Article
It will take clear-eyed leadership to determine which healthcare problems demand AI solutions, and whether the benefits are worth the potential costs.
“The AI Doc: Or How I Became an Apocaloptimist” is a recent film that explores and tests three perspectives on the obviously transformational (and occasionally terrifying) technology that’s teetering on the brink of dominating our personal and professional existence. Of those interviewed in the film, there are the “pessimists/doomsayers” who predict societal collapse if AI eclipses human agency, the “naïve optimists” who have adopted a utopian view of AI’s potential, and the AI founders and industry CEOs who acknowledge some of the risks but are incentivized to keep their foot firmly on the accelerator.
When you consider the wide spectrum that exists with AI deployment, it’s difficult to know how to feel. On one hand, there have been undeniable advancements: At MIT, researchers used AI to discover novel antibiotics for the first time in nearly 60 years to combat drug-resistant infections like MRSA; Johns Hopkins Medicine deployed its Targeted Real-Time Early Warning System to flag sepsis risks up to 48 hours earlier than standard clinical methods; and Henry Ford Health applied AI to target cancer cells with radiation while avoiding healthy tissue.
On the other hand, dozens of companies have laid off thousands of workers while their leaders point to the same culprit; security incidents have nearly tripled in the past three years, including a recent occurrence where one AI model autonomously broke out of its testing environment while attempting to cheat on its evaluation; and confoundingly, an artificial intelligence model can win an International Math Olympiad while still failing to reliably read an analog clock.
In healthcare, we must strike a practical balance, especially when faced with a rapidly accelerating technology that also presents a wide range of potential societal impacts. The idea of balance transcends the notion of optimism vs. pessimism or hype vs. skepticism, instead necessitating a more precarious tightrope walk: balancing incremental, efficiency-oriented thinking with transformative possibility while also respecting the sensitive nature of data and care within the health industry.
That balance is precarious for three reasons: 1) Healthcare is likely to change at an astronomical pace over the next quarter century, 2) as an industry, healthcare is an integral part of local and regional economies, especially in rural communities, and 3) healthcare’s role in life-and-death situations, and its contributions to social welfare, bring a tremendous amount of weight to these conversations. As AI promises near-term efficiency gains in an unpredictable and rapidly changing environment, leaders face a difficult set of decisions about whether potential financial gains outweigh the downstream effects of disruption to the local labor market.
We concede that a balanced, long-term view can feel like a luxury when health systems are dealing with razor-thin margins and the continued pressure to move faster and perform better. So, how can we expand the aperture to include more than “low-hanging fruit”? And how can leaders strategically balance the promise and the risks of AI to increase the likelihood of a better future for the organization and communities you serve?
Tech doesn’t cement your strategy so much as enable it. Organizational maturity starts with an informed and aligned leadership team, and a laser focus on how AI can advance your enterprise strategy. This requires the knowledge and judgment to understand what AI can do, where its current limitations lie, and when it may not be the right tool at all. That means leaders across the organization—not just those in IT or analytics—need enough fluency to ask informed questions, challenge assumptions, and recognize the difference between a compelling demonstration and a durable solution.
That’s why education should be a core part of your organization’s AI infrastructure. An AI maturity assessment should look beyond technical capabilities to consider leadership understanding, workforce readiness, data quality, measurement capabilities, and governance. Do we have the processes necessary to act on what the technology produces? Do clinicians and employees understand when to trust its output and when to question it? Does someone clearly own the consequences when it contributes to harm?
Perhaps most importantly, maturity means knowing when to say “no.” Every potential AI deployment competes for capital, attention, implementation capacity, data, energy, and organizational trust. The art of strategy has always involved deciding what not to pursue; AI doesn’t change that. If anything, the sheer volume of possible applications makes discipline even more important.
As one top-performing health system leader recently noted, “If it doesn't give time back, reduce risk, or improve decision-making, we have to ask why we're doing it.”
The healthcare industry has no shortage of big problems to throw at AI: access, navigation, burnout, administrative burden, care variation, chronic disease, population health, persistent disparities. These are the problems we’ve been talking about for years precisely because they’re hard (and most worth solving).
The more enticing path, of course, is to focus on near-term, incremental gains. Save a few minutes here and automate a task there. Make an existing workflow slightly less painful. There’s value in that, especially in an industry where time is perpetually scarce. But if that’s where AI begins and ends, we may wind up paying many of the technology’s costs without capturing its promise. There also is a risk of saving time through single-use cases, but passing bottlenecks or inefficiencies downstream.
The more important opportunity is to use AI to make headway on problems that have resisted every other solution we’ve tried. Take Google’s AMIE, developed for clinical history-taking and diagnostic dialogue, which has evolved beyond answering clinical questions to synthesizing patient histories and multimodal data, reasoning through diagnostic uncertainty, engaging patients in dynamic conversations, and supporting clinical decisions across the care journey. Or Stanford Medicine’s SleepFM, an AI model that can predict future risk for more than 100 health conditions using data collected during a single night’s sleep. Or Mayo Clinic’s AI-assisted phlebotomy system, a model intended to help mobile care teams perform blood draws in high-need rural communities by identifying optimal veins and guiding needle placement in real time. Or the emergence of self-driving laboratories that combine AI, robotics, and autonomous experimentation to accelerate the discovery of new medicines.
These examples represent only a fraction of what may be possible. AI could expand access to primary care in underserved communities, help identify disease before symptoms emerge, guide patients to the right care at the right time, and help healthcare leaders make better strategic investment decisions through increasingly sophisticated simulations.
Of course, thinking bigger only works if your technology partners can think bigger with you. Challenge your partners and vendors to understand the problem you’re trying to solve and demonstrate how their technology will move the needle. Choose the outcome and evaluate the end-to-end domain or process(es) to target. Then select the technology (and partner) best equipped to deliver it.
A tool might reduce administrative expense while also eliminating jobs. It might improve clinical decision support while introducing new opportunities for bias or overreliance. It might make a workflow more scalable while increasing dependence on energy- and resource-intensive infrastructure.
Some of those tradeoffs are relatively easy to put a number on. Others aren’t. But just because an impact is harder to quantify doesn’t mean it isn’t real.
Health systems are more than businesses; they’re stewards of community resources and health. If part of the mission is to improve the health and well-being of the communities you serve, then it’s hard to argue that the effects of your technology decisions on those same communities should sit outside the business case.
Ask who benefits? Who accepts the risk? What happens when technology gets it wrong? Could it deepen an existing disparity? And are the benefits meaningful enough to justify the disruption required to get them?
As with any new technology, AI hype may lead organizations to aimlessly execute AI for the sake of AI. This approach can create confusion, encourage organizations to focus on smaller initiatives, and introduce unnecessary cost and risk. A common indicator is the inability to move from pilot projects to scaled implementation—or a tendency to build individual use cases instead of reimagining the broader processes they are intended to improve.
One of the biggest risks inherent in AI is mistaking adoption for progress.
That necessitates measuring AI at two levels. First, is it working? Is it accurate, reliable, usable, and helping (rather than complicating) the workflow? Look for time saved associated with token cost; accuracy measures associated with drift and maintenance cost; and adoption metrics associated with user access.
Then comes the more important question: Did it solve the problem we paid it to solve? If the goal was better access, did more patients get care? If it was less burnout, are clinicians actually less burned out? If it was more capacity, did we do anything meaningful with that capacity?
Those answers shouldn’t be reverse-engineered after launch. Define success upfront, decide how you’ll measure it, and be equally clear about what would make you change course. There’s no prize for keeping an AI tool alive simply because it was exciting enough to announce six months ago. In fact, one key commonality we’ve observed among health systems reaping the most return on investment is the willingness to sunset projects that lead to diminishing returns.
“Did the AI solution work?” is probably the wrong question. The better one is: “Did it make anything that matters meaningfully better?”
Maybe being an “apocaloptimist” isn’t such a bad place to land. You can believe deeply in AI’s potential while also being clear-eyed about its costs. Healthcare may in fact need both instincts at once: enough optimism to pursue what’s possible, and enough skepticism to make sure what’s possible is worth pursuing.
That requires thinking beyond adoption for adoption’s sake: Establish the discipline to know where AI belongs, aim it at problems worth solving, look beyond short-term ROI, and measure whether it’s working and if key success metrics improved.
For all the sophistication of the technology, the hardest questions AI raises are still distinctly human ones: What do we value? Who benefits? Who bears the cost? And what kind of healthcare system are we ultimately trying to build?
AI can help us answer many questions. Those, however, are still ours.
Managing Director and Innovation and Insights Leader
As managing director of the consulting and advisory firm Kaufman Hall, a Vizient company, Dan Clarin leads Innovation and Insights for the firm. In this role, he guides Kaufman Hall’s growth through the development of new solutions and capabilities, and by leveraging data and research to equip Kaufman Hall’s clients and team members with proprietary insights.
Managing Director, Practice Leader, Advanced Technology
Terry Hemken is managing director and practice leader for Advanced Technology at Kaufman Hall, where he brings more than 25 years of experience in healthcare analytics, data and technology innovation.
