Improving maternal healthcare with AI.
About this project
Despite advances in technology, maternal mortality rates in the U.S. have dramatically increased from 17.4 deaths per 100,000 live births in 2018 to 32.9 in 2021. There is a clear need for increased accessibility to health care for pregnant mothers. I worked with University of Michigan's school of nursing to create an AI-powered platform to allow nurses to provide 1:1 coaching to clients in need.
Role
Product design lead
Timeline
Jan 2025 - ongoing
Team
4 engineers
2 data scientist
1 faculty
2 sponsors
Skills
Product design
UX Research
Stakeholder communications
Problem & Opportunity: The Crisis of Fear and Inefficiency
Our initial user and secondary research confirmed 2 main barriers to effective maternal healthcare support.
The Stigma Barrier (Mother)
Fear of Judgment: Mothers avoid seeking crucial help due to fear of stigma, causing them to isolate and struggle with fragmented information.
The System Barrier (Nurses)
Workload Crisis: Nurses struggle with managing workload, limiting time and capacity for personalized coaching.
The core challenge
How might we reduce the clients’ stigma barrier and increase nurses’ coaching capacity to improve maternal education and reduce mortality risks?
The solution for nurses
An online maternal health platform with 1:1 coaching, personalized care plans and verified educational materials.
The solution for nurses
AI integration in workflow to reduce adminstative task and increase working capacity.
The solution came with many challenges
Proving the Value of UX
I inherited a project defined by five basic Sigma screens that stakeholders and the computer engineering faculty member believed were sufficient. My first critical task was demonstrating the necessity of a full, end-to-end UX process.
Ambiguous Stakeholder Needs
Our discovery phase was initially hampered because our stakeholders themselves were ambiguous about their core needs and vision. I realized we were spending too much time trying to figure out what they wanted instead of why they wanted it.
Laying out the user journey to identify our opportnities
The Critical Strategic Pivot
The initial plan was to use our AI engine, UM Maizey, for geospatial analysis to find local resources BUT from testing with prompt engineering, we found that Maizey can't reliably work with numbers and is not good with addresses. Based on the user journey map, I identified the necessary pivot to using the LLM for text-heavy care plan.
Design details
Incorporating AI into nurses workflow responsibly.
I referred to IBM’s AI principles as guidelines to develop the care plan creation workflow. The design’s goal to help reduce nurse workload but still making sure the information provided is accurate and personalized.
Design for Mental Models
I used the care plan structure familiar to nurses, and inserted the resources finder in the implementation steps where nurses most likely to need it.
Provide rationales for outputs
The generated care plan links its content with sources such as meeting transcripts, notes, or survey results. User can hover over to see the quote or click on the annotation for the full document.
Design for Co-Creation and Variability
The generated texts are insert into text input boxes that users can click on and make direct edits that auto saves. Users also have the option to clear and regenerate text if needed.
Design for Appropriate Trust & Reliance
Notice of generated care plans are included at the top and bottom of the form.
Balancing different needs
A scheduling flow with structure and autonomy.
Young Vietnamese people are conscious about saving, they use multiple bank accounts, credit cards, and e-wallets to manage their spending, savings, and investments.
Prioritizing the platform’s values
Focusing on the user needs and platform value proposition.
From my competitive analysis, I listed down health trackers and milestones checkers as one of the desired features. The initial version of the client dashboard covered these features BUT considering the web-app