The Future of Health System AI: Challenges & Innovations
Table of ContentsSystem Integration as the Foundation for Health System AIOne Workflow, Not a Series of HospitalsRethinking Build-First Versus Buy-First Health System AIA Shift Toward Custom-Built. . .
Table of ContentsSystem Integration as the Foundation for Health System AIOne Workflow, Not a Series of HospitalsRethinking Build-First Versus Buy-First Health System AIA Shift Toward Custom-Built SystemsClinical Applications of Health System AI in the EDNudges Against Anchoring BiasHealth System AI as a Force Multiplier for StaffRedeploying Staff Toward Human ConnectionTackling Workflow Friction With Health System AIA Voice AI Handling Real Call VolumeUsing Health System AI to Expand CapacityAfter years of pilots and proof-of-concept deployments,health system AIleaders have sharpened their focus on a harder challenge: deploying the right tools at scale, against real operational problems, with measurable outcomes. The question driving health IT strategy now is not whether AI works, but which big problem it should tackle next. Across the country, chief digital and innovation officers are answering that question differently, depending on where their systems’ biggest bottlenecks sit. System Integration as the Foundation for Health System AIDwight Raum, chief digital information officer at Rochester Regional Health, sees system integration as the precondition everything else depends on. Rochester Regional spans nine hospitals, a large ambulatory network and post-acute facilities across upstate New York, and Raum argues that getting those components to operate as one integrated system has to come beforehealth system AIcan meaningfully close the gap between patient demand and available supply. One Workflow, Not a Series of HospitalsRaum described the goal as ensuring one workflow for appropriate functions across the full continuum of care, rather than treating each facility as a standalone operation. Without that foundation, he suggested, AI tools risk optimizing isolated pieces of a fragmented system rather than the system as a whole. Rethinking Build-First Versus Buy-First Health System AIAt Jefferson Health, Luis Taveras is challenging a premise that has governed health IT for four decades. After his system crossed the 1 million ambient note mark nine months after launch, Taveras is now questioning whetherhealth system AIcapable of generating functional code quickly enough changes the calculus between building and buying technology. A Shift Toward Custom-Built SystemsTaveras framed the shift directly, asking whether health systems should now switch from a buy-first to a build-first approach, given how quickly AI can help write and deploy code. He described this as an opportunity to reimagine computing in healthcare altogether, rather than continuing to default to vendor-built platforms. Clinical Applications of Health System AI in the EDPaul Testa, chief health informatics officer at NYU Langone Health, represents a different frontier. Having moved past the extraction and summarization use cases that defined AI’s first chapter in clinical settings, Testa is now building toward what he calls “the flow” — injectinghealth system AIinto the moving parts of emergency department care, not just isolated documentation points. Nudges Against Anchoring BiasNYU Langone has deployed models in the emergency department that watch documentation, lab results and imaging in real time, surfacing a nudge when a pattern suggests a high-risk diagnosis may be worth revisiting. Testa described this as a powerful opportunity to offer a nudge away from anchoring bias, addressing a longstanding source of diagnostic error. Health System AI as a Force Multiplier for StaffAt Lurie Children’s Hospital of Chicago, Susan Goodson has focusedhealth system AIon optimizing ambient documentation and generative tools for in-basket messaging, aiming to reduce the cognitive overhead that pulls clinicians away from patients. She described the goal as solving boring problems so people can focus on more interesting aspects of their day-to-day work, calling AI a force multiplier rather than a novelty. Redeploying Staff Toward Human ConnectionAt El Camino Health, Deb Muro is pursuing a similar goal throughagentic AIin the call center, handling appointment scheduling, prescription refill inquiries and routine patient questions. The plan is to redeploy staff currently doing that work into care navigation roles, with Muro emphasizing a desire for humans to meet with humans in ways that meaningfully impact patients’ healthcare journeys. Tackling Workflow Friction With Health System AIAt the Medical University of South Carolina, Crystal Broj is targeting the friction clinicians face daily, pointing to disconnected systems, excess handoffs and low-judgment tasks as root causes of both staff burnout and poor patient experience. She emphasized thathealth system AIonly succeeds when paired with operational redesign and strong governance, not deployed as a standalone fix. A Voice AI Handling Real Call VolumeMUSC’s voice AI platform, a bot named Emily, operates across more than 100 patient access lines, handling about 25% of incoming calls around the clock in English and Spanish without human intervention. The deployment has freed up the equivalent of roughly 30 full-time employees of staff capacity, redirected toward more complex patient interactions. Using Health System AI to Expand CapacityAt Children’s Hospital Los Angeles, Omkar Kulkarni is focused on a different constraint entirely: capacity. With more than 90,000 emergency visits and 750,000 total patient visits per year, demand consistently outpaces available slots. Kulkarni is deployinghealth system AIto match patients on waitlists with open appointments across ambulatory clinics, the infusion center and the operating room faster than manual processes allow, calling it a huge enabler for maximizing limited resources.