Healthcare is investing in AI. But most operating models haven’t changed
Health systems have long had more manual work to do than staff to perform it. Now, these health systems are rapidly adopting AI under the promise that it will take on more autonomous work and deliver outcomes at a greater scale than their previous digital initiatives.
They’re piloting chatbots, deploying AI phone agents, testing predictive models, and moving clinical documentation to AI assistants.
Yet in many organizations, the core operating structure remains reactive.
A referral arrives. A nurse reads and routes it. A document waits in a work queue. A cancellation opens capacity that no one recovers. A care gap sits untouched until someone intervenes.
That’s the acceleration of individual tasks, not the transformation toward autonomous workflows that health systems are looking for.
The Dividing Line: System-Level Execution
Most AI initiatives optimize isolated tasks: they generate summaries. Draft responses. Send reminders. Surface recommendations.
Human teams still reconcile schedules, monitor handoffs, work with patients, and trigger the next step manually.
Even best-in-class task-level AI still operates within a model built on manual coordination. Incremental efficiency improves performance. It does not redesign the operating model.
The dividing line is system-level execution; whether AI can run workflows end to end, even when a step stalls or data is incomplete.
This shift is already underway. Across more than 50 health systems, Luma’s Operational AI platform now runs workflows end to end across Access, Engagement, Intake, and Payment Capture, saving millions of staff hours and coordinating hundreds of agentic workflows directly within the EHR and downstream systems.
With system-level execution, the onus is no longer on staff to connect a workflow that’s only partly automated.
Referrals progress. Eligibility is verified. Outreach is initiated. Scheduling occurs. Documentation updates. Care happens. Closure is confirmed. Escalation rules activate when thresholds are met. That’s what it takes to consistently deliver the right outcome.
But too many AI systems lack this system-level execution. Without orchestration embedded into enterprise infrastructure, AI optimizes around structural inefficiency instead of eliminating it. Health systems will invest time, money, and change management into yet another system that doesn’t change the core challenges underneath.
Operational AI enables this needed orchestration.
From Automation to Orchestration: What It Looks Like in Practice
Healthcare’s first digital phase improved access points. The second layered automation onto manual workflows.
The emerging phase is architectural.
Operational AI embeds workflow execution directly into enterprise infrastructure. Work no longer waits in queues or depends on manual routing. Referrals, intake, financial clearance, and follow-up advance automatically, with shared context across the electronic health record and connected systems.
Instead of surfacing work for humans to complete, the system completes the work.
Instead of tracking activity, leaders measure closure.
The difference becomes clear in organizations that have treated AI as infrastructure rather than as a feature.
At Northfield Hospital + Clinics, for example, AI adoption reflects a broader operating model shift. The organization first built its digital front door, embedding self-service scheduling, intake, and more directly into the electronic health record. Northfield then connected patient-facing access to real-time operational visibility. With urgent care wait times, “save a place” functionality, and real-time updates for staff, Northfield enabled urgent care capacity to be balanced across locations rather than managed reactively after bottlenecks formed.
From that foundation, Northfield began applying EHR-integrated AI to high-friction operational workflows. Manual fax processing, prescription refills, patient outreach, and coordination with outside reference labs were redesigned to reduce queue-based work and relieve constrained staff capacity. These changes eliminated more than 250 hours of manual fax handling each month.
Rather than treating AI as a standalone tool, Northfield’s leaders are applying it across interconnected domains of access, capacity, and staff bandwidth. The goal is to move beyond isolated task automation to a system designed to advance work continuously across the healthcare journey.
The Architectural Choice Ahead
The next five years will not be defined by how many AI pilots health systems launch. They’ll be defined by whether AI is embedded into the operating model itself.
Organizations that treat AI as a feature will continue optimizing at the margins. Organizations that embed AI into enterprise infrastructure will redesign how work advances across the healthcare journey.
Without architectural redesign, acceleration reinforces fragmentation. When orchestration is embedded into infrastructure, workflows no longer stall. They advance.
Operational AI represents this shift.
Platforms built around this model embed workflow logic directly into enterprise architecture, replacing layered automation with coordinated execution.
The health systems that make this architectural choice early will define the next era of healthcare operations.
