Project Structure
As a team, we selected three themes to investigate for potential features: World-building, Accountability, and Personalized Modules. My objective was to uncover opportunities in these thematic areas where generative AI could elevate the conversational experience, prototype and test new features, collaborate cross-functionally to validate ideas, and ensure the implementations drove increased user engagement, satisfaction, and positive outcomes.
To address these objectives, I spearheaded workshops with key stakeholders including the translational science team, machine learning department, and clinical leadership. We utilized a variety of research and design methods, including affinity mapping, user story mapping, and an impact-effort matrix to generate and prioritize ideas. The insights from the translational science team, with their extensive experience in therapy and behavioral science, were invaluable during the ideation process and in evaluating the feasibility and potential impact of the proposed concepts.
The insights from the translational science team, with their extensive experience in therapy and behavioral science, were invaluable during the ideation process and in evaluating the feasibility and potential impact of the proposed concepts.
The design process followed an iterative approach for each of the three focus areas - world-building, accountability, and personalized modules. I worked closely with the cross-functional teams to develop initial conceptual designs, create clickable prototypes in Figma, and conduct user testing through Dscout to gather feedback on the usability, desirability, and perceived effectiveness of the new features.
This chart illustrates our three-week iterative process for research and prototyping during the Woebot Health project. Our approach was designed to efficiently cover multiple themes and quickly determine which ideas held the most promise.
What We Built
Some of the key design solutions I implemented included:
🗺️ World-Building Enhancements:
Designing conceptual wireframes for a revamped Woebot homepage experience, incorporating generative AI-powered elements to create a more immersive and personalized environment.
Developed concepts for Woebot to adopt various personas (e.g., artist, educator, nature guide) to engage users more meaningfully and offer tailored interactions.
🔔 Accountability and Notification Features:
Creating a behavioral activation module that included personalized follow-up reminders and a lightweight, celebratory feedback loop to encourage users to take real-world actions.
Leveraging generative AI to craft empathetic and motivating notification messages to support users in achieving their goals.
🕸️ Personalized Modules:
Collaborating with the translational science team to develop information-rich user profiles that could be dynamically updated based on preferences and behaviors.
Prototyping adaptive modules that could suggest relevant content, exercises, or interventions tailored to each user's unique needs and progress.
Example Flow: Behavioral Activation
One path we explored in the 'accountability' theme is a method of psychological intervention called Behavioral Activation, which is similar to exposure therapy.
With support from the head of translational science, I considered how to structure a conversation so that we could get buy-in from our user. How can we structure a chatbot conversation to encourage users to step out of their comfort zones and participate in new experiments? How can we imagine an experience to encourage users to take actual, real-world actions, motivated solely by a conversational experience in the Woebot app?
.png)
This pathway outlines how Woebot aims to encourage real-world actions through emotional tracking and personalized engagement, setting the stage for future integration with generative AI.
Challenge: 'My Happy Place'
One notable challenge I encountered was with the "My Happy Place" feature under the 'World-building' theme, where users could describe their personal happy place and have it visualized through generative AI-created images. User testing revealed that these representations often fell into an "uncanny valley," failing to accurately capture the users' precious and personal mental images. To address this, I conducted additional research and user testing to better understand the nuances of how people conceptualize and relate to their happy places. The findings showed that the visual representation of these personal spaces was highly sensitive and could not be adequately captured by the current state of generative AI, leading me to deprioritize this feature and focus on other areas where the technology could be more effectively and thoughtfully integrated.
Impact
Despite the challenges faced, the implementation of generative AI-powered features in Woebot Health led to noticeable improvements in user engagement and satisfaction. The new world-building and personalized modules created a more immersive and tailored experience, which was well-received by users. In user testing, 75% of participants responded favorably to generative AI-powered notifications that were personalized based on their rich profiles and recent activities. These AI-generated notifications were rated 40% more effective in driving user engagement compared to standard, non-personalized notifications.
The key lessons I learned throughout this project included the importance of close collaboration with cross-functional teams, particularly the translational science experts, to ensure design solutions are grounded in behavioral science and clinical best practices. I also recognized the need for iterative testing and feedback loops when incorporating emerging technologies like generative AI, to validate the feasibility and desirability of the proposed features.
Moving forward, I believe there is significant potential to further integrate more advanced generative AI models to create even more personalized and adaptive conversational experiences for Woebot users. This could include the development of personalized mental health interventions, the generation of empathetic and contextual responses, and the exploration of multimodal interactions. However, continued research and testing on the boundaries and appropriate use cases for generative AI in mental health applications will be crucial to ensure a responsible and ethical implementation of these technologies.
Much of the work involved in this project is confidential, but I’m happy to provide further details during a live discussion.
