Changing how billions of phones communicate risk

In early 2020, contact tracing apps could tell people only one thing: exposed or not. Mila, the world's largest academic research center in deep learning, was building COVI, an app whose AI model calculated a graded risk score from symptoms, comorbidities, occupation, and contact patterns. A more precise score is only useful if people understand it, trust it, and act on it. Mila asked us to turn a research instrument into a consumer product with user wellbeing, privacy, and autonomy at its core.

We ran weekly surveys of Canadians throughout the crisis, measuring attitudes toward COVID-19, trust in tracing apps, privacy concerns, and anxiety about reopening. Combined with a review of crisis-messaging evidence, this fed a behavioral map of the specific actions the app needed to support and the barriers to each.

We codified the findings into a five-pillar decision model covering user preference, comprehension, wellbeing, empowerment, and inclusivity, then translated it into the product itself: tailored English and French messaging by region, occupation, and health history, graded risk guidance in place of blunt alerts, and positive reinforcement designed to protect mental health. We carried the model into working sessions with Health Canada and public health leaders.

The work reached beyond one app: it contributed to Apple and Google changing their exposure notification API from binary alerts to multi-level risk, live on billions of phones worldwide. The project's findings on digital contact tracing were published in The Lancet, and the approach was documented in a public white paper covering its design, ethics, and privacy strategy.