AI-driven personalization vs. privacy: redesign choices at wearable sleep coach Somni

AI · 6 min read

Somni, a sleep-coaching wearable startup, relied initially on continuous raw-data collection to power its coaching models. Early user interviews flagged privacy concerns: users wanted insights but were wary of constant sharing. The product team split the personalization pipeline into two tiers — local-only personalization that runs on-device and server-side models that require explicit consent — and redesigned the settings flow to make this distinction clear.

The UI change introduced a simple segmented control during onboarding: 'Local insights' or 'Enhanced insights with cloud personalization.' Each option included succinct bullet points about data stored and retention windows. The design also added a one-tap 'privacy snapshot' where users can see what types of data the cloud model would access, and a preview of the additional recommendations they'd receive.

After rollout, Somni saw a 42% opt-in for enhanced personalization within 30 days — higher than expected — while daily active users increased by 9% due to improved trust signals. The case demonstrates that transparent UX and tiered technical design can reconcile the tension between AI personalization and user privacy.