BOSTON - U.S. health systems are scaling artificial intelligence by focusing on frictionless, transparent tools that automate administrative burdens and allow clinicians to focus on patient care, according to panelists at the HIMSS AI in Healthcare Forum here. The discussion, part of an enterprise AI strategy session, also explored how to redesign care delivery around patients and reduce the cost of AI deployments.
Automating the invisible work
Nick Yaitsky, chief AI officer at Lulav AI, framed AI as the next layer of automation that handles tasks humans have always wanted to avoid. "We've always been trying to automate the things that we as humans don't want to do," he said. "For example, we don't want to dig in the ground, so we made a plow." The mechanical, behind-the-scenes work of clinicians - the things patients never see - are prime candidates for AI. "The things that are transparent, the things that you don't even see, really become key projects for AI," he said.
Eric Poon, chief health information officer at Duke University Health System, said the health system has been moving into patient-facing AI after building an online patient community through its portal. Duke is now considering how to use tools like ChatGPT and whether to deploy a chatbot. Patients often choose convenience over physical touch, he noted, which pushed Duke to offer virtual primary care. "I think that the most exciting and probably most challenging opportunity is for us to leverage this moment of crisis and the opportunity afforded to us by this amazing technology to really rethink and redesign with the patients how we should deliver care," Poon said.
Rethinking care delivery with patients
Poon argued that healthcare has spent recent years using AI to "tinker around the edges," which helped establish trust. Now the sector needs to think about delivering care in different ways. "Historically, [patients] have been the most underutilized resource in the healthcare system," he said. Co-designing the new medical system with patients is essential, he added, and AI can give patients a more active role in their own care.
Sandra Powell-Elliott, chief innovation and commercialization officer at Hackensack Meridian Health, cautioned that vendors often design tools for technology-enabled people because that is where immediate financial return is easiest to prove. "Revenue cycle, improving efficiency through automating processes, right? So that's why vendors are focused on those elements, because they can sell those," she said. The result is a market that oversimplifies software around easily measured administrative tasks, making it harder to drive clinical outcomes, patient satisfaction, or readmission rate reductions.
Building business cases for AI requires a shift in thinking, Yaitsky said. He warned against using "number of hours saved" as the primary metric. In an industry already facing severe labor shortages, saving 20,000 hours could become a justification to cut headcount rather than to let the existing workforce handle more volume. The ROI question breaks into two parts: generating new business and reducing overhead. "AI has an incredibly important role to play to remove some of those back office, let's call them tasks, annoyances, things that the system makes me do, remove those to allow the nurses, physicians, allied health teams to do what they do best," he said.
The cost and architecture shift
On the cost side, Yaitsky pointed to a future where local desktop hardware could lower AI expenses. While the cost per AI token is falling, new models consume significantly more tokens. Moving away from cloud-consumption models could give organizations access to "infinite tokens" on their own machines, altering both the cost structure and the security profile of AI. This shift would also change how health systems build their AI for Executives & Strategy roadmaps, tying investments more closely to business needs rather than cloud bills.
Why this matters for healthcare
The panel underscored that AI in healthcare is not about replacing clinicians but about removing the friction that keeps them from patients. Health systems evaluating AI tools should look beyond administrative efficiency metrics and ask whether a tool makes clinical work transparent and patient-centered. As Snowdon said, "Technology is simply a tool, and one of many." The real opportunity is to work with patients to design a system that serves them, not to prescribe solutions from the top down. For healthcare leaders, that means building AI for Healthcare strategies that start with the people they serve, not the processes they want to automate.
Your membership also unlocks:
