Beyond technology implementation: What’s required for AI activation to transform care delivery

Sara: Every organization is a bit different, but there are some common themes. First, AI deployment often isn’t aligned with enterprise strategy. Instead, it’s off to the side. 

In our last episode, we talked about how it’s not “AI strategy,” it’s “AI in your strategy.” Your enterprise strategy and AI need to be woven together. I think that’s the initial challenge. If you don’t achieve that, your AI strategy might not focus on your organization’s top priorities. You may be able to experiment, but scaling it will be difficult. As a result, it will be difficult to get transformative value from it.

A related theme is that health systems don’t necessarily pull all the pieces together. It’s not just technology. Real transformation comes from the old adage of “people, process, and technology.” I would argue that there are other factors as well, such as systems, structures, workflows, and compensation. Ignoring those things will hamper your ability to get real value beyond the experimental or consultative phase.

Tom: You’re articulating an important distinction. When you think about technology alone, you focus only on how to deploy it. Vendors will talk about deploying a solution, but there’s a difference between deploying something and activating it and managing how it changes your process, how your people use it, and how it integrates into your organization.

Let’s talk about that distinction. Many health systems have a laundry list: “Here are the things we’ve deployed. Here are all the vendor solutions we’ve got.” But it’s important to distinguish that from activation and what truly changes how healthcare is delivered and experienced by patients, providers, and family members.

Sara: That notion of how healthcare is delivered and experienced is often called the “last mile” of digitally enabled transformation. One of the best examples of closing that last mile with the consumer is Amazon. 

Amazon has thought about every aspect of the value chain: From when a consumer makes a purchase online (or even starts to think about that purchase) until they have that item in hand, ready for use, Amazon has thought about price, convenience, and availability. And it has knit everything together in terms of inventory, how consumers can find the product, and how it is delivered to them.

Recently, Chartis published a point of view about the future model of care delivery, which we call HealthCare360. In that emerging model, healthcare will be transformed into a human-centered experience that works—affordable, accessible, safe, and high-quality. To achieve that kind of meaningful last-mile change, you can’t just put technology in place or sprinkle some AI into the mix.

Tom: You identified Amazon as an industry disruptor that blew up the retail model. The danger in healthcare is that people tend to think about the last mile in the context of what we already do today. 

As we talked about in previous episodes, the current approach for many organizations is continuing in the current delivery model but just dropping some magic AI dust on the process and hoping it gets better. We have to think about the last mile as something different. Talking about this in the context of access, the last mile involves building an entirely new process. 

It goes back to your earlier comment regarding people, process, and technology. I know it’s an old consulting cliché, but it becomes the consideration point when you rebuild a process. For every touchpoint, what processes have to happen? It depends on what you’re trying to do. 

Amazon said, “Our last mile won’t be the customer coming to a location to get something. Instead, we’re going to deliver to every American’s door.” Eventually, it became the door of every consumer in the world.

First and foremost, it’s about asking, “What’s the goal?” The healthcare industry has spent decades trying to make access better. But every objective metric—time to appointment, outcomes, patient perceptions—indicates that it has only grown worse. We’re not trying to get a patient a reimbursable appointment with a clinician, but that’s what the existing model was built to do. Instead, we’re trying to get a correct, useful, actionable insight to a patient about their discrete care needs. 

When you start to think about it that way, your last mile is totally different. 

Now you’re thinking, “How do we get the patient to the right place? How do we proactively engage them?” Now you’ve got legal considerations, clinical pathways, a whole host of partnerships, operations, and the rev cycle. You’ve got a completely different set of last-mile considerations.

We use a lot of clichés in our industry, but the one that drives me most batty is “changing the engine on a plane while it’s flying,” which we’ve talked about before. Nothing is impossible, but it’s very close. 

At the end of the day, all you’re going to do is replace parts of a janky old system with moderately better ones. Replace the old prop engine with a turboprop. Replace the turboprop with a jet. Replace the jet with a higher output. But with every change, you’re changing the balance of the elements of the last mile that the consumer experiences, and you’re putting the plane at risk of crashing.

We have to think about what we’re trying to do. Using that analogy, we want to transport people safely, efficiently, and quickly, with the lowest carbon footprint, at the lowest cost. We have to do the same thing with healthcare. 

Sara: Trying to change the engine on a flying plane is a perfect example of how this piecemeal approach might not work. Let’s take the example of primary care within the context of healthcare and health systems, which I know you’re very passionate about.

The traditional view, and what we did at my previous institution, was to think about it from discovery through delivery. That meant shepherding someone through the process of, “I need care. Now how do I get it?” Then we put them in a physical setting or a virtual visit. 

We didn’t have many degrees of  freedom or many tools at our disposal to do anything beyond the process from discovery to delivery. As a consequence, primary care was a heavily subsidized loss leader for the organization—and that is probably the case for pretty much every health system in the country.

Tom: The first issue that makes primary care unsustainable for most health systems is that they are hemorrhaging money on it. So, start with that as our first requirement: We can’t hemorrhage money into primary care as an overall set of solutions.

We often start with a patient raising their hand and saying, “I need help.” Most patients know how hard it is to get into healthcare, so psychologically, they’ve probably gone through the process of, “I know something’s wrong, and it’s bad enough that I’m going to deal with getting on a 9-month wait list to see a primary care physician, or maybe a shorter wait time to see an advanced practice provider (APP).”

But what about all the patients who say, “I think I’m probably okay. I haven’t had my mammogram in 4 years. I’m probably fine.” Or, “I haven’t had my colonoscopy. I’m probably fine.” Or the patients who say, “I think I’m good. Everything’s good. I’ve done everything I was supposed to.” And then there are the patients who actually are good.

Primary care addresses only one of those segments: We wait for patients to say something’s wrong. Then we make them go through scheduling, we make them wait to see a clinician, and we make them figure out everything that happens after that.

If we disaggregate the elements of primary care, we can decide which ones have the greatest yield. This is a bit of a sensitive subject, but we need to ask, “Which parts of the traditional primary care interaction add the most value?” I often hear primary care doctors say it’s the diagnostics, the labs, or components of the physical exam. Or they say it depends on the patient. Physicians typically say it takes cognitive intelligence to put some things together based on the questions the patient asks or the concerns they raise.

But why does that have to be an in-person interaction? 

I just had my once-every-3-years primary care visit—because I travel, and it was hard to get in—and I totally forgot one of the things I wanted to ask my doctor about. So now what do I do? Do I send him a MyChart message? How do I follow up? What if we think about doing this more asynchronously? Let’s take all the waste out of a physician’s time so they can manage the team of APPs and staff around them and allow them to interact with patients in much lower-intensity ways. 

This is where the last mile comes in. You can’t just say to a physician, “Do all this other management.” You’ve got to take the load off elsewhere. 

It requires changing the entire economic model. We have to address policy standards. For example, some payers won’t reimburse for labs if the patient hasn’t had a physical interaction with their primary care doctor. Do we just take it at a loss because we’ll maybe lose less money that way? Do we take it on with the payers?

Last-mile considerations like these are predicated on the best way to serve this patient in a timely, convenient, and financially sustainable way. There are lots of solutions, but we have to move multiple things at once. If you make it easier to find out information about primary care doctors in the local health system but don’t have any availability and the patient can’t schedule online, you haven’t really made it any better. 

This is true of every functional area. And clinical service lines—such as cardiology, oncology, and mental health—don’t have to operate under the legacy constraints. 

When we build a new model, we’ll have entirely new last-mile considerations.

Sara: There are a few themes that are important for health system leaders to understand and start to build strategies around: 

• One is this data-driven view of supply and demand. It’s not sufficient to put your finger up to see which way the wind is blowing or have a dated view without real-time data or a real-time understanding of the inventory or the modalities of your supply. (Some inventories aren’t traditional. They’re asynchronous or AI-powered, such as an agentic experience, for instance.)
  
  On the consumer or patient side, you need to understand what’s driving them at a deeper level than just, “They’re looking for care.” It’s intent, preference, motivation, and clinical, social, and financial needs all put together. The data-driven view is critical.
• Another is that a lot of stratification and personalization will go into getting the right patient the right care at the right time. As the degrees of freedom associated with AI-powered experiences increase, so will complexity. And the systems and instrumentation required for stratification will need to take advantage of the latest tooling.
• I also really loved what you said about aligning on the economics and affordability for the patient and what’s sustainable for the health system. Why do we accept that primary care or a lot of ambulatory care is unprofitable? By the way, in an H.R.1 environment, with all the changes with Medicare and Medicaid that are coming up in January, accepting unprofitability is going to be less of an option. Leaders will have to lean into creating this alignment.
• There is a major trust component here to AI evaluation. To some extent, the sensitivity with clinical teams you articulated hinges on whether we can trust AI.

Tom: Each element has multiple dimensions. People often focus on entry-point access—how to create more reimbursable slots for patients to see clinicians. But if you look at the cost of that slot, that unit of interaction—whether with a physician or an APP, whether virtual or in person—you’re losing money. Everyone knows that. Health systems lose money here and assume they’ll make it up on the back end.

But why do we accept that that’s the world we should operate in? In broad public research, physicians agree that in the majority of interactions, the patient could have been served in a different way. Physicians report that in the majority of interactions, patients could have been served asynchronously. They didn’t actually need to see the patient. The patient just needed some general guidance or insight. 

By disaggregating some of the insight into healthcare delivery, we can solve the economic problem differently. The question is how to reduce the cost of the reimbursable slot we lost money on so we can provide more slots. Because if the cost per unit is high, we’re not making it up in volume. We have a finite number we can provide—unless we’re going to change the entire economic system of healthcare and go all-in on risk-based models, and we’re just not there yet.

How do we make the entrance cost much smaller? How do we proliferate it or even make it positive? Imagine if, all of a sudden, those ambulatory interactions were cash-flow positive and we could make money when serving patients on their primary access needs. 

It would change the economics of all healthcare. It would then be a case of who could get to the patient faster and who could give them a better experience. The entire economic landscape for a provider would change. And we can do it. AI can drive those costs down dramatically. It’s the same when you think about evaluation and monitoring. It’s, “How do I learn to trust it as a patient? How does the clinician learn to trust it? How does a family member trust it?”

Some of that is technical, but some is just, “How do we make it easy, transparent, and clear what role the AI tools have?” There’s a lot of detail in that, but it all starts with not being constrained by the traditional delivery model and not giving in to the naysayers— because it can be done. 

Sara: We did it at my previous institution in a very narrow sliver around more on-demand care, where we understood deeply the intent a patient was coming in with. Using an intent recognition engine, we were able to parse out their need and point them in the right direction (and in some cases in a self-serve manner). 

Sometimes folks are just looking for information or content. Sometimes they have a very simple ask, and they don’t need to talk to a physician or even another clinical care team member. We directed them accordingly, speeding up response time and meeting patient needs. Patient satisfaction associated with those experiences increased.

On the health system side, patients with more complex needs were no longer having difficulty getting the care they needed. We freed up tons of appointments that historically were filled by patients who had sniffles, a follow-up question, or something that could have been addressed differently. Now everyone had better access for their needs. 

Tom: Two examples of provider organizations I’ve worked with spring to mind. One health system started bydissecting the pieces. The first piece is the assumption that patients are coming into the wrong site of care, the wrong place, which is constraining the number of slots. This health system looked at which patients could have gone to a different site of care, which would increase the scale and opportunity for access.

In this case, the clinicians and care teams were presented with the analysis. The health system then built the tooling to do shadow routing in the background, mapping out where the patient could go. They started to see what AI routing would recommend. Then the health system measured it against the variance, saying, “When we implement this, this is the return we expect.”

Then the health system had to identify where the technology touches the humans around it. That starts with focusing on patients coming in. Once leaders start to think about this on a bigger scale, the question becomes, “How do we get the patients coming in and serve them as efficiently as possible?”

After that, the health system should have a care plan for every patient who interacts with them. The health system can have all the data to know what should happen and when, then prompt that next action and activate people at the right time. 

More and more people have Apple Watches, Fitbits, Whoops, Ouras, or whatever their tool of choice. Those tools can provide data that health systems can use to say, “There’s something here you might want to check out.” It’s preventive, low initiation, and easy to do. So how do we start to think about moving further upstream? 

Health systems taking this approach have a much more defined population with lower-intensity needs that they can serve with a lower-cost operating model—and the entire operating structure changes. The cost model is entirely different.

Another organization we’re working with wants to create a digital concierge for all its patients coming in and moving between locations. To be clear, the digital concierge doesn’t provide care; it helps answer routing questions. Patients can reach out to it, and it will reach out to them. It’s a consistent input-output for all interactions. 

This organization has a narrow scope of care delivery, but its approach simplifies the patient experience. Patients moving between sites now have a consistent place to go, and it feels like a more cohesive experience as the organization changes some elements of the last mile and how it delivers care.

So there are real examples of healthcare organizations starting to transform operating models, and I think it comes back to the analogy of flying the plane. 

The right way to do this isn’t to change your entire multibillion-dollar health system or mission-critical aspects of the entire system that keep the plane in the air. It’s building the right solution for the future, building a new plane on the ground with the right experience and assets, getting it in the air as fast as possible, and letting it win on preference. Let physicians, patients, families, and administrative leadership choose it. 

Let everyone see this is the way we should be flying. 

Over time, you sunset the old plane, and you retrain the pilots to fly the new plane. You show patients, “Now you’re not in a metal tube. You’re in this voluminous open space. You can experience it differently.” It will take time, but we have to build that future model and show that it works, rather than trying to iteratively improve the old model.

Sara: And it’s not just about technology. What you described is all-encompassing. The new plane is not just the engine or the wings. It’s the whole thing. Effective AI activation is much more than just technology deployment. It’s much more than a partnership with a foundation model company. Those are elements, but there’s much more to it. 

Executing across all those pieces requires a much more comprehensive approach to redesigning the components so that they work together.