This is the third in a blog series on strategic planning in health care. The first piece described health care as an equation of great computational complexity with a variety of internal actors on one side reacting to large external social forces on the other and focused on the value of adopting a scenario planning approach. The second in the series looked at the effects and implications of blurring boundaries – how distinctions among health care categories continue to change. In this piece we will look at the seemingly pivotal, but impossibly vast subject of AI and its influence on health care.
AI presents knotty challenges to strategic planning in general and health care strategy in particular. It is vast, increasingly ubiquitous and expanding furiously. AI, as with any major technological breakthrough – think of the internet – will not evolve along a neat, predictable path. Nor from a scenario-planning perspective does AI fit comfortably into conventional socio-political-economic-regulatory scenario narratives. AI innovations and applications could conceivably accelerate even in the most hostile business conditions, much like the iPhone, which was introduced on the eve of the Great Recession.
Alternatively, AI could quite plausibly get derailed or at least hit speed bumps if politicians, regulators, and the public push back against its unconstrained growth and influence. Or if there’s some kind of AI catastrophe. And it is a safe bet that health care is the one sector where AI’s influence will come up for rigorous scrutiny, sooner rather than later.
AI Roots in FSG Scenario Planning
FSG scenario planning consultants have been engaged in strategic planning with health care organizations for nearly 15 years. Back when we started working in the category AI hadn’t erupted, but its precursor and necessary partner – what we used to call “Big Data” – was salient, and just as we were to see subsequently with AI there was a mixture of optimism and concern about its effects. At that time, the health care sector was about two generations behind the commercial/private sector in its adoption and use of IT and was just beginning the transition to electronic health records (EHRs). Fast forward to 2026 and health care has leapfrogged into the lead in its adoption of AI and is reportedly now deploying it at more than twice the rate of the broader economy.
Mirroring the way that the breakthrough into modern AI happened only when computing power met massive, diverse datasets (Big Data), an important driver of health care’s rapid adoption of AI has been the sheer amount of data from EHRs, and the need to manage and process this efficiently, mitigating clinician burnout by removing the administrative burden from physicians using ambient documentation, and at the same time reducing costs for provider systems.
So far, so good. But wait…
While administrative AI has been widely adopted, diagnostic AI has advanced more slowly than enthusiasts had predicted even though the underlying technology has advanced faster than almost anyone expected over the past three to four years.
And therein lies the tension between algorithmic capability and real-world clinical deployment. Powerful deep learning models and large language models (LLMs) can routinely outperform human doctors in controlled tests. They can analyze images faster than radiologists and accurately catch rare diseases or occult cancers before human eyes can spot them. However, technology optimists have drastically underestimated the “deployment gap”—the time it takes for a functional tool to safely enter routine hospital operations. While the underlying science is moving at a breakneck pace, actual adoption at the patient’s bedside remains slow and cautious.
So why is this?
AI Adoption hurdles at play
The integration of diagnostic AI into actual healthcare settings faces several persistent barriers: Doctors bear the ultimate legal responsibility if a diagnosis is wrong. Because complex deep learning models cannot easily explain why they reached a conclusion, clinicians are naturally hesitant to trust them over their own judgment.
Then there is siloed and ‘messy’ data: Medical information is trapped across fractured EHR platforms, billing databases, and disconnected imaging archives. AI requires unified, high-quality data streams to function, which most legacy hospital infrastructures cannot provide.
Lastly there are distributional shifts: An AI trained on pristine data from a top-tier academic research center frequently degrades when exposed to the messy, atypical patient demographics or older hardware found in local community clinics. Rare cases are by definition rare and therefore rare pathologies and early-stage disease are underrepresented in most medical AI training datasets. A model trained predominantly on common presentations will encounter atypical cases in clinical practice without sufficient exposure to learn from them.
Not surprisingly, diagnostic AI has the highest rates of adoption so far in specialties where image recognition is central — like radiology, pathology, dermatology, ophthalmology, and cardiology. It is lowest in psychiatry, rheumatology, physiatry, and much of primary care —specialties where diagnosis rests on narrative history, conversation, physical exam, ambiguous multisystem synthesis, and longitudinal interpretation of symptoms and function.
Ultimately, diagnostic AI is transitioning from an era of hyperactive hype into a period of systemic normalization. The technology is not failing because it lacks capability; it is moving slowly because reshaping the highly regulated, risk-averse infrastructure of medicine takes a long time.
But assuming that all these barriers to adoption are likely to be resolved over the next few years, then what? Where could AI have the greatest impact?
Could AI accelerate preventive medicine?
Historically, US healthcare has been organized around treating illness. One long-term goal for AI could be to help keep people healthy, making preventive medicine financially attractive. It is already accelerating a broader shift that was underway, driven by value-based care, aging populations, chronic disease, wearable technology, and reimbursement changes. At the moment several barriers remain:
- Most payment in the U.S. still rewards treatment more than prevention.
- Preventive benefits often take years to materialize, while investments are immediate.
- AI models require high-quality, interoperable data that many organizations still lack.
- Clinicians remain cautious about relying on AI for preventive recommendations without compelling evidence – for legal and professional reasons – but also because patients are not yet sick, so recommendations are being made based on a risk assessment.
- Patients may not consistently engage with preventive tools or wearable technologies.
AI could strengthen the business case for prevention by helping organizations in a number of ways: identifying who will benefit most from interventions; deploying staff more efficiently; reducing avoidable hospitalizations; and managing chronic diseases proactively. Perhaps AI’s greatest contribution to preventive medicine may not be inventing new preventive therapies but making prevention operationally and economically viable at scale.
How trust could shape AI’s future deployment in health care
The defining role of physicians may shift from being the primary source of medical knowledge to being the primary steward of judgment, ethics, trust, and human relationships in an increasingly AI-enabled health care system. But will AI be used to make physicians more human by removing administrative burdens and enhancing clinical decision-making? Or will it be used to make medicine more industrialized, protocol-driven, and impersonal?
The consensus view among the health care professionals we have worked with is that AI’s impact will depend less on the technology itself and more on the values, incentives, governance structures, and professional norms that shape its use. If this view is broadly correct, the future of medicine may therefore hinge not merely on technological innovation, but on preserving trust, professionalism, and human connection while embracing AI’s considerable capabilities.
But scenario thinking challenges us to consider alternative outcomes, including that AI leads to new workforce models that in turn drive a fundamental rethinking of what professionalism in medicine looks like in the future.
Who knows? No one, really. Which is why scenario thinking that is both rigorous and humble is more than ever necessary to anticipate the complex road ahead for AI in health care.