LRVHealth’s Keith Figlioli on the Evolving Landscape of AI Procurement

LRVHealth is a venture capital firm whose investors include health system and payer partners. Managing partner Keith Figlioli, M.B.A., M.P.H., recently spoke with Healthcare Innovation about the need to take an architectural approach to the evolving landscape of AI procurement in healthcare.

Before joining LRVHealth, Figlioli was senior vice president of healthcare informatics at Premier Inc. (NASDAQ: PINC), where he led the company’s enterprise healthcare technology business unit and helped the company raise $820 million in its IPO.

Healthcare Innovation: Keith, you wrote something recently that I found intriguing. I think your point was that healthcare leaders have to rethink how they evaluate, procure and govern AI across their organizations. I am hoping we can dig into some of the implications for those health system leaders, for vendors and perhaps for venture capitalists like yourself, too.

You described a framework for AI procurement that had three distinct layers: core enterprise platforms, foundation model platforms, and specialized use cases. The core enterprise platforms include both the hyperscalers and the enterprise platforms such as Epic and others. Can you talk about how that layer is changing?

Figlioli: The first wave of any technological change is frantic and crazy. Regardless of what people think, they don't really know what's going to happen. But as we get into the second wave, or the next maturation cycle, I think there is some clarity here. You have to think about architecture to get the most out of AI. You have to think about architecture first before you can think about a point solution or even a platform solution. That involves thinking about security, privacy, and the  provenance of data.

If you think about layer one, it involves all the systems of record that we have in healthcare, like Epic, Salesforce, or ERP vendors. 

The thing that I think people are missing, especially in healthcare, is the hyperscalers, but in particular Microsoft. The reason why Microsoft matters so much, in my opinion, is that health systems and payers — almost all of them have an enterprise license from a productivity suite standpoint for Microsoft. So all their e-mail, all their calendar, their team's infrastructure, everything is in that environment. So what is Microsoft doing? Microsoft is building Copilot. They're building Copilot Studio. They're building Work IQ. They're building all these things to make you lock into their environment even further, which I think is a smart strategy, by the way. 

Google is doing the same thing with their coding environment and their agent studio. If you just take one step further out, all of these systems of record and hyperscalers, are starting to build agent studios. To me, the question over the next few years is how do you hook into that? 

HCI: OK, let’s move on to layer two…

Figlioli: Layer two is the foundation model platforms. These are players like Anthropic and Open AI, and they don't just want us to embed their models into the hyperscalers. They want to go to market to healthcare systems, payers and life science as well. You know, Dario [Amodei, CEO of Anthropic] said this past week that his No. 2 industry is going to be life sciences in terms of the amount of effort and time and energy.  How this second layer stitches into your core systems of record is quite complex, because your systems of record and your hyperscalers are also embedding these foundational models into their offerings.

HCI: I think you described those foundation model platforms as wanting to function as orchestration layers. Can you describe what you mean there? 

Figlioli: If you have multiple systems of record serving up data and also starting to serve up various pieces of agents, those are all different types of companies. Who's going to be the orchestrator of all of that? Could that be the foundation model companies? Could that be a third party? There's a handful of recently funded third-party companies that are thinking about this problem. Something is going to have to become a coordinating agent. There's never been true middleware in healthcare IT. I think we're at the early stages of experiencing that. I have no idea who's going to win, but I do believe that this orchestration layer is going to become a true middleware category for healthcare technology over the next three to seven years.

HCI: OK, so layer three is the specialized use case innovators such as people focusing on prior auth or claims management or revenue cycle automation. What are some of the implications for startups that want to attack a relatively narrow problem like that? Do they have to account for what's going on in these other two layers? What do they have to be aware of in terms of how their solution might fit into a broader picture?

Figlioli: I think that's where a lot of people who knock on our door got a big-time free pass for the last couple of years. There weren’t many questions asked. Now in this layer, there are going to be a lot of questions asked — architecturally, how do you fit into workflow? How do you integrate? 

There is a reason why some vendors in spaces like specialty pharma or prior auth are doing so well. It’s because it's so important to the bottom line of some of these players, and they're going to allow them certain permission factors to go faster, because it's helping them on their bottom line and helping patients get access to medications faster than otherwise. 

The lion's share of the investment dollars have gone into two categories of the first wave. One is ambient scribe, which we all know, and the other has been all the things around rev cycle, coding, documentation and denials management. We talk to these startups every day in our job. They’re going to be asked a lot more hard questions as the maturation of the industry takes hold and the architecture takes hold. And the question is, who holds the keys of the kingdom? We have seen historically with Epic they have held a lot of keys to the kingdom. But now, if the hyperscalers wedge in and start doing more or the foundation model companies in layer two start doing more, they may hold more keys to the kingdom t or at least gain some parity with Epic. And I think that's still to be determined. 

HCI: Will this all have an effect on how CIOs and CMIOs think about build-vs.-buy decisions? Is it going to be easier for them to build applications or agents internally than it has been?

Figlioli: That’s the best question yet. I think we're right back to where we were 20-plus years ago, when people said maybe we should just create an internal development team and build all this stuff ourselves. And some people did. I think the debate is still wide open. For many people who have had vibe coding experience with some of these new tools, it's still not easy, connecting it to a back-end database, getting into a runtime, enterprise caliber. It is still a lot of work. You still need a lot of developers to do this stuff the right way and stitch it into an enterprise architecture. I still have a hard time thinking that a lot of our health systems are really going to end up building a lot of things on their own, but we’ll see.

HCI: So you say to startup execs: ‘Hey, you don't just have to build a good mouse trap; you have to figure out how you're fitting into the architecture of the health system.’ Are they surprised to hear that? Or has that occurred to them already? And if so, do they have an answer?

Figlioli: I think some have thought about it. To be really direct about it, that’s our screening criteria. So if you have a very sophisticated startup founder, they're thinking through a lot of this stuff. We didn't get smart on a lot of this stuff by ourselves. We talked to our network, and we talk to 800 to 1,000 companies a year, and they're helping educate us on some of the things that they're having to go through to be able to get a contract through. I think both sides of the circle — the incumbent industrial medical complex and the NewCo startups — are getting smarter about all this. It is no different than in the post-Meaningful Use electronic health record era, when people got smart and realized they had to integrate with Epic or be in Epic’s app store to be relevant. 

HCI: Well, it seems like some of this hasn't shaken out yet. For the startup exec, what’s the right answer when you ask: Are you set up to fit into the architecture that's developing?

Figlioli: I think you're 100% right — it’s not sorted out. This is a little bit of the Wayne Gretzky moment for us, which is trying to tell you where we think the puck is going.

People are getting more sophisticated about what scale looks like. It no longer looks like a thin-veil UI on top of an LLM. I think this past week Anthropic dropped 10 agents in financial services that can wipe out a handful of companies. That is the type of the disruption you're going to see in multiple sectors. 

The real lesson is that this stuff is just moving so fast. But getting back to our architecture point, you can move even faster and scale even faster if you understand the environment you're about to drop into.

HCI: Within these sophisticated health systems that LRVHealth works with, who's usually leading the charge? Have they hired a chief AI officer? Or is it the CIO? Or is it a committee?

Figlioli: I don't think there's one model that’s taken hold. But I had dinner with a bunch of CEOs of health systems last week, and every single one of them is thinking about this, and they're also thinking maybe it's not just an IT thing. What may distinguish the winners and losers in this is whether you have a very enlightened CEO who is driving a huge portion of this agenda, and not just leaving it to the technology and information team.

HCI: Will the changes you are describing have an impact on the number of VC deals that we'll see in healthcare? 

Figlioli: I think we've already seen it. You’re seeing fewer deals, and you're seeing more concentration of capital with not much scale yet. People are trying to place early bets on perceived platform companies. The first inning is all about how much capital you can pull. And if you can pull that much capital, can you make yourself a winner? We're seeing, rounds of capital come together of $50 million-plus, and some of these businesses don't even have $1 million or $2 million of revenue, while others are accelerating into revenue faster than they ever had in the past. 

HCI: Earlier, you mentioned AI scribe companies being one of the early success stories. But do those companies have to evolve to do much more than that to survive? 

Figlioli: It’s a very open-ended question, because of what Epic and Microsoft announced. We had a bunch of forward-leaning vendors that were way ahead, and they even partnered with Epic and others. Then all of a sudden, Epic decided that this is part of their core workflow — just like a page out of Microsoft's book. So to your question, what else do they have to do? Every single one of them from a strategic lens is trying to figure out, what else do I have to do now? Do I have to get into revenue cycle? Do I have to get into prior authorization? You can see some of these things that are very logical for some of those vendors, like coding. There are many autonomous coding vendors. An ambient scribe vendor picks up the first set of signals on an episode. The next logical step would be the autonomous codes that come off that. Most of those ambient scribe vendors, with the exception of one, have not built that. Could they go acquire that? Could they go build that? I think you're going to have to watch that space pretty closely to figure out who's going to do what to stay relevant, and, more importantly, to grow into their valuations.