SickKids and SKAI:in-house AI capability, not vendor dependency
What they're doing
SickKids launched the SickKids Artificial Intelligence (SKAI) program in 2025 as a flagship of its 2030 Strategy. SKAI runs three things in parallel: an in-house team that develops and deploys clinical AI tools, a research arm where scientists use AI to accelerate discovery, and a community-building arm focused on AI literacy across the org.
The output is already concrete. Co-led by Dr. Devin Singh and Dr. Anna Goldenberg, the SKAI Service has built models that identify which diagnostic imaging and lab tests a patient needs before a physician sees them. The model is expected to expedite care for over 22% of ED patients and save two to three hours per visit. A cardiac arrhythmia detection model is in silent trial at SickKids and is moving to CHU Sainte-Justine. A hydronephrosis screening model is also in development.
Why it matters
The structural decision SickKids made is the interesting one. Most hospitals buy AI from vendors. SickKids built the muscle to develop, validate, deploy, and govern AI in-house, then complement that capability with selective partnerships. That's a different bet about where the value sits over the next decade.
What's working
- Named, senior leadership: Robert Greer as technical director; Devin Singh and Anna Goldenberg as co-leads of the SKAI Service
- "Silent trial" methodology where models run alongside care without influencing it, building confidence before deployment
- A literacy stream so the org's culture moves in parallel with the technology
- Concrete, measured outcomes (22% of ED patients, 2-3 hours per visit) that the board and the public can see
What's harder than it looks
- In-house capability is expensive. SKAI's headcount and tooling burn is not affordable for most hospitals
- The hand-off from research model to clinical deployment is where most AI projects die. SKAI's solution is the silent trial pattern, but it's slow
- Pediatric AI carries additional ethical weight. SKAI's literacy work is partly insurance against future scrutiny
Three takeaways for similar orgs
- 01Name a senior leader. Without a Robert Greer figure, AI work scatters across teams and stalls
- 02Silent trials beat pilots. Run the model alongside care for months before it touches a decision
- 03Match the technology pace with a literacy pace. If your tools move faster than your people, the deployment fails on humans, not on math