Discussion

Singapore hospital AI lead: fix the workflow before reaching for AI

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#13084

Singapore General Hospital’s Adam Chee is urging healthcare organisations to define the problem before adopting AI, arguing that technology can accelerate a broken workflow rather than fix it. His advice is practical: start with the people, need and decision involved, not the model’s sales pitch.

Watch Desk analysis

What happened

Chee, director of SGH’s AI Development Office, spoke at HealthTechX Asia 2026. In an article published by GovInsider, he argues that a promising pilot may fail in clinical practice if it was tested against predictable cases but cannot handle unusual ones. Strong model performance or high usage figures, he says, do not by themselves show that a tool improves care.

Chee recommends asking who a proposed solution is for, what unmet need it addresses, what decision or workflow should change, and how success will be measured. He also distinguishes a model from a product and a complete solution: the latter must fit existing clinical systems and workflows, handle technical failures, and have people and funding for ongoing support.

Why it matters

The advice shifts the test of healthcare AI from “does the model work?” to “does this intervention improve the right part of care, in the setting where it will be used?” That matters when edge cases can expose weaknesses that a controlled pilot did not, and when automation leaves staff fewer chances to intervene.

It also gives organisations a way to challenge vague proposals. If a project promises faster turnaround, teams should identify what currently causes delays and how the proposed change removes that barrier. “We need AI” is not yet a problem statement, however confidently delivered.

Our read

Chee’s case is for better problem definition, not a blanket ban on healthcare AI. His strongest point is that success depends on workflow fit, exception handling and support after launch, as well as model performance. Healthcare teams should be able to explain the need, the intended change and the measure of success before committing scarce development resources.

What to watch

  • Whether hospitals define clinical outcomes and workflow changes before starting AI pilots.
  • How teams plan for unusual cases, system failures and staff intervention after deployment.
  • Whether organisations budget for integration, maintenance, licences and ongoing oversight.

Discussion spark: Should hospitals require an AI proposal to show how it changes a specific clinical workflow before approving a pilot, or would that make useful experimentation too difficult?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.