For many years, providers have watched AI arrive with a mix of optimism and restraint. Most AI deployments were deliberately constrained––focused on pilots, narrow use cases, and approaches that kept humans in the loop. That caution was warranted as the technology matured and organizations built confidence in where it could be trusted.
It is now becoming a limitation.
What has changed is both the pace of improvement and the scope of what AI can credibly do. Work that once required a human is starting to be handled, in part, by AI, and is moving closer to the core of care delivery and operations. As that boundary moves, so does the implication: parts of the current operating model that once felt fixed are now open to redesign.
This will not play out as a series of incremental gains. For academic medical centers (AMCs) and nonprofit health systems, which have historically been more risk averse and slower to adopt digital and technology solutions, the gap between how they operate today and what is now possible is widening.
What this shift points to is a more fundamental redesign of access, care delivery, operations, physical infrastructure, competitive positioning, and governance. Leaders who approach AI as a tool for transformation, not just augmentation, will define what the future operating model looks like.
The Next Phase of AI
The need to act is not theoretical. It is driven by changes already taking hold across providers. Four convictions will shape how provider leaders approach the next phase of AI.
Conviction 1: The most valuable AI in health care won’t always have a human in the loop. Health systems have largely restricted AI to advisory roles––flagging a risk, surfacing a recommendation, prompting action. This approach is logical, but insufficient. The challenges associated with capacity, cost, and access cannot be addressed through augmentation alone.
The larger opportunity is moving selectively into automation: AI that can triage, diagnose, monitor, and follow up without waiting for a physician to approve each action will not only expand systemwide capacity but also improve outcomes.
This is not a fringe view. Autonomous chronic disease monitoring and AI-directed low-acute care delivery are already in early deployment. The organizations that lead over the next decade will systematically identify where autonomy is viable from a clinical and regulatory perspective––and build the quality, safety, governance, and liability infrastructure required to support it.
Who needs to see a physician—in person, virtually, or not at all? Health systems and AMCs that answer this strategically will retain physician trust and patient confidence.
As AI lowers the barrier for patient engagement, total utilization is more likely to increase than decrease. That makes one question unavoidable: Who needs to see a physician––in person, virtually, or not at all? Health systems and AMCs that answer this strategically will retain physician trust and patient confidence. Those that let it be defined by others will face a harder transition.
Conviction 2: Differentiation will come from the operating model, not the AI itself. The current arms race across health systems is focused on selecting the right AI vendor and the right LLM. This choice matters, but it is less important than leaders think.
Within five years, foundational AI capabilities will be broadly accessible, competitively priced, and largely interchangeable. Early signs of tool commoditization are already visible in generative AI use cases such as ambient documentation. Many health systems have already piloted multiple vendor tools and found that the differences among them are hard to see, both in performance and in clinician satisfaction. As a result, pricing is starting to matter more.
What will not be commoditized is what sits on top: the workflows, orchestration, clinical logic, and operating model changes required to translate AI into better performance.
Health systems should be deliberate about avoiding vendor lock-in and owning the elements that will differentiate them––particularly their data layer and orchestration logic. The durable advantage for health systems and AMCs will not come from the model itself, but from how effectively it is embedded into the system.
Stay ahead with BCG insights on health care payers, providers, systems & services
Conviction 3: Facilities being built today might not serve the health system of the future. Capital planning cycles in health systems still assume assets will operate for decades. This is no longer relevant or cost-efficient, particularly as health systems expand access closer to patients in smaller, lower-cost formats. Independent operators, particularly in ambulatory surgical centers, are already building facilities at a fraction of the cost, designing for shorter lifecycles, and in some cases eliminating traditional administrative and front desk space altogether.
Many capital plans still assume a model that is human-staffed, procedure-centric, and primarily in-person at every touchpoint.
Facilities being designed and financed today will operate through a period in which AI and robotics materially reshape how and where care is delivered in clinical settings. Yet many capital plans still assume a model that is human-staffed, procedure-centric, and primarily in-person at every touchpoint. These are the elements of the care delivery model that AI and robotics are actively disrupting.
Leaders are beginning to ask harder questions:
- Which clinical service lines will see volume migrate to home, ambulatory, or AI-mediated sites?
- Which spaces will need to accommodate robotic systems that do not yet fully exist?
- How do we design flexibility and modularity into spaces when future workflows remain uncertain?
Misaligned capital commitments in this window will constrain strategic options. Leaders do not need to predict exactly which spaces will become obsolete, but they need to design for an environment that will be defined by rapid change. The successful health systems of the future will build for the optionality to repurpose and reconfigure.
Conviction 4: Intelligent access 24/7 will become table stakes. Patients are already accustomed to on-demand access in nearly every domain except health care. That gap is closing fast.
In the near term, leading health systems will offer real-time scheduling without human intermediaries, AI-navigated triage, continuous availability, and more precise matching of supply and demand across specialties. Over time, these capabilities will stop being differentiating and become expected.
What is changing more quickly is who sets the expectation. A growing set of tech-forward startups are building around 24/7, digitally mediated access from the start of the patient journey. When that becomes the baseline experience, legacy health systems that are still organized around traditional business hours will feel increasingly out of step. They will soon lose ownership of the patient relationship and face disintermediation.
When access becomes easier, more immediate, and always available, demand does not stay fixed––it expands.
There is also a second-order effect. When access becomes easier, more immediate, and always available, demand does not stay fixed––it expands. Patients who might have waited, delayed care, or gone without it altogether begin to engage: a new symptom gets checked that night instead of next month; a chronic condition gets monitored more closely; a question that would have gone unasked now gets answered.
The constraint shifts from attracting demand to absorbing it. That will require new pathways to engage patients outside traditional settings and hours, not just more efficient versions of current operations.
Implications for Health System and AMC Leaders
Given the cross-functional nature of the shifts we’ve just outlined, CEOs at health systems and AMCs will need to take the lead. BCG’s recent 2026 AI Radar survey shows that in many organizations, they already are: nearly three-quarters of CEOs say they are the primary decision maker on AI, twice the share as last year.
CEOs are recognizing that AI is not just a technology shift. It opens the door to a fundamentally different way of running organizations––touching strategy, operations, culture, risk, and talent. Only the CEO––coordinating closely with the executive team––has the scope to connect those decisions end-to-end.
In practice, critical decisions show up across a variety of areas and will likely require actions including the following:
- Reframe the governance question. Rather than starting with where AI can be applied safely, begin by defining where human involvement is truly required, and why. Over time, the default in some domains will shift from human-led to AI-enabled, with clear exceptions and guardrails.
- Redefine the clinician's role proactively. The question of who needs to see a clinician in person, virtually, or not at all, will need to be answered. Health systems that define this proactively will win the hearts and minds of clinicians and patients and be best positioned for the future.
- Set an access ambition, not just a target. Benchmark against technology-native entrants, not just peer systems. What patients experience as “good access” is increasingly set elsewhere. Define what that standard looks like in your market and build toward it on a timeline that outpaces competitors.
- Stress-test the capital plan. Evaluate major capital commitments against structured scenarios: What does this space look like if AI-mediated care captures 20% of current volume? Or 40%? Of the remaining care that requires human involvement, what proportion of patients should be seen in-person versus virtually? What check-in steps will be completed before the patient walks in the door? Design for flexibility and modularity rather than fixed assumptions about future demand.
- Audit your architecture now. Map current vendor dependencies, data layers, and orchestration logic. Identify where lock-in creates long-term risk––and build the internal capability to own and evolve the intelligence layer over time.
- Move from pilots to platforms. The organizations pulling ahead are not running more pilots. They are scaling. To safely shift from proof-of-concept to operational deployment requires new governance, risk tolerance, and leadership attention.
The timeline for these shifts will vary. But the direction is already visible. The health systems and AMCs that lead will be those that made bold, well-reasoned bets while there was still time to shape the outcome.
