverse content and infringing on their ability to make independent decisions.

• Privacy Issues: These systems rely on extensive data collection, including viewing history, search queries, and personal preferences. Without proper safeguards, this data accumulation can lead to unauthorized use or breaches, compromising user privacy.
  
  

For instance, Netflix faced criticism in 2016 for making assumptions about users' preferences based on race, leading to concerns about stereotyping and biased content delivery.

Therefore I designed the carousel with a few key considerations:

Ensuring Fairness & Diversity

To keep workouts fresh and engaging, I designed the recommendation carousel to balance personalized suggestions with diverse class options. The “Surprise Me” card encourages users to try new workouts beyond their usual picks, adding variety to their routine.

To give users more control, I introduced a customizable surprise feature, letting them filter workout categories to fine-tune randomness while still discovering something new. I also added a card-flipping animation for a fun, gamified experience. To ensure fairness, the system suggests three relevant classes plus one completely random pick, keeping recommendations both diverse and personalized.

Protecting User Privacy

In addition to diversity, protecting user privacy and ensuring transparency were key considerations in the design. To help users understand why a particular class was recommended, I included a “Why This Class?” tooltip, which provides context. I introduced a simple toggle switch that allows users to opt out of AI-based recommendations entirely. This gives users direct control over how AI influences their class suggestions while ensuring they can still explore options manually.

Audit, Implementation, and Testing

With the design finalized, I conducted a thorough audit of user flows to identify the best entry points for maximum impact. The home dashboard and search page were identified as high-traffic areas where recommendations could drive the most engagement.

We then integrated the trained model into the app's backend. Lastly, I conducted multi-variable testing to evaluate the performance of the AI-driven recommendations at different entry points. This involved A/B testing to compare the effectiveness of personalized suggestions against a control group.