From pilot to platform: Johns Hopkins researcher on healthcare AI’s next chapter

Ritu Agarwal, a professor of information systems and health at the Johns Hopkins Carey Business School, shares her views on how healthcare organizations can get the full benefits of AI.
LAS VEGAS — Healthcare has spent nearly 30 years preparing for AI. During this time, the industry has moved from paper records to digital systems, adopted digital health tools and gone through many waves of new technology.
But even with all this progress, healthcare organizations still have not figured out how to use AI across their operations, according to Ritu Agarwal, the Wm. Polk Carey distinguished professor of information systems and health at the Johns Hopkins Carey Business School.
Using AI at scale means making it part of everyday clinical and office work across an entire organization. For example, hospitals would use the same tools across different locations, follow shared rules for managing AI and connect AI tasks with electronic health record systems. Many healthcare organizations are trying AI, but struggling to move beyond small pilot programs.
“Healthcare has many known problems, including poor quality, high costs, limited access and unfair differences in the care people receive. I believe AI could help address all of these problems,” she said. “But we still have a lot of hard work to do before we can achieve these benefits.”
She said one reason for this is that AI in healthcare is still very new. Important questions remain about how to fit AI into clinical work, how to manage it, how to earn doctors' trust and how to measure whether it is worth the cost. These questions will take time and research to answer.
Healthcare Dive spoke with Agarwal at the HLTH 2026 conference. They discussed what is stopping healthcare organizations from using AI more widely, how hospital leaders should measure the value of AI investments and what may happen next in the industry.
Editor's note: This interview has been edited to make it clearer and shorter.
HEALTHCARE DIVE: What are the biggest differences between how an AI model works during a pilot program and how it works when a health system puts it into everyday use?
RITU AGARWAL: I don't think we have made the move from pilot programs to using AI across healthcare organizations. That's because healthcare is complicated, people are complicated and medical conditions are complicated.
When we build AI models, we need enough reliable data to give us confidence that the models will work well in different situations. The difficult part is making sure they work under all conditions.
For example, a model trained with data from Stanford Medical School may be based on a very different group of patients than those treated at a rural hospital in Kentucky. This can lead to what we call model drift. Model drift happens when a model built for one setting is used in a different setting, where it may not work as well.
The short answer is that we need much more development and testing, along with data sets that better represent different patients and settings. This will help us build models that we can trust in everyday medical care.
Many organizations are still far from using AI across their entire systems. I imagine doctors' trust in AI plays a role. What makes doctors accept a new AI tool instead of letting it disappear after a pilot program?
We still have not worked out exactly where and how AI should fit into clinical work. As with any new technology, we need to think about how work processes should change. We should not simply add AI to a process that is already poor or inefficient. At the same time, we should not take away doctors' ability to make their own decisions.
The difficult part is deciding how and when AI should help doctors make decisions. When should it make a recommendation? When should it provide information? When should it bring different pieces of information together? When should it question a doctor's decision? I think doctors will push back if AI is simply added to the way they already work.
Doctors also need to be involved in coming up with ideas for AI tools and helping design and build them. I see many companies that have a chief medical officer and a few doctors on their development teams. But, in my view, they still let technology experts make most of the decisions. Doctors should not just be involved in developing AI tools. They may need to lead that work.
Let's talk about return on investment, or ROI. If a hospital leader asked you how to measure whether an AI investment is worth the money, what would you tell them to measure?
We have been studying the results of healthcare technology for a long time. But AI is still at a very early stage, so we do not yet have enough results to show its effects. What we need is a careful study or evaluation process that measures results before AI is introduced. This gives us a starting point for comparison.
If I were advising a healthcare leader today, I would tell them to set up a system for measuring results before introducing AI. They should also clearly decide what they expect the AI to achieve. For example, they might want it to improve efficiency by reducing the time spent on certain tasks. Or they might want it to improve effectiveness by helping serve more patients or improving patients' health.
They should not try to change everything at once. Instead, they should choose specific diseases or medical conditions and study how AI works in those areas. Then they should follow a clear, step-by-step process. They should not expect to see improvements in revenue or profits right away, but early evidence suggests that benefits may appear over time.
We can see this happening in other industries, so there is no reason it cannot happen in healthcare too. The possible benefits are large. We will have to wait and see whether they become reality.
How do you think the market for healthcare AI companies will change as AI improves? Large technology companies are entering healthcare and releasing tools designed for the industry. What will happen to companies that offer highly specialized, single-purpose products?
I think many companies in the industry will join together. The companies that succeed will be those with the strongest relationships with healthcare systems and access to the largest amounts of data.
Google and Microsoft already have strong positions in the market because they are large companies with many customers. If they can build very large sets of health data, they will have an advantage. Electronic health record companies also have huge amounts of data that they have collected and organized over the years. However, compared with technology companies, they are less quick and flexible when it comes to creating new solutions and technologies.
I think we will see competition and some companies joining together. The companies that succeed will probably become even larger. Companies that offer only one specialized solution may be bought by larger companies or become part of them. The idea of building large platforms that offer many services in one place, known as platformization, could become a very powerful business model in healthcare technology in the future.
