AI Defaults at Scale: Why NovaHealth Automated Medication Reminders and How UX Handled Consent

AI · 5 min read

NovaHealth, a telemedicine startup, launched an AI defaults feature that auto-suggests reminder schedules and dosing cues for patients with chronic prescriptions. The engineering team used a hybrid model combining on-device personalization embeddings with server-side clinical rules, while UX focused on how to present an algorithmic default without eroding autonomy.

Designers explored three patterns: opt-in suggestions, passive defaults with clear modals, and interactive 'explain & accept' flows. They settled on an inline suggestion card that included the predicted schedule, a short plain-language explanation of the model’s reasoning, and two clear actions — accept or customize. Privacy was handled by running personalization locally where possible and surfacing a compact privacy note using progressive disclosure.

In a four-week pilot, adherence improved by 14% among users who accepted defaults; only 8% opted out after seeing the explanation. The team logged edge cases where defaults conflicted with patient preferences, prompting a tweak that allowed quick manual overrides and surfaced an alternative suggested by a clinician. The article closes with guidelines for designers: test defaults with real tasks, make undo frictionless, and co-design explanations with clinical stakeholders.