CASE STUDY: SHAPING AI EXPERIENCES
Context
Omada is an app that helps people manage chronic health conditions. The company was exploring how AI could support members in two key areas:
• Nutrition Education: Answering food-related questions with evidence-based guidance
• Motivational Interviewing: Supporting behavior change through conversation
The opportunity
When members had immediate health or nutrition questions, they often waited to hear back from their Omada coach or turned to Google for faster but less reliable answers. With AI, we could give members help in the moment.
Goals
From a metrics standpoint, our goal was to increase engagement and retention. From a product perspective, we aimed to make the AI experiences genuinely useful and grounded in Omada’s clinical expertise, clearly differentiating them from other options in the space.
My role and process
I led content design across both AI experiences, defining how the systems should communicate and, in partnership with clinical leads, how they should behave to deliver the right care.
I partnered with the team on system prompts and evaluation frameworks to assess quality, safety, and brand fit.
I also helped create the in-app entry points for both experiences, which involved mapping flows and writing UI content (headers, value props, etc).
Step 1
I started by grounding both AI experiences in a shared voice. Omada's existing brand voice was based on four characteristics:
I adopted these voice characteristics but defined distinct tone guidance for each experience to serve different user needs.
• Nutrition Education: More casual and upbeat in tone, sometimes light or clever, with direct, structured answers for quick understanding
• Motivational Interviewing: More serious, respectful, and formal in tone, using open-ended questions and gentle prompts to support user-led exploration
This gave us consistency at the brand level, while still allowing each experience to feel unique.
I documented these voice and tone decisions to align cross-functional partners and ultimately translate them into system prompts.
Step 2
I then moved into writing system prompts for both experiences, reviewing outputs to understand how the AI behaved in practice.
The engineering team set up a space where I could add a prompt, quickly test the results, and make adjustments.
This iterative process was key, and the prompt evolved rapidly as I tested and made refinements.
I ran this loop across both Nutrition Education and Motivational Interviewing, using output review as the main way to test and improve the experience.
Step 3
Because clinical safety and brand consistency were extremely important, the team used a second behind-the-scenes LLM to judge the output of the user-facing LLM. To train this judge, I created a rubric for evaluating voice and style, while the clinical team created a similar rubric for clinical quality.
I also partnered closely with clinical and legal teams to design how the AI handled guardrail moments. I created the messaging for these scenarios, ensuring responses were clinically appropriate, legally sound, and still clear and supportive for members.
Step 4
I created the in-app content members saw before starting a session, setting expectations and clearly communicating key value props.
Here is the Nutrition Education experience:
And here is the Motivational Interviewing experience:
Impact
This work helped bring both AI experiences from concept to launch as the company’s first major step into AI ahead of its IPO.
For Nutrition Education, over half of members who logged a meal engaged with the experience, contributing to an overall 4% engagement lift.
Motivational Interviewing was more challenging: only 63% of members who started a chat completed a full session. Unlike Nutrition Education, which was embedded in the food tracker, Motivational Interviewing appeared as a home screen tile, which likely didn’t match the timing or mindset needed for a reflective conversation.
Bonus content
The legal team needed to make it clear that care team members could view chats. The original version (left) framed this in a way that could feel scary or invasive. The revised version (right) reframed it as a benefit, since care team members can follow up and provide more in-depth support.
Hearing from members
It was especially rewarding to see members actively use the features and share their experiences.
