Everyone in U.S. healthcare has heard the phrase “people, process, and technology.” Indeed, the phrase has been repeated so often that, when newly referenced once again, its articulation often leaves audiences losing attention. Yet the concept of “process” appropriately took center stage on Monday morning at HIMSS25 at the Venetian Sands Convention Center in Las Vegas, during a panel discussion entitled “Lead Your AI or It Will Lead You.”
Brian Spisak, Ph.D., of Harvard University, led a panel that also included Graham Walker, M.D., of the Permanente Medical Group, Brenton Hill of the Coalition for Health AI (CHAI), Rohit Chandra of Cleveland Clinic, and Tanay Tandon of Commure. The key point around which all the panelists sat in consensus was the inevitability of addressing process in the context of working towards success with artificial intelligence adoption. As Dr. Walker stated early on in the discussion, “Process is friction. Friction is, do you have to turn or slow down? Process is where that friction exists. What we’re really trying to do with AI, he said, is to reduce the friction stuck on humans, and offload some of that onto AI. You can can’t get rid of all the friction, because you’ll go flying off the roller coaster, but you can get rid of a lot.” He cited the example of using analytics to streamline the triaging of emergency department patients, so that EDs are no longer overwhelmed with patient traffic on a day-to-day basis.
Per that, Hill told the audience that, “Apart from the general understanding of process, when I think about it in terms of AI, I think about continuous improvement. You look at the AI landscape today, and not a single person or organization is doing it all 100-percent right. So on this journey, how can we continuously improve, in a responsible way? How can we learn from one another, and take things and distill them down to being able to help others? Test one thing, try it out, figure out where you were getting it wrong, and continuously improve,” he recommended.
What’s more, when asked a follow-up question around monitoring implementation development, Hill advised the audience that “People underestimate the amount of monitoring needed. And we’re trying to figure out ways to streamline processes. Practicality is the key in advancing this. From a legal perspective, there’s not a single legal case that has yet emerged that defines AI liability. We just have to do the best we can right now to continuously improve; people are going to make mistakes, and we’ll learn from those mistakes,” he emphasized.
Inevitably, the panelists ended up discussing where some of the most concrete progress is being made these days around AI adoption. And, in that, two of the panelists agreed that the emergence of the use of AI scribes is in their view one of the best examples of the practical, concrete application of AI in clinicians’ worklives. “AI scribes are one key area” in which a focus on process is bringing forward progress, Chandra asserted. The implementation of AI-scribe solutions “addresses an acute pain point for physicians. Second, it’s relatively simple. You don’t need a huge amount of organizational realignment. It’s the right technology at the right time, and is easy to use,” he said. That’s why AI scribe technology is being adopted faster than other types of technologies.
Walker agreed, saying that “We’re in a real time crunch now, and absolutely need all the tools at our disposal that address pain points. I think the AI scribe tools are extremely valuable, to understand context, so that by the time I get back to my desk, I’m finished with my note. It can make my interactions with patients better. And I’ll speak in real time to the patient and suggest orders, and the AI scribe will be taking that information down in real time and will save me time.”
Per that, Brenton said, “In the work we’re doing, we’re rolling our 3,000-plus community members, to ask them what they want. Sepsis prediction is one big thing. EHR retrieval is another one: how do you summarize what’s in the patient’s record, and supporting the physicians so they’re not reading through tons of pages. How do we make an immediate impact with these tools? AI scribes, EHR retrievals, and predictive models, are what we’re focusing on,” he said, noting that the governance and project management elements around the adoption of those types of solutions are relatively uncomplicated.
And, more broadly, Hill emphasized, “There’s not one silver-bullet governance structure out there. There’s no single process that can be followed, healthcare organizations are starting to have to think like manufacturers, because this is a manufacturing-like process. So some high-level principles we’re working with,” he said, “are, do you have people accountable for specific AI tools? Do you have a governing body? Lawyers, physicians, and nurses should be on there. And, per that, nurses are often forgotten. It’s important that we increasingly put nurses into these positions; they will effectively make or break adoption. And what resources and time are we putting into development? How are you classifying and quantifying risk?”
And it was very clear as the discussion moved forward, that the panelists felt that one key to success in AI involves satisfying end-users with early wins, wins that don’t require particularly intensive governance work, and with a focus on benefits for clinician end-users. The panelists agreed that, over time, norms around process are emerging; in other words, the foundations of streamlined process are beginning to be laid, in the AI adoption arena. No one on the panel doubted that things will take time; but they also agreed that process is an issue that simply cannot be ignored.
