Abby Burns (00:13):
From Advisory Board, we're bringing you a Radio Advisory, your weekly download on how to untangle healthcare's most pressing challenges. I'm Abby Burns. We talk a lot on Radio Advisory about the provider and the plan landscapes, but there's another sector with meaningful influence in our industry that we need to talk about, and that's big tech. Because for as much as we have examined AI and how it is or isn't yet shaping our businesses, we really haven't taken the time to unpack exactly how we're getting this technology, or, to be more precise, courtesy of whom. The majority of health systems and plans report using at least one, if not more AI solutions today. The vast majority expect AI to become a more integral part of doing business moving forward, and there's certainly no shortage of companies that will help them do that. But who's ultimately powering that business?

(01:08):
We have a habit in this industry of only looking in the mirror. What's happening with AI and healthcare? Let's look at what other organizations like mine are doing. But at Advisory Board, we'll argue we need to expand that lens. When it comes to AI, this is particularly important because we know that there are a million and one vendors in this space, so it can feel like as the big fish, we're driving the bus here. But when we look under the hood, we see a little bit of a different picture. So, to unpack the shape of the health tech landscape and talk us through the implications, we've invited Advisory Board AI expert Ty Aderhold back to the podcast.

(01:46):
Hey, Ty. Welcome back to Radio Advisory.

Ty Aderhold (01:48):
Hey, Abby. Thanks for having me back.

Abby Burns (01:52):
Ty, it's been a few weeks since you were last on. Is that a record for you?

Ty Aderhold (01:59):
As long as there continues to be AI in the news I sort of expect to be back regularly.

Abby Burns (02:04):
That's what we call job security. Ty, you called Ray and me a few weeks ago, admittedly after the Hugging Face incident, and you said we need to talk about the AI vendor landscape. We certainly talk about AI a fair amount on Radio Advisory. You talk about AI a fair amount on Radio Advisory, because in reality it does sort of permeate every strategic facet of healthcare. But we spend most of that time talking about the what and the how of AI, and not so much the who, as in who is actually supplying the tools and the underlying technology. Why did you want to focus on this angle specifically?

Ty Aderhold (02:43):
Yeah, I think in healthcare we, or I would say most leaders don't necessarily appreciate the extent to which the healthcare AI ecosystem that I am always talking about and it's on a lot of people's minds is dependent on big tech. When I say big tech, I am including OpenAI, Anthropic, these big companies with the foundational models as part of big tech.

Abby Burns (03:10):
Why is it important to understand how and maybe where these players are showing up?

Ty Aderhold (03:16):
The biggest thing to understand here is that many of the AI applications that healthcare organizations are using today, are hoping to use in the future are dependent on some of the foundational models from these companies. And so as you think about the world of applications that are available today in healthcare, these organizations and what happens to these organizations, the decisions they make, whether they go public or not, all has an outsized influence on these healthcare AI applications, whether or not you are partnered with them directly.

Abby Burns (03:56):
And to kind of close the loop there, what happens with these companies affects the AI applications that you're using, which means it affects ultimately your business. Ty, to put you on the spot for a minute, we talk a lot about AI as I said, but I don't expect that everyone has a robust AI dictionary. So to level set for our listeners, when you say foundational models, can you just be clear about what that means?

Ty Aderhold (04:20):
Yeah, so this is your GPT-4, this is Claude, a couple of other big ones that are out there, Google Gemini, X has one, Meta has one. There's only a few of these big, very large tech companies that have developed these massive foundational models, mainly because the training and the resources that are required to build these are massive. And so when you hear about all of the investment that is going into it and how much money these different large companies are spending, it's all around that model training. And that's why there's only a few models out there that we see being powerful in a number of different areas like a ChatGPT or a Claude.

Abby Burns (05:09):
I appreciate that you spelled that out because it kind of rounds out the question of who we're talking about. One name I will say I expected to hear in that list, you said OpenAI, Anthropic, you said Google Gemini, you didn't say Microsoft.

Ty Aderhold (05:24):
I think Microsoft belongs in this big tech group. In many ways, they are serving as the main pass through right now for a lot of OpenAI technologies. They've had a deep partnership with OpenAI for years. It used to be exclusive, but still many organizations are accessing OpenAI chat tools through Microsoft Azure. And so there is a classic connector piece there. And so very much I would say Microsoft is tied in to some of these big tech and large language models in a real robust way.

Abby Burns (05:59):
And that's almost a foil to where you started, which is what happens with OpenAI ultimately impacts what happens to healthcare businesses, in a very similar way it sounds like what happens with OpenAI matters for Microsoft's business. So you said that these organizations have outsized influence on payer and provider businesses that is likely to only ramp up. I don't hear about a lot of major divestment or slowing down of healthcare organizations looking to use AI. Let's make this real. One way we could do this is go solution by solution and show how and where OpenAI or Anthropic is involved. I'm guessing, knowing you as I do, that you have a more constructive way to think about this.

Ty Aderhold (06:42):
I think there's two big buckets that we need to be thinking about here of how they're shaping healthcare organizations. First is they are powering a lot of the AI solutions you are using or want to use for business operations. The second is stepping away from business operations and into the consumer world. They are shaping how patients, how members interact with healthcare organizations and how they engage with healthcare.

Abby Burns (07:12):
I want to talk about both of these and let's start with powering business operations. What are a couple examples here?

Ty Aderhold (07:21):
Abby, I'll give you three. The first one, let's take Abridge, a common ambient listening solution that's out in the market. They are working and partnering directly with OpenAI. So when you are working with Abridge, you are somewhat dependent on OpenAI solutions to help power all of the different impactful things Abridge is doing in the Ambient space.

Abby Burns (07:44):
And I think if I'm remembering correctly, Abridge does obviously do Ambient, but they do other things to support provider operations as well.

Ty Aderhold (07:51):
Right. And I think that's a trend we're seeing in general from Ambient solutions. It's not just Ambient, there's a whole package around it and a variety of solutions they're giving you. And most of the time that is because they are using some of these foundational models from different partners they have in order to build out those other capabilities.

Abby Burns (08:10):
Okay, so that was one example. What's another?

Ty Aderhold (08:14):
The next one I'll give is a little more hands-on. So if we take Palantir, for example, they've started adding healthcare offerings. Well, they'll come into your organization, help you with data, but also build AI solutions. They are not using Palantir specific models most of the time in those scenarios. They're reliant on some of these foundation models from different solutions as they go in and help you build new solutions for you. So even a big company like Palantir, we would think, oh, maybe they're doing everything internally. They're also still using some of these foundational models.

Abby Burns (08:51):
It's interesting that you went there because I actually don't think of Palantir when I originally think health tech.

Ty Aderhold (08:57):
Right. I think it's a changing world, Abby, in many ways, and I think we're seeing organizations that are great at data or great at AI move into healthcare. And that actually leads me to the final example here. We're seeing Anthropic out in the market directly trying to partner with healthcare organizations, health systems and health plans as a, "Use our models and use Claude and Claude Code to develop your own agents internally to help with your business operations across a variety of needs." And so you might choose as an organization to go partner directly with one of these big tech companies to help you on the business ops side.

Abby Burns (09:40):
There's one use case that you didn't name, and it's somewhere I can sort of hear in my mind's eye pushback, well, hey, our predictive analytics use machine learning without being partnered by something like OpenAI, for example. How should we think about that?

Ty Aderhold (09:55):
I think it's a fair pushback in some ways. Many organizations probably do not realize the power and the extent of using just predictive analytics and machine learning. The big caveat here is your ability to scale those repeatedly. Do you have data scientists in-house? Are you able to build these yourselves? Are you reliant on other organizations building them? And what will that start to look like? I think if we look a few years down the road as well, the other thing in my mind is how are these big tech foundational models going to change how solutions are built in the first place in terms of dependence-

Abby Burns (10:33):
Oh, interesting.

Ty Aderhold (10:34):
... for coders and other things and creation of new solutions on the backend?

Abby Burns (10:39):
So if that's on the business operations side, let's talk about the other side of the coin and the influence with the patient, member, consumer relationship.

Ty Aderhold (10:49):
Yeah. So on this side, I'm going to use OpenAI as our example here, and there's two different offerings that I think are going to paint the picture for what is happening.

Abby Burns (11:02):
I think I know what they are.

Ty Aderhold (11:02):
Yes, you might be ready for this. The first one is ChatGPT Health. They recently rolled this out this year, and this is a direct to consumer. You can go with your healthcare questions, interact, get information. It's very much marketed and built to be a replacement for the, "Oh, I need to go Google what I have and figure it out," for consumers. And this is going to, and we're already seeing it happen, shape healthcare interactions. It's going to change what patients and members decide to do with their healthcare, and they're going to bring those responses from ChatGPT into their healthcare visit or into the phone call with a plan.

Abby Burns (11:44):
A couple thoughts on that, Ty, because ChatGPT for health or health in ChatGPT is distinct from generic ChatGPT. This is a designated consumer offering, but we also know, at least according to OpenAI's numbers, that 300 million users a week ask ChatGPT Healthcare related questions. So I think it just goes to show there's an enormous market as if anybody needed convincing of that at this point for chatbot-based interactions to influence the care journey, if you will.

Ty Aderhold (12:14):
Yes. I think we are in a new world of what patients and members expect or want in terms of their, "I have a question about my health" first line of next interaction. And I think we're seeing healthcare organizations realize this as well and OpenAI is realizing, and so they are also rolling out what I would describe as B2B2C offerings. So this is OpenAI for healthcare, for example, which is OpenAI saying, "Hey, healthcare organization, we know you want to have a chat interface for your patients and members. You can use our chat interface as your chat interface for members. So we're going to give you the same solutions that we're putting out D2C in the market and we're going to let you use them, we're just going to sell it to you."

(13:06):
And so what you're seeing is at the same time, OpenAI is both saying, "Hey, we're going to go direct to consumer with his offering, and we're going to go to businesses who are trying to reach those same consumers and offer it that way." So they have ChatGPT for Healthcare and then broader APIs that allow you to connect in and pull various things from OpenAI's offering into your own organization.

Abby Burns (13:32):
So these are two commercial offerings that OpenAI is taking to market. Outside of what OpenAI is directly selling, whether they white label it or not, are other direct to consumer or B2B2C solutions still powered by these same small selection of companies?

Ty Aderhold (13:49):
At this point, these large language models from these big tech companies are the gold standard of what a consumer expects in terms of interaction. So if you are a healthcare organization out there that is trying to and wants to have an AI solution that patients or members interact directly, I almost guarantee you're going to be building it with help or support or somehow a pass through to one of these foundational models because that is what is expected from the consumer preference and experience side.

Abby Burns (14:24):
Let's weave Epic in here because I think when we're talking about healthcare and technology, we can't go too long without referencing Epic. And I know that Epic is rolling out a lot of AI enablement, including consumer-facing or patient-facing tools.

Ty Aderhold (14:41):
With Epic, they have Emmie, which is their patient-facing suite of offerings, and that is going to be reliant on OpenAI and some pass throughs for ChatGPT. I don't know if long-term we're going to see Epic stay at just the mostly an OpenAI partnership or expand their partnerships offerings. They recently rolled out their agent suite, which allows the organization to pick which foundational model they want to use as they build their agents.

Abby Burns (15:10):
Oh, interesting.

Ty Aderhold (15:11):
So that's more foundational model neutral. But even when you are model neutral as an offering, there's only so many choices you're going to pull from. So at the end of the day, it's still dependent on this group of large tech companies.

Abby Burns (16:19):
The other organization that we haven't talked about yet that comes to mind for me is actually Mayo Clinic, because I remember seeing a few months ago that Mayo Clinic is investing in building out its own foundational model, deciding to invest the resources and the training that you mentioned at the start of our conversation to build out specifically a medically trained model.

Ty Aderhold (16:40):
Yes, I think it's exciting news. I think it's something that in healthcare AI circles has been talked about as an interesting place to be heading. The one big thing I will call out though is they are partnering with Microsoft in order to do that. I think that shows the required resources and scale necessary to build one of these models almost forces an organization into some sort of partnership.

Abby Burns (17:09):
When we're talking about such blanket dependency on a few organizations, we've got to be talking about risks. What are the major flavors of risk that we should be thinking about here?

Ty Aderhold (17:21):
The first one that I will start with, because I'm guessing it's on listeners' minds based on news cycles in the past few months, is the cyber security side of things. We've all probably at this point read the stories or heard the stories of Hugging Face and other instances of some of these agents going rogue and potentially hacking different organizations. The big caveat I will start with upfront is all of the incidents we've seen so far are on these models that are still being trained. So these are research models, they're internal to these organizations. It's still, I would say, somewhat scary, but these are not the publicly available models that healthcare organizations are out there using today.

Abby Burns (18:04):
Okay, that's a helpful clarification.

Ty Aderhold (18:07):
I do think it still has to be a part of the conversation here. We have seen OpenAI recently announced that they were not going to roll out the latest model that they had planned to roll out, so they are taking precautions. I think it is something to be aware of or part of the overall risks you need to be considering.

Abby Burns (18:27):
Yep. Okay, so cyber is a major category. What's another one?

Ty Aderhold (18:32):
Before we move on, there's a separate, I think, cybersecurity concern for organizations that need to be considering, which is the dependency that we've been talking about. There's only a few organizations here that we're discussing in terms of who is offering these foundational models. So we're very dependent on an OpenAI or an anthropic. And some listeners might already know where I'm going here, but this is not a new to healthcare of the potential risks of being highly dependent on one or two organizations if we think back to Change Healthcare a few years ago and the ripple effects of that and how many organizations were impacted because of that single dependency.

Abby Burns (19:13):
And it was a single dependency on an integral function to how individual organizations worked and how organizations worked together because it's foundational, not to confuse our words here, but it's foundational to the way we pay for healthcare.

Ty Aderhold (19:26):
Exactly.

Abby Burns (19:27):
Okay, so plenty of cyber risks that we need to be cognizant of. This is one of the reasons why chief information officer, chief digital officers, lots of healthcare roles are becoming more and more important. What are some of the other risks?

Ty Aderhold (19:42):
The other big one that I think everyone needs to be thinking about is on the economic side. To name it directly, these organizations right now are not making money. They have certain prices set for how much it costs to spend a token, but those prices are not such that without investor money, they could survive as an organization. They're dependent on investors right now. Now, there's theories out there of how with computing getting better, it'll drive down costs, all these other theories, but as of now, these companies are still dependent on investors. That is especially important as we are hearing all of these stories in the news lately around safety and concerns. It's also very important as we've now seen these organizations, OpenAI, for example, saying they're going to delay their IPO until they feel like they have safe offerings that they can go out to the market with, which all sort of adds into a picture of, I would say, economic instability for these organizations.

Abby Burns (20:44):
I think there's another economic risk here, Ty, which is more on the pricing side of things.

Ty Aderhold (20:51):
Yes. With this economic instability, there is a unknown what is the cost in the future going to be of these solutions? Will they have to raise prices at some point? I think that matters in two big ways. One, you might become dependent or you might have staff who really love a solution that you have rolled out. Let's take Ambien listening, for example. I think it's a classic one of most clinicians love it, they want it, they're asking for it. I think it's becoming table stakes offering. But, let's say it's two years from now and it's double the price it was today. How are you as an organization going to be able to make that decision of, "Well, do we take it away now that our clinicians are used to it for two years, now they've changed their workflows towards it?" That's a tough situation to be in.

(21:44):
I think the second piece is understanding the ROI of these solutions over time. So ROI, very much dependent on the input costs here. Let's take some of these revenue cycle support solutions. There's a lot of solutions out there that help revenue cycle staff, coding assistance, other things make them much more efficient. Those have a price today, and we see organizations say that is worth it to us, and you might be able to say, okay, we're going to bring this in and we're not going to need to hire more revenue cycle staff over the next couple years because we now have this solution in place. But if the prices go up there on that solution, you could end up looking at a scenario where it makes more sense for you to be paying for another coder, another human to be in seat as opposed to these solutions, but you'll have gone years without that hiring, without having those individuals.

(22:41):
So I think the other risk here is you're betting on AI solutions to make you more efficient with an unknown ROI from the cost side of what these solutions are going to be in the future.

Abby Burns (22:53):
What I imagine is maybe more reliable to bet on, Ty, is that over time, compared to where we are now, the prices of these solutions will go up.

Ty Aderhold (23:02):
I think they almost have to. If you look at the history of huge new tech companies coming out of Silicon Valley that have been investor propped up for years and they pivot into being profitable, there's two real ways that they make that pivot. It's either ad revenue or increasing prices.

Abby Burns (23:26):
Add like advertisement, not additional revenue.

Ty Aderhold (23:28):
Yes. Advertisement revenue or raising prices. I think a classic example of the raised prices is Uber. For years, Uber prices super cheap, really changed the consumer perception of what getting from point A to point B via a vehicle should cost, but then prices went up after they left being dependent on the investors and wanted to start turning up profit themselves.

Abby Burns (23:54):
Ty, as we start to wind down our conversation, I want to switch from almost the skeptic or the pessimist lens and look at big tech involvement in healthcare through more of an opportunity lens. These are large and very well capitalized players. They are definitely moving fast, but they're also in the early innings. I mean, the solutions that you talked about before, ChatGPT for Health, ChatGPT for Healthcare, they were rolled out in January and April respectively of this year. That means they are still figuring out their healthcare solutions.

(24:23):
How should health systems, health plans, some of the more incumbent players in the healthcare industry be thinking about the opportunity to inform these players' healthcare solutions both now and over time?

Ty Aderhold (24:35):
I think there's two obvious ways to me. The first is very much have the conversations around this economic instability and some of the ROI concerns you have. I think a pretty common refrain I'm hearing in the market is that the token-based pricing that most of these organizations offer is really, really hard for organizations, especially as they scale solutions across their whole organization. When you go from 10 users to a thousand users, all of a sudden it's a lot more unpredictable what your month-to-month cost is going to be. And so have that conversation with any vendors you're working with, whether it's one of these big tech vendors directly or a more specific healthcare vendor.

(25:24):
The second thing is these companies are out there looking to partner directly with healthcare organizations.

Abby Burns (25:30):
Yes.

Ty Aderhold (25:31):
I'm not saying that is the right answer, but certainly they are coming to the table and wanting to partner and wanting to have conversations. So I would say if you're considering it, use that to your advantage to help shape the potential offerings, to have a better understanding of where these organizations might be headed, because I don't think we're going to see them move away from that direct partnership type approach in healthcare. I would expect your OpenAIs, your Anthropics, they're still going to do the direct to consumer stuff, but increasingly we're seeing signs they want to work with healthcare organizations directly.

Abby Burns (26:08):
I think it's also worth noting when we look at... Stick with the provider side for a minute. When we look at the provider landscape in the US, we've got a lot of different types of providers, whether it's different sizes, different geographies, different patient populations that they serve, and it's important for these solutions to be able to cater to or tailor to all different types of providers. I think there's almost this conception or misconception that it is only going to be the largest academic systems that are going to have the door open to them to partner with these companies. I would love to encourage listeners, if you're coming from a critical access hospital, if you're coming from a safety net hospital, these companies need to understand your realities, your challenges, et cetera, as well. I think there's opportunity for provider types across the spectrum.

(26:55):
Ty, we've covered a lot of ground today. If we had to distill down, what is a message you want listeners to leave remembering?

Ty Aderhold (27:05):
Abby, I want folks less focused on the news stories and specific incidents over the past few months, but more want to use those to actually shine a light and show folks how big an influence these organizations have on your business today and where you're hoping to go with your business in the next few years, even if you're not directly contracting with them. I also think it's important to note that these big AI companies are going to be who is setting the rules of the industry and how the industry is using AI in healthcare now and in the future, and they can't be ignored.

Abby Burns (27:47):
Well, Ty, thank you for coming back on Radio Advisory. I'm sure we'll have you back soon.

Ty Aderhold (27:52):
Yeah. Until next time, Abby.

Abby Burns (27:57):
To have a more nuanced conversation about AI, we have to understand the state of the AI players. There are hundreds of AI applications for healthcare at this point, but when we look under the hood, they all come back to a small handful of companies. The foundational models from OpenAI, Anthropic, and Google are not just shaping your patient's experience with direct to consumer tools like chatbots, they're powering the AI tools you use to interface with your patients or members and to run your own business. That means these companies actually have a ton of influence over your organization, whether or not they show up explicitly on your vendor registry.

(28:35):
The last thing Ty said is sticking with me. These are the companies who are going to be setting the rules for the rest of the industry. As that happens and as you figure out your place in that shaping, remember, as always, we're here to help.

(29:14):
New episodes drop every Tuesday. If you like Radio Advisory, please share it with your networks, subscribe wherever you get your podcasts, and leave a rating and a review. Radio Advisory is a production of Advisory Board. This episode was produced by Rae Woods, Chloe Bakst, and me, Abby Burns. The episode was edited by Katy Anderson with technical support provided by Dan Tayag, Chris Phelps, and Joe Shrum. Additional support was provided by Dominique Del Gaudio. Special thanks to Cooper Hageman. We'll see you next week.