AI-Guided Patient Onboarding

The Challenge

A digital health platform was struggling with a patient onboarding process that had grown organically over five years into a 47-step nightmare. Patients were abandoning mid-process at a 38% rate. Those who completed it took an average of 3 days — not because the process required 3 days of work, but because it was so confusing that patients kept stopping and restarting.

The process was also one-size-fits-all: a 25-year-old with no chronic conditions went through the same flow as a 70-year-old managing multiple medications. The irrelevant steps frustrated patients and eroded trust before they'd even used the product.

The Solution

We built an adaptive onboarding system that uses a lightweight ML model to predict which steps are relevant for each patient based on their initial responses. A patient who indicates they're managing a chronic condition sees a different flow than one who's using the platform for preventive care. The system learns from completion patterns to continuously improve its predictions.

The UX was redesigned around progressive disclosure — showing patients only what they need to see right now, with clear progress indicators and the ability to save and resume at any point. We integrated with the platform's existing EHR connections to pre-populate fields where possible, eliminating redundant data entry.

HIPAA compliance was a core constraint throughout. All ML inference happened on-device for sensitive health data, with only anonymized behavioral signals sent to the training pipeline. We worked closely with the client's compliance team to ensure every architectural decision met their regulatory requirements.

Tech Stack