The Future of Medical Innovation with AI: Yonsei University Health System Uses AI Agents to Enhance Efficiency and Professionalism in Administrative & Support Processes

Emergency TIPS procedure scheduled for Patient John Doe. Please prepare.

Late one evening in Seoul’s Sinchon district, as a night shift handover winds down at Yonsei University Health System, Juhui Jung, a registered nurse, checks a message on Microsoft Teams. Faced with an unfamiliar procedure and complex preparation steps, she would once have needed to search through guidelines and reach out to other departments to find the information she needed.

Today, the process looks different. Without leaving Teams, Jung opens the Nursing Test Q&A Copilot Agent. The test purpose, preparation steps, required consent forms, and reference images appear together on her screen. Even when faced with unfamiliar procedures, Copilot is there as a reliable guide, helping staff respond quickly and accurately around the clock.

Designing AI for Yonsei University Health System

Demand for AI utilization existed within Yonsei University Health System for some time. In practice, many employees were already using AI tools on an individual basis. However, formally applying AI in a hospital environment came with practical constraints related to security requirements and internal policies. Under these conditions, there was a clear need for a role that could deploy AI safely while complying with hospital security policies and connecting the technology with the needs of frontline departments.

As a result, the role of the Digital Health Strategy Center gradually began to take shape. Yonsei University Health System is expanding the use of Copilot agents, with some frontline departments already recognizing that this will lead to positive changes such as reduced repetitive work and faster access to information.

Hyungjin Lee, Director of the Digital Health Strategy Center, explains the context behind these considerations as follows.

Rather than focusing on AI adoption itself, we had to start by thinking about what form of AI could actually be used in the field while maintaining network separation and security.

This consideration soon led to AI use in administrative and operational support areas. AI had already been actively used in clinical care and research, but expanding its use into administrative and support departments proved more challenging. This was because both the complex structure of a tertiary care hospital and strict security guidelines requiring a network separated environment had to be met.

In this context, a wide range of AI initiatives have been actively moving forward in clinical care and research with organization-wide support. By contrast, administrative and support departments have taken a more cautious approach to AI adoption.

The challenge facing the Digital Health Strategy Center was clear. It needed to build an AI platform that staff could readily use in their day-to-day work, while fully safeguarding the security of internal documents and guidelines. Rather than deploying a single, centrally managed AI system across the organization, Yonsei University Health System set its sights on a citizen developer ecosystem, where each department could design and continuously refine its own agents to reflect the specific nature of its work. This direction reflected a realistic assessment that approaches considered in the early stages, such as building internal GPU infrastructure or focusing on in-house model training, would involve a high-cost structure and significant operational burden to scale across the hospital. With the technology landscape evolving rapidly, the goal was to move away from an operating model that required repeated system rebuilds and instead secure a more sustainable approach over time.

Today, the Digital Health Strategy Center provides a shared foundation built on Copilot Studio and the Power Platform, focusing on lowering the barrier to AI adoption by building on existing experience with tools such as Power Automate. Through training and guidance, the Center is also working to establish a low code environment where frontline teams take the lead in response design and quality validation.

As part of this approach, Yonsei University Health System selected a SaaS based architecture that builds on the hospital’s security policies and access control framework, while using external AI endpoints for its services. This model makes it possible to adopt the latest capabilities through configuration changes rather than complex rebuilds, securing both flexibility and efficiency as performance continues to improve.

These efforts began in Microsoft Teams, the familiar environment staff use every day. As many employees were already familiar with Microsoft tools, integrating AI into existing workflows was a more logical choice than introducing an entirely new platform. For AI to be used consistently, it had to enable staff to ask questions and receive immediate answers within the tools they use for their daily work.

Early Changes Felt First in Nursing Teams

Nursing teams were among the first to experience the impact of this approach. As professionals working closest to patients, nurses rely on timely access to accurate information, which directly affects the level of service provided to patients.

In daily practice, nurses face a wide range of situations. They may assist foreign patients in unfamiliar circumstances, respond to repeated inquiries from multiple departments, provide care for tests or procedures they do not routinely perform, or confirm guidelines from other teams while preparing necessary treatments.

Much of this information exists in different formats, including text, images, and tables, and the sheer volume of guidelines can make it difficult to locate what is needed quickly. In some cases, it is not always clear which department to contact. When materials are scattered across departments, efficient sharing becomes difficult, and repeated questions place an added burden on primary clinical teams. In a hospital that operates on a 24-hour shift system, delays are more likely during night and weekend shifts, further increasing the pressure on frontline staff.

To reduce this complexity and make critical information easier to access, testing of the Nursing Guidelines Q&A Copilot Agent began.

The Q&A agent was designed to surface relevant guidelines and images together on a single screen in response to a question. During testing, nurses actively suggested improvements, noting that adding certain details could help further reduce repeated inquiries. Many also said that finding the information they needed had become far simpler. In a hospital that operates 24 hours a day, the ability to check information immediately regardless of time or location proved to be a meaningful improvement.

For Juhui Jung, Copilot has helped reduce the time and effort spent on documentation, allowing her to focus more on projects and system improvement work.

I used to spend nearly three weeks entering and organizing Excel-based materials, she says. Since using the Copilot agent, I can now finish the same work in a single day.

In the past, preparing data for AI training required a lot of manual effort to convert and process materials. Today, however, the team can organize information more efficiently using original source documents. Copilot can also structure meeting discussions and easily turn them into staff notices or system announcements, making information sharing far smoother across the hospital.

Testing the agent in real clinical settings also underscored the importance of document management. Because hospital guidelines and handover materials are constantly updated, a one-time, static approach is unsuitable. Jaerin Jung, from the IT team, points out that document format and structure are critical factors that influence agent performance.

Even documents that seem intuitive to people can be hard for AI to understand, he explains. That’s why we continue to carefully consider how documents should be structured and conduct ongoing testing.

Now, each department reviews its materials based on frequently recurring questions and incorporates feedback from pilot use to refine documents over time. Through this process, both the accuracy and the practical usefulness of agent responses continue to improve.

From Hospital Administration to Clinical Care: Yonsei University Health System’s AX Roadmap

Yonsei University Health System is advancing AI adoption through a two-track approach. In administrative and support areas, AI is being expanded quickly to reduce repetitive workload and information delays. In clinical settings, where patient safety and legal responsibility are critical, adoption is being reviewed carefully and expanded in stages, from data architecture to clinical validation.

Efforts are currently focused on exploring how Q&A agents based on departmental documentation can help organize the hospital’s internal knowledge system. Over time, this work is expected to narrow gaps between departmental guidelines and serve as a foundation for a shared institutional knowledge base across the organization.

In frontline departments participating in the pilot, staff have responded positively, noting that they now turn to the agent first for questions they used to confirm repeatedly, allowing them to stay more focused on their work. Discussions are also continuing around what kinds of changes may be possible in daily workflows going forward.

Yonsei University Health System’s AI Transformation roadmap is clear. Hyungjin Lee explains that the ideal structure is one in which each department builds AI that best reflects its own work. The goal is for AI to move beyond a single function and become a practical tool used in the daily work of both clinicians and staff.

Yonsei University Health System views organization-wide AI adoption as a single, unified direction. AI is already being used across multiple areas where medical data is involved, including medical record documentation, patient risk prediction, and research, and this momentum is expected to extend further across operations, research, and clinical care.

Starting with a Microsoft Copilot Studio based agent ecosystem, Yonsei University Health System is gradually advancing its AX roadmap, step by step, toward innovation across hospital-wide operations.