Earlier this week, I had the privilege of moderating a fireside chat at the HIMSS AI Preconference Forum with two people who have shaped my career: Suresh Balu and Eric Poon. Suresh and Eric are former colleagues and mentors. Our relationship was forged over the 10 years the three of us spent together at Duke. That shared history gave our conversation in Las Vegas a depth that allowed us all to be candid about the issues health systems continue to face with innovation.
The thread running through all our work together (and through the HIMSS anel) is a question that anyone serious about healthcare AI has wrestled with: With so much opportunity for improvement, why is it so difficult to leverage a transformative technology like AI to solve clinical and operational problems?
So many clinical and operational workflows in healthcare are manual, inefficient, and incredibly high stakes. Mistakes can have grave consequences. These tasks are obvious candidates for AI-driven automation. But hesitation on the part of health systems, or (often justified) skepticism of end users, slows adoption.
Bridging the gap between the impact innovation should have in healthcare, and where the system is today is a focus area for all three of us, and the panel was an opportunity to share what we've learned along the way.
Building Clinical AI Solutions: Duke Health Model Implementations
Over the last decade, Suresh, Eric, and I worked together to build and integrate more than 60 AI models across Duke Health. These ranged from an early sepsis detection deep learning model to a flexible electronic health record (EHR) search and synthesis tool.
The Duke Institute for Health Innovation (DIHI) was founded in 2013 with a mission to catalyze innovation through high-impact research, leadership development, workforce training, and the cultivation of an entrepreneurial community. All our work is grounded in a design-thinking approach. That mission continues to animate the work Suresh, Eric, and their teams do today.
One of the things that makes DIHI distinctive, a credit to Suresh and Eric’s leadership, is their deliberate cultivation of a diverse portfolio of projects spanning the full spectrum of care delivery and operational use cases. Duke has become a recognized leader in AI applications that directly improve patient outcomes as well as enhance the efficiency of care delivery. That breadth of focus has been central to Duke’s success.
Why Healthcare AI Adoption Fails: It’s Not the Technology
During our fireside chat, Suresh and Eric spoke openly about how they've worked to align emerging technology opportunities with meaningful clinical and operational improvement at Duke. Perhaps the most important insight we surfaced, and one that might surprise people outside of health systems, is that the barriers to effective AI adoption are rarely technical. Existing technology capabilities are often more than sufficient to make meaningful changes in healthcare. What holds organizations back is a deeper problem: misaligned incentives, poor coordination, and entrenched inefficient processes.
At times, realizing the full value of a promising technology required Suresh and Eric to advocate for changes in organizational structure or incentive design: not just workflow tweaks, but more fundamental shifts in how the institution was organized around care delivery.
An example: a Duke team set out to reimagine the front desk operations for Duke Health’s ambulatory clinics. The process for scheduling patients and checking them in, while seemingly mundane, was creating operational inefficiency and expending precious resources.
Having Duke leadership’s buy-in was the most critical aspect of successfully transforming the front desk operations. Eric’s team chose an out-of-the-box solution from their EHR to solve the scheduling and check-in workflow issues. While the technology should have been simple to turn on, it still took months to deploy the technology in clinics because of the change management required by the staff. To date, the solution has been integrated in about half of Duke’s ambulatory clinics. Without operational and organizational support, the integration of the new technology would still be far behind where it is today.
Early on, Duke leadership created a board-level metric to track if AI technologies were providing tangible benefits once used in practice. The visibility into the technology’s performance from senior members of the clinical, operational, and technical teams has created a shared understanding of the technology that has allowed the organization to find opportunities to use innovative technology.
Practical Advice for Healthcare AI Leaders: Lessons from Duke
Suresh and Eric offered direct guidance, grounded in experience, for health system leaders trying to build an AI strategy today.
They recommended that an organization’s AI strategy must be directionally aligned with its overall business strategy. Duke has aligned some of its focus areas like revenue cycle automation, burnout reduction, and nursing care model redesign, with AI.
The innovators at DIHI have also prioritized working side-by-side with clinicians and technology leaders to think about what technologies they should introduce to solve organizational problems, try them out, and scale them where appropriate.
Both Suresh and Eric were adamant about the value of shorter planning horizons: traditional one-to-three-year technology strategies simply don't work for AI, given how rapidly the landscape is shifting. Health systems need to operate on a much shorter planning cycle — thinking in terms of months rather than years — while maintaining a longer-term directional vision.
They also advocated for health systems to establish a structured process for surfacing and prioritizing pain points, such as a formal mechanism for identifying where innovation can make the most meaningful difference, informed directly by the clinicians and operational staff living those problems every day.
Scaling AI Innovation: DIHI’s Build-Buy-Partner Approach
Looking ahead, Duke is pursuing several paths to remain on the leading edge of healthcare AI and ensure that what it learns reaches the broader industry.
The Health AI Partnership (HAIP). DIHI co-created HAIP in 2022 as a multi-stakeholder collaborative to help healthcare organizations, particularly those in rural and community settings, adopt AI safely, effectively, and equitably. Through HAIP, the team has developed a key decision-point guide for health systems implementing AI, a community-informed framework to mitigate the impact of AI on health inequities (HEAAL), and an AI vendor disclosure framework.
A "build, buy, partner" strategy. Duke is formalizing a framework to guide decisions about when to build proprietary tools, when to purchase from the market, and when to partner externally.
Commercial partnerships are a key path to scaling innovation through external commercialization. Duke has already partnered with Artisight, Tres, Abridge, Microsoft, RedCell Partners, and Vega Health to bring its innovations to market.
Suresh and Eric emphasized the importance of partnerships to help them stay on top of the rapid pace of change in AI and quickly develop new tools. They also emphasized the need to cultivate the right partnerships with companies that align with the long-term vision of Duke and its values, rather than contracting point solutions to solve siloed problems.
Finding the right partners will allow Duke to be agile and rapidly iterate if AI fails. Because it can be hard to tell whether a piece of AI will really work, Duke has created a process to quickly test new tools through low-cost proof of concepts.
DIHI also ensures it has organizational buy-in and adequate resources before scaling the technology to successfully integrate AI.
As the healthcare industry moves deeper into the AI era, the organizations that will lead are those that invest not just in the technology, but in the infrastructure — human, organizational, and analytical — needed to integrate it responsibly and sustainably. I'm grateful to work alongside innovators like Suresh and Eric who are not only building that future at Duke but sharing what they learn with the broader field.
Thank you to HIMSS for the opportunity to moderate this conversation, and to Suresh and Eric for their generosity, candor, and leadership — on stage and off.
