A hospital's competitive edge in AI will depend less on which model it chooses than on how it runs the technology in daily clinical work, according to Lee Dae-wook, chief strategy, marketing and operations officer at GE HealthCare Korea. Speaking at the K-Digital Healthcare Summit at COEX in Seoul on Thursday, Lee said two hospitals using the same AI model, vendor and contract can see very different results after deployment.
"The key battleground in hospital AI today is not which model you use, but how you operate it," Lee said. "We can see two hospitals with the same model, vendor and contract achieve very different results after deploying the same technology. The difference comes not simply from the performance of the algorithm, but from how hospitals use the infrastructure and resources they already have in place."
Outcomes also vary by the complexity of care, patient volume, and how consistently clinicians actually use the AI systems, he added. "The focus is shifting from technology to people."
From AI-attached to AI-native
Lee argued hospitals must move from being "AI-attached," where new tools are layered on top of existing processes and add more accounts and logins, to becoming "AI-native," where workflows and staffing are redesigned around AI from the start. "What matters is not simply introducing technology and having end users use it, but reorganizing work within the hospital based on that technology," he said.
That includes reallocating staff and rethinking how departments interact. "Hospitals need to discuss how to improve operational efficiency based on the technologies they introduce, including reallocating staff. That is how they move closer to becoming AI-native hospitals."
For operations professionals, this means AI implementation is an operations redesign project, not an IT purchase. The same principle applies in other industries: adding AI tools to unchanged workflows rarely delivers the expected gains. AI for Operations training typically emphasizes process redesign and workflow integration, which matches Lee's argument that the technology itself is only one layer of the system.
Five layers of a hospital AI operating system
Lee divided a hospital AI operating system into five layers: data and infrastructure, models, agents, workflows and governance. He stressed that governance should not be a separate step for reviewing ethical issues after deployment. It should be embedded throughout the structure, starting from the initial adoption stage.
He also said hospital AI investment has focused too narrowly on buying high-performing models. Future assessments should look at what agents those models enable, which workflows those agents change, and how much those changes improve hospital efficiency. "An AI model is just one component of AI utilization," Lee said.
Physical AI and the patient experience
Lee said hospitals need to prepare for physical AI, as AI's role expands from judgment to action. While conventional medical AI supports decision-making - interpreting images, predicting risks, drafting documents - AI is expected to move into physical activity through autonomous logistics robots, fall-detection sensors, wearable monitoring, rehabilitation and surgical assistance, and medication dispensing. Once AI takes physical action, hospitals will need stricter standards for safety, accountability and data integration.
The ultimate measure of AI's success should be how much it improves the patient experience, Lee said. Patients often experience problems not within individual stages of care but in the gaps between them: appointment delays, slow communication between departments, changes in surgery schedules, delayed discharge documentation. "What patients really see are the empty spaces and gaps between each area," he said. "How much we can resolve these issues through AI will become an important theme in improving the overall patient journey."
That focus on workflow gaps and patient-facing outcomes is central to AI for Healthcare training, which examines how AI tools change clinical and administrative processes rather than treating deployment as the end goal.
Three questions before adopting AI
Lee outlined three issues hospitals should examine when adopting AI. First, define how to measure AI performance. Rather than counting the number of systems deployed, hospitals should establish metrics patients and staff experience, such as shorter consultation times or faster interpretation of medical images.
Second, define the boundaries of tasks that can be delegated to AI. These boundaries should be embedded in hospital systems, not just written policies, so the point where human judgment takes over is clear. Third, hospitals should determine how to reinvest the time and resources saved through AI, and how to use the data accumulated during deployment.
"It is important to consider how hospitals can further use the time and resources saved through AI and the data accumulated in the process," Lee said. "Hospitals need to think about where they can effectively reallocate the time and resources that AI-driven efficiency gains make available."
"AI models will continue to improve," he added. "But without a stable and specifically designed operating system within the hospital, AI can never be successfully implemented."
Why this matters for operations professionals
Lee's central point - that the same AI model produces different outcomes depending on workflows, staffing and infrastructure - applies well beyond healthcare. Operations leaders in any sector should treat AI adoption as a process redesign project, with clear success metrics defined before deployment, and a plan for reallocating the time and resources AI frees up. Without that operational foundation, the model itself won't deliver results.
Your membership also unlocks:
