Role

UX Designer and Product Manager
Led a team of three developers and two data analysts
Conducted user research, prototyping, and testing

Empathize

I conducted user interviews and surveys with select hospital clients, uncovering a key issue: “We can’t predict when or where adverse events will happen.” Risk managers were frustrated by the unpredictability of incidents like falls or errors, lacking tools to identify patterns. Through research, I learned users wanted to leverage diverse data—hospital records (e.g., staffing levels, department activity), plus external factors (e.g., time of year, weather)—to build their own AI models. Usage data showed low adoption of existing AI tools due to complexity, and client feedback emphasized a need for clear, customizable insights to enhance patient safety.

Define

Problem Statement: Non-technical healthcare staff need an intuitive interface to build and compare custom AI models for adverse event prediction, because current tools are complex and don’t leverage diverse data effectively.
I developed personas (e.g., “Risk Rita,” a risk manager seeking actionable insights) and user journey maps to pinpoint pain points, like confusion over data inputs and model reliability. Collaborating with Product Management, I set a goal: create a ReactJS dashboard that simplifies data integration (hospital and external) and model-building, empowering users to predict incidents with confidence.

Ideate

I facilitated brainstorming sessions with my team and clients, sketching low-fidelity wireframes to design an AI-driven dashboard. We explored features like:
- A data input panel to select predictors (e.g., staffing, weather, location).
- Trend graphs to visualize incident patterns over time.
- A model comparison tool to evaluate custom models by accuracy.
- Guided prompts to simplify data choices for non-experts.
Client workshops helped prioritize intuitive features, like visual summaries over raw data tables.

Prototype

Using Figma, I crafted interactive prototypes for a dashboard featuring:
- A prediction input panel to choose data (e.g., “Low Staffing: 95% impact,” “Winter: 90% impact”).
- Line graphs showing incident trends (e.g., 2017–2018 data).
- A model comparison section (e.g., Custom Model A: 87% accuracy vs. Model B: 85% accuracy).
- Tooltips to explain terms like “predictor weight.”
Usability testing with risk managers revealed that technical labels confused users, so I simplified terms and added color-coded visuals.

Test

I ran usability tests with our selected client risk managers, observing how they built models and compared predictions. Task success rates hit 88%, with users appreciating the dashboard’s clarity. A/B testing pitted our visual prototype against a table-based version—92% preferred the visual design for its ease. In client trials, users created custom models confidently, sharing feedback like, “I can finally see what drives our incidents.”

Results

Though paused due to company restructuring, the project shone in client trials:
- 88% task completion rate in usability tests, up from initial prototypes.
- 92% preference for visual dashboard in A/B tests.
- Client feedback: “This helps us better understand our risks.”
- One client stopped all operations in their hospitals after 2pm, because they found fatigue to be a factor in adverse events in the OR, with staff and long shifts.

Reflection

This project taught me to bridge complex AI with non-technical needs. By empathizing with users’ struggles to predict adverse events, I designed a tool that gave them control and clarity. Leading cross-functional teams and iterating based on client trials honed my ability to balance functionality with simplicity. Though deployment didn’t happen, the project’s success in trials inspires me to explore AI-driven UX further—perhaps with real-time data prompts to guide users even more.