Voice UX & Oracle Health Clinical AI Agent
Project Overview
With the recent news of healthcare giant, Oracle, integrating top LLMs into its future products, I aimed to explore what this might mean in practice. How might this impact consumers? And where can Oracle play to drive the most impact and value for its business?
UX & UI Design, Voice Design
Industry Research
Figma
Healthcare
A personalized AI assistant for diabetes patients
Providing 24/7 personalized assistance such as answering questions, medication reminders, morning and evening check ins, lifestyle guidance and best practices. Continuous interaction allows for real-time data collection, aiding in proactive management and early intervention.
Voice UX
Using simple, conversational language, offering helpful alternatives, confirming intent for sensitive actions, providing clear exit points, and leveraging stored user data with transparent feedback, make interactions relevant and user-friendly.
Research & Analysis
The impact and value question led me to research the most prevalent diseases affecting Americans today:
Chronic Diseases
In the United States, the CDC reports that 6 in 10 adults have at least one chronic disease, and 4 in 10 have two or more. In addition to market size, chronic diseases incur $3.8 trillion in annual healthcare costs in the U.S and economic burden is set to grow. By 2030, chronic diseases are projected to reach $47 trillion by 2030 (World Economic Forum). Furthermore, healthcare systems face many challenges and resource constraints that limit their ability to provide personalized care at scale, making it difficult for patients to follow through on their treatment plans. Often, this results in less than optimal patient outcomes.
Not only was this the biggest market, it had significant opportunity.
Concepts & strategy
In my personal life, I use ChatGPT's advanced voice mode all the time. It is a (pretty) great sounding board for all kinds of projects, use cases and needs. How might this look as a healthcare vertical LLM specifically trained on medical inputs and conversations between doctors and patients?
To explore this space, it was important for me to converge on a specific use case.
A personalized AI assistant for diabetes patients
Providing:
- Personalized Support: Provide 24/7 personalized assistance such as answering questions, providing guidance and helping patients learn about best practices to manage their condition.
- Medication Reminders & Adherence: Conversational reminders can improve medication adherence and lifestyle modifications.
- Data Collection: Continuous interaction allows for real-time data collection, aiding in proactive management and early intervention.
- Morning & Evening Check in: Daily check-in to see how the day went, medication adherence, provide motivation and support, and answer any questions.
- Health record setup (if not connected to an EHR)
In additional to a set of initial features, it was important for me to define a list of UX considerations for voice and chat.
The following items are a list of practices critical to ensure positive interaction that drives adoption, retention and engagement.
- Use short, easily understandable phrases, especially for common interactions. Include details only when prompted.
- Offer helpful alternatives to users. Instead of a generic error messages like, “I didn’t understand." Use “Could you repeat that or ask about another topic?
- Proactively learn and store user information, while providing clear end user feedback.
- Leverage context and stored data to make content more relevant.
- Avoid overly formal language, especially medical jargon. Use conversational language.
- For important tasks involving sensitive data or actions (like payments), confirm user intent before proceeding.
- Provide clear exit points in the experience so users don't feel stuck in the experience.
Success Measures
To test this idea in practice, the north star metric I'd recommend is DAU with secondary metrics of WAU, and retention rate after 30 days.
