Daily Recommendations
Personalised run and workout recommendations that adapt to users’ fitness levels and routines, with clear rationale behind each suggestion
Role and contribution
Led design for Daily Recommendations across Android, iOS and Pixel Watch from discovery to launch. Defined the design direction, contributed to product strategy and drove delivery, aligning multiple product/ design teams on a sequenced coaching framework.
Contribution
Strategy, Discovery Research, UX/ UI Design, Testing, Delivery
Team
Product managers, Program managers, AI Engineers, Android & iOS engineers, QA, Data scientist, Researchers
Timeline
July 2024 - Sep. 2024
Outcome
- Established the coaching plan framework and launched the feature.
- Improved usage on the content and recommendations.
Key features
-
Onboarding
Allow users to share their training goal, routine, preference, available days and equipment
-
Daily recommendations
Daily personalised run and fitness workouts based on user’s workout history, preference and readiness
-
Execute the Run
Provide further rationale to users so that they understand why this is today’s recommendations
-
feedback loop
Users can provide feedback about the workout to make the personalisation even better
-
Preference
Users can adjust their daily recommendation based on the context so they received timly updates
Full case study details are available upon request.
Below is a breakdown of how we arrived at the solution. ↓
Growing content library, widening discovery gap
Fitbit Premium gives users access to hundreds of workouts and audio sessions from certified trainers and well-known fitness brands. But as the library grew, so did a clear gap.
Three core user problem spaces
Struggle to discover
relevant workouts
Large, overwhelming library makes it hard to find workouts that fit everyday habits, time and training style.
Unsure which content
support goals
Even when users find something they like, it is unclear whether it meaningfully supports their goals.
No connection to
fitness data
The library is disconnected from users’ fitness level and historical data, which breaks trust and reduces perceived value.
Leverage Google AI to build a fitness coach
After Fitbit merged into Google, the organisation aimed to leverage AI to evolve from a static content library into a dynamic and personalised fitness experience that adapts to each user’s fitness level, routine, and preferences.
Grounding the vision in user insights
Partnered with UXR to analyse insights from multiple previous studies. We synthesised these into five key themes that revealed what users need from a personalised fitness experience, which shaped our assumption for the coaching plan.
Humanise the data
Raw numbers don’t motivate. Translate data into meanings users can act on.
Make it personal
Prioritise what matters in the moment. Tailor metrics to context, goals, and routine.
Unveil what’s behind
Expose key drivers behind advanced metrics so users understand cause and effect.
Use familiar anchors
Map metrics to familiar references to clarify scale and effort.
Motivation and support
Celebrate wins and balance activity with recovery.
Goal setting
Set goals and metrics based on fitness level and routine.
Guidance and optimisation
Provide clear guidance on form, timing and intensity.
Adaptability and flexibility
Adapt plans to progress, recovery and changing needs.
Progress assessment
Show progress through clear benchmarks and summaries.
Motivation and support
Celebrate wins and balance activity with recovery.
Designing a five stage coaching journey
With the research insights in place, I created concepts and key user flows to test our assumptions.
We validated which experiences resonated most and learned that users valued a simple five-stage coaching journey:
ONBOARD
Set aspiration & goals
ASSESS
Evaluate fitness levels
PLAN
Create workout and recovery plans
EXECUTE
Follow the plan and stay on track
REFLECT
Measure progress and improve the plan
Ground the experience in behavioural science
Map the user journey to a COM-B behavioural loop that helps users build sustainable habits, adapt over time and progress with the plan.
Focusing the theme 1 for a foundational launch
To deliver this in a scalable way, we created a phased roadmap for the coaching vision. After company restructuring and a reduced team, we needed to focus. We decided to ship the workout recommendation experience with Pixel Watch 3 first. This was the most foundational piece and set up the long-term coaching roadmap.
THEME 1
Workout recommendations
THEME 2
Real-time workout coaching
THEME 3
Workout plan & real-time assessment
THEME 4
Weekly recap & Celebrations
Helping users decide faster with smart recommendations
Provide the top recommendations that allow users to save their time and effort to make their workout decisions
Provide the top recommendation on the primary page that allow users to save their time and effort to make their workout decisions
Previous version
Users are not sure which is the top recommendations.
Current design
To save users time and effort, Lead with one clear, personalised recommendation.
Previous version
Long explanations obscured why the workout was relevant.
Current design
Use a concise LLM generated rationale to explain its purpose and relevance.
Previous version
Users still felt they had to put in effort to choose from four recommendations.
Current design
The updated information hierarchy clearly convey the primary and secondary recommendation
Streamline the rationale to let users understand it at glance
Created a rationale strategy with UX writer that connects each recommendation to the user’s recent workouts at the primary page making its relevance clear and encouraging deeper exploration.
Reduce upfront setup through progressive onboarding
Maintain user engagement by minimising upfront commitment to three questions, then progressively collecting user data through workouts or by prompting users to return to onboarding
Previous version
A lengthy setup increased effort and risked early drop-off.
Current design
Ask three essential questions to unlock a recommendation, then capture additional preferences over time.
Calibrating the journey with a lightweight assessment
Lightweight assessment to help users set the right level to start with the process
Previous version
Users struggle to understand the definition of each option
Current design
With a clear one-line explanation, users can understand the meaning of each level more easily
Leverage AI to build the foundation of the visual system
To create illustrations without dedicated illustration resources, I used Gemini and Midjourney to generate systematic patterns and visual directions. This helped us rapidly explore options, define a coherent visual language, and set the foundation for the final visual system.
Cardio load was central but poorly understood
Coaching plan anchor Cardio load is one of the biggest concerns in the coaching metrics.
When we launched it in 2024, we assumed a daily goal range would give users flexibility and a clear target, but in reality the goal felt confusing and hard to interpret.
Scaling visuals with the Material Design system
Partnered with Material Design system team to further define the visual system for AI generated workouts.
Define a distinctive visual system for AI-recommended workouts
Visual System Anatomy
-
Runner
Runner illustration indicating intensity of run shown in stance of runner
-
Pattern
Pattern indication type of run (Interval sprint,Tempo interval, Tempo run, Easy run)
-
Colour
Colour from warm to cool subtly indicating intensity of run based on intensity (low, medium, high)
Launched Fitbit’s AI recommendation system,
driving a 25.3% uplift in repeat runs
Scaled Fitbit’s AI recommendation system, increasing repeat runs by 25.3%
Launched Fitbit’s first AI driven recommendation system in 2025 and scaled it across sleep and health content, driving a 25.3% uplift in users completing 2+ runs over 6 months.
Core recommendation model