PRISM-Coach's privacy-by-design AI boosts adherence 94% in coaching
New system achieves 5.2 kg avg weight loss without leaking personal data
PRISM-Coach tackles a core tension in digital wellness: personalizing peer support over time while preventing personally identifiable information (PII) and sensitive health data from leaking into analytics and AI pipelines. The architecture splits each user into four bounded views—Identity, Operational, Learning, and Coaching—each with distinct access controls and risk profiles. It then uses vault-based controlled identity restoration and a privacy-constrained contextual bandit to assign users to eligible peer groups under coach-capacity and stability constraints. A human-in-the-loop coaching assistant generates de-identified summaries and draft messages without sending raw PII or PHI to external AI services.
The system was deployed in a commercial lifestyle coaching platform and evaluated over three years with telemetry from approximately 2,800 users. At the population level, daily check-in adherence increased from 0.35 to 0.68, and engagement rose to 1.35 times baseline. In a matched 19-week comparison, the AI-enabled workflow achieved adherence of 0.74 versus 0.48 under static grouping and higher average weight loss: 5.2 kg versus 3.1 kg. Survey results show 82% reported positive perceived benefit, and 92% reported increased privacy confidence after transparency disclosures. PRISM-Coach offers a practical blueprint for scaling personalized wellness programs without sacrificing user privacy.
- Adherence jumped from 0.35 to 0.68 daily check-ins, a 94% increase over baseline.
- Users lost 5.2 kg on average vs 3.1 kg with static grouping (68% more weight loss).
- 92% of users reported higher privacy confidence after transparency disclosures.
Why It Matters
Enables scalable, personalized wellness coaching without compromising user privacy or exposing sensitive health data.