Forget building the plane while you fly it—healthcare AI transformation is like changing the bus in Speed

Sara: Everyone is thinking and talking about embedding AI in healthcare delivery. It’s captured everybody’s imagination. In fact, Chartis’ 2025 Digital Transformation Survey found that 9 in 10 executives are prioritizing AI to transform healthcare for more proactive care delivery. But are organizations making progress toward this goal?

Tom: Health systems are looking at how they can do everything they’re doing today but use AI to make it better. 

We’re seeing some areas of real improvement: In the revenue cycle, we’re seeing lots of progress, new vendors, new capabilities, and good results. We’re also seeing progress in some incremental tasks. Ambient listening has certainly been the “flavor of the month,” and research studies are looking at that.

But many health systems are stuck in pilots.

Tom: They’re not really thinking about it as, “How do we transform our business?” They’re thinking about it as, “How do we employ AI in something we do today?” In some ways, saying “we can test it out and see what we learn from it” feels like a less daunting way to approach it. 

But it’s not simple in a big, complicated health system. Any process that seems consistent across the breadth of the health system often isn’t. Even for things like denials processing, the inputs, outputs, and process may be slightly different at each hospital. 

While it’s easy to spool up small-scale pilots in pockets, it’s really hard to scale across an entire organization, across all the variations, and to systematically drive that kind of breadth of change management. Which pilot are you going to scale? What’s the return on investment (ROI)? How does it fit into the process?

Many organizations are thinking, “How do I use AI to do what I’m doing now but better, faster, cheaper?” Not many are asking, “How do the abilities that AI presents allow me to do what I’m doing completely differently—transforming how I do business—to achieve the same ultimate endpoint?” 

The former approach is inherently constrained and less risky, but it’s also a less transformative approach to value creation.

Sara: It’s a bit like picking a random smattering of things and hoping you get something out of it—in a way, valuing quantity over quality without having a real strategy behind it. 

Sara: One of the big challenges is that you can’t do it for the whole system, but everything is very interconnected. Where do you even begin?

Tom: Primary care is a great example. If you’re a primary care physician or part of a primary care team, you know all the steps. And you know how each step impacts the next step and understand the nuances of your operating model. A lot of times, people look at a function within that and think it’s relatively easy to change. 

But they don’t think about it in the context of the full system, so they underestimate the scale of change management required. 

Alternatively, a health system could decide, “We’re going to reconceive primary care. We’re not going to change the legacy model that hundreds of clinicians and thousands of staff support. Instead, we’re going to build something discrete and smaller in scale but fully scoped.” 

The power is in how you now think about primary care fundamentally differently. 

For instance, in the old world, simple things like asking a patient how they are feeling and tracking sentiment require call center reps to reach out. Now, you can do much more, more effectively, and at a cost that’s coming down toward zero. You can respond in real time to patients’ needs, inquiries, questions, concerns, and data. You can proactively prompt a patient to do something. 

In this new world, you could reconceive a different kind of primary care.

You can define what that model will do and the guardrails you’ll put in place. You can build that model, scope it, and deploy it. Then, let it compete with the legacy model. Promote that new model and let people opt into it. If your marginal profit is essentially better than the other model’s, every patient who goes in creates economic value for the health system. This is intrinsically more scalable. 

That also allows you to start thinking about how you manage population health, how you proactively route people to elevated care needs, how you put them into chronic care management programs, and how you dispatch remote patient tracking tools.

You can take this approach alongside AI process improvement pilots. But always pilot with intention: How will you get to process improvements that apply across the enterprise? The “field of a thousand flowers” concept doesn’t yield a result. It just adds cost. You have to be able to identify and cultivate what works.

You can think about it in two discrete forms: (1) How do you turn these pilots into process improvements that apply across the enterprise? (2) But also, how can you change what you’re doing fundamentally, build a defined scope with defined value levers and measurement goals? And have both things running in parallel.

Not looking at how you transform the business is a huge, huge gap.

Sara: You said a couple things that really resonated with me. One is that you have to conduct process improvement and transformation concurrently. The other is that in the transformation categories, you’ve got to go super deep.

Realistically, there are only a handful of things that you can do at once because you’re going to uncover so many new things. Not only do we have the operational and business model implications, but we also have a frontier of technical things all happening and converging at once. You’ve got to match all these things.

For instance, whether you’re using generative AI or agentic AI, there’s a non-deterministic aspect to figure out: A defined input doesn’t always give you the same output. You’re going to have to understand how you account for this in an operational context and how you model it out financially in aggregate. 

These things can be tricky. You can’t do a hundred of them at once. For a given enterprise, you might start with three.

Sara: I find health systems are generally in a few categories when it comes to AI:

• Some health systems are still waiting, or they are conducting process improvement, but they don’t have the resources or capabilities to engage at a deep, transformative level.
• A lot of health systems are experimenting with a high volume of pilots, potentially without a bigger plan or strategy.
• A very small number of health systems are connecting strategy to actual testing, implementation, transformation, process improvement, and measurement.

For health systems that don’t have it all buttoned up, from strategy through delivery, they’re often stuck in a common scenario: They want to prioritize AI, they want to make progress, but they don’t have the necessary foundation, or they have major technical or financial blockers.

Tom: The first thing you have to do is define the strategy for what you’re trying to do with AI. If you don’t identify the parts of your business that are most important to impact, you just peanut butter around everything.

That’s where these organizations get stuck in pilot hell. They can check the box with lots of things. For instance, when their board asks if they are doing anything with ambient listening or automatic inbox, they can say yes because they are trying it with some of their physicians. 

But the box-checking adds no value. And if you’re using incremental third-party partners, it actually adds cost because now you’re supporting the manual process and the automated process, which you’re paying a massive premium for because it’s not scaled. 

Some people have gotten pretty far just from the sheer will of a visionary senior executive or chief executive officer, and they have built a leadership team to drive that forward. But if you haven’t tied it to what’s strategically most relevant, how you’re going to make investments, and how you’re going to measure it, you won’t achieve transformative impact.

There’s no silver bullet. But it’s that last mile of how you are actually going to monitor and manage this and say, “This is what is most important to us across the leadership team.” Otherwise, none of your management metrics—how you track it, how you measure progress, how you look at ROI—are going to be broadly supported and believed by the rest of the organization. 

Sara: How many enterprise strategies are specific enough to drive an AI vision or allow people to understand an AI vision within that context? As a former leader in both strategy and AI and technology at a health system, I really had to work super hard to bridge that gap. What level of specificity is required? Are people getting there?

Tom: It’s a really good question: How do you think about the strategy that a health system may have built 2 years ago? A lot of elements will be useful input and valuable. But in some ways, it’s already obsolete in the context of AI strategy. 

What AI could do 6 months ago is radically different from what it can do today. So it calls into question how health systems are going to think about an AI strategy, which is going to have a much faster cycle time and iteration. There’s a series of questions leaders start asking here:

• What is our AI strategy? 
• How will it allow us to advance our existing enterprise strategy?
• Is our enterprise strategy today the right strategy when we take into account the competitive landscape that’s going to be using AI (our partners, payers, and competitors in other health system domains)?
• Is our enterprise strategy still the right strategy?
• Do we have an AI-informed strategy for our business?

If the starting enterprise strategy isn’t specific enough, it won’t be helpful.

But if you’ve got a really well thought-out strategy that says, “These are the areas we’re going to invest in, and this is how we differentiate,” it can certainly be a powerful driver for identifying the AI initiatives that will advance those strategy points. That back and forth is healthy: “These are the AI initiatives. Knowing what’s coming in the AI landscape, does that change how we thought about the underlying strategic goal initially?”

Going back to our primary care example, a health system might say, “We want to grow our primary care offering but reduce our operating losses by 50%.” That’s a very specific goal. Primary care is important, and it’s a great strategic ambition. It positions the health system for value-based care and drives referrals. 

But now if you start the program saying, “How do we use AI to advance that?”, you can say, “We can change that goal altogether: We could change the entire economic model and move more substantively. We want to move the model so the net performance of our primary care offerings is not subsidized.”

Then it’s a question of how you build a model for which the marginal value is positive versus the legacy negative. How do you adjust the ratios quickly over a defined period to get to a sustainable operating chassis for this service? That, in turn, may change how you think about the definition of the strategy.

Sara: We—and when I say “we,” I mean the “royal we” as an industry—haven’t redesigned the system. We’ve just added to the mix. We’re still doing things that maybe shouldn’t exist in the first place, and we’re just making them more efficient at best.

On the administrative side, where a lot of AI has been applied to date, we’re not able to consolidate the gains yet because there are a lot of hands in the pot, a lot of different vendors. How do you actually change the workforce dynamics and all these different things?

Tom: Any time something is incremental, you run the risk of just more cost and more technology poorly applied. It’s not going to drive real value.

It comes down to: Are we looking at a discrete function and just adding incremental cost, or are we looking at a new business model with its cost and revenue implications? And you’re building it de novo in a way.

Then the questions become, “Do we believe we can build that? Can we execute against that? Can we operationalize against that? And how are we going to measure the progress toward this new operating model?” 

A lot of times in healthcare, it’s the old cliché that we’re going to change the engines while we’re flying the plane. Can health systems really transform everything they’re doing while trying to keep razor-thin margins, rebuilding, and retooling?

It helps to look at how other industries have done this. The airline industry, for instance, changed its entire consumer front-end experience. They didn’t say they were going to blow up the way they did it piece by piece and all of a sudden, it’s going to be completely different. Instead, they created a parallel capability. 

In the early days, it was online ticketing, then it was online check-in, then online bag prep. And now, they will pick up my bags the day before and deliver them to my location. 

They said, “This is the old model. And this is the new model.” The new model is much more cost-effective, and once consumers started going to the new model, the old model became catastrophically expensive because now the volumes aren’t even there. They’ve got a big, fixed-scale infrastructure, so they started charging extra for using the old model to move even more people to the other side.

We can learn a lot in healthcare from this experience. We don’t have to transform this moving vehicle while we’re driving it. We can build a parallel vehicle and then—good 1990s pop culture reference—do what they did in Speed and go from bus to bus. We don’t have to change the entire bus we’re driving. We just have to get another one that we can get the people onto.

But you have to then be able to define what you’re putting them onto with detailed specificity. You have to be able to go deep. It’s not just rolling out a technology and the skies will part, the sun will shine, and the birds will sing. It’s a detailed operating model. It’s an investment.

But it’s a different cost platform than adding more beds, which may have a 10- to 40-year functional life. These are much more capital-light investments. They are much quicker in the grand scheme, but they do still require investment. 

With razor-thin margins, a lot of health systems say they can’t pick all these bets. No, but you should pick a few of them. Go deep, then execute like hell. Measure it, track it, value it. If it works and you’ve done it well, you’ve transformed your organization. If it doesn’t, you learn from it and you try the next one.