237: Inova Health's CIO/CDO Matt Kull: Why the Five-Year Vendor Contract Is Dead, Catching AI Model Drift Without Doubling Headcount, and How to Protect Trust with Good Governance

Subscribe on: RSS | SPOTIFY | APPLE PODCAST | GOOGLE | BREAKER | ANCHOR

On this episode of The Digital Patient, Joshua Liu, MD, Co-founder & CEO of SeamlessMD, and colleague, Alan Sardana, chat with Matt Kull, Chief Information and Digital Strategy Officer at Inova Health, about "Why the Five-Year Vendor Contract Is Dead, Catching AI Model Drift Without Doubling Headcount, How to Protect Trust with Good Governance, and more..." Click the play button to listen or read the show notes below.
‍

Audio:

‍

Guest(s):

• Matt Kull, Chief Information and Digital Strategy Officer at Inova Health
• Joshua Liu, MD (@joshuapliu), Co-founder & CEO at SeamlessMD

‍

‍Episode 236 - Show Notes:

[00:00:07] Episode preview

[00:06:00] The test Matt applies before betting on any AI investment — with technology moving as fast as it is, he asks two questions: are the inputs and outputs something he's comfortable with a machine deciding, and is a machine likely to be capable of deciding them in the short to medium term? He points to the day the iPhone launched and nobody predicting BlackBerry would be a relic three to five years later, and to the CEO of Citadel saying agentic bots now do quantitative analysis in hours that used to take teams of people a month. At Inova, the majority of appointment cancellation calls are already handled by a conversational AI agent, but booking net-new specialty appointments is not — he isn't yet comfortable that the inputs ("here's what I'm complaining about") reliably produce the right output of the absolute right doctor for that condition. "Where I don't see there being a human machine solution in the near field, we make that a human-oriented task."

[00:09:24] Why Inova's risk framework leads with trust rather than risk mitigation — three prongs drive every decision: how risky is it, how safe is it, and what does it do for trust. He's blunt that risk management isn't why anyone chooses a health system: "People don't come to any healthcare system because they've done the best job managing risk. They come there because, 'I trust you.'" Trust gets inspired by explainability — it isn't hocus pocus, it doesn't live in a closed box, you explain how it works, you get consent, and you can explain it internally too. He also notes he isn't suggesting his own no-go areas should be off-limits for everyone: people have to have their own risk schematic.

[00:11:38] What actually generates efficiency versus what generates headlines — the back office is frankly where most of the efficiency can be found, but the clinical space gets the attention. "Nobody cares that we're, like, automating denial reports. They care that we're finding tumors in images faster than anyone else, and they're both responsible." The clinical places where AI is trustworthy and explainable today are reading images, reading labs, and decisions on static data — not a unique, individual patient circumstance that should have a physician looking at it. His larger thesis: the real value in AI today is the "human middleware," the disorganized and disentangled systems people are sneaker-netting information between.

[00:13:35] Why Inova avoids the word standardization and uses "systemness" instead — with five main hospitals and hundreds of individual sites, the goal is for everyone to experience Inova care the same way, with predictable outcomes, experience, and quality, because in Lean and Six Sigma terms variance is where the bad stuff happens. Crucially, systemness doesn't start with technology or digital. Matt repeats one line to every executive in the organization: "Old processes plus new technology equals a really expensive old business with absolutely no gain."

[00:15:06] How Inova built its hip fracture care pathway, and the outcomes that followed — 130 stakeholders came together across the organization: clinicians, nurses, patient navigators, finance, advocacy groups, and patients themselves, looking at hip as a condition across inpatient, periop, EVS, nutrition, and physical therapy. The gold standard process they produced targeted faster identification-to-surgery (the national average is three days; Inova is under 24 hours), pain controlled without heavy narcotics use, patients walking straight away, discharge to home, and full independence within six months. Against two to three hundred thousand US hip fractures a year where those metrics are largely not attained, "At Inova, 100% of our hip fracture patients have metrics that are similar."

[00:17:30] Why technology comes second — and what it does once it arrives — technology was brought in only to anchor the pathway: automatic surgery scheduling on identification, appropriate prioritization of hip fracture patients, nutrition in place, physical therapy at the bedside the same day of surgery, and a physical therapy appointment the day of discharge rather than two weeks later. The workflow technology also flags outliers who have deviated from the pathway and automatically instructs someone to go engage and review the case. Inova has run the same exercise for hip, back, heart, and everything else, always starting with the clinical enterprise because it owns both the clinical and the experiential outcome. "When we think about systemness, it's not a tech-led initiative, it's a culturally led initiative."

[00:19:58] Whether true systemness is achievable at a 30- or 50-hospital system — Matt says it depends entirely on what the enterprise priority is. If the priority is outcome and that drives every decision, it's totally possible. If the priority is growth, outcome may not be first and foremost, and it may just be number of lives served. His intuition is that a system running many different clinical processes depending on site will find it pretty hard to anticipate comparable outcomes by site, or to really and truly have a standard of care.

[00:23:22] How he separates a visionary frontier bet from a distraction — the test is whether it ties directly to value or produces a lot of knowledge. The world's first 5G-from-the-ground-up hospital construction site at Cleveland Clinic was deliberately a knowledge-gaining exercise: a willing telephony provider, a construction company that wanted the profile, and a building still in foundational stages meant high visibility with very low impact and little extraneous cost. What it bought was an information base on how quickly laptop, PC, and medical device vendors would incorporate 5G, what continuous connectivity means for patients moving in and out of the hospital with remote monitors, and the answer to whether he could ever build a hospital without structured data cabling in it.

[00:25:35] Why quantum computing was, in his view, the only option that could be true — no two people's proteomes react to the environment or to therapeutic agents the same way, and at some point the in-silico discovery being attempted is frankly beyond the reach of classical computing; the only way to simulate the natural world in its real sense is inside a quantum state computer. When the initiative was announced he took roughly a hundred phone calls, "Half of them were, 'Matt, you're insane. This is going to end your career,' or, 'Matt, I can't believe you thought of that first.'" It fit a research institution comfortable being misunderstood for a little while, and it produced the first DARPA grant for quantum and healthcare, the IBM research relationship, and the Discovery Accelerator partnership.

[00:30:11] Why he thinks the long-term contracting cycle in health IT is over — Matt is candid that as a health system leader, "the plight of the startup, I'm not always certain is mine to bear"; his responsibility is to spend money wisely on solutions that benefit patients. He uses public-market examples: signing a massive long-term deal with a company whose stock price says the market doubts its viability can leave you embedded with a partner cutting expenses, headcount, and R&D, unable to innovate because it's trying to stay above water. The second risk is the exit — a great, highly innovative startup sells, and now you're attached to a conglomerate squeezing revenue out of an embedded customer base while the product dies off. "You think there's a lot of happy PeopleSoft customers these days? Think there's a lot of happy VMware customers?"

[00:34:56] How shorter contracts are meant to function as an innovation incentive — his logic is that the best way to keep a company pushing forward on its mission is to make it earn the business over and over again; a hundred customers locked in for five years is an annuity that inflates exit value, whereas re-upping every year or two happens because value is still being delivered. He also thinks investors need to change strategy and invest for the long term so companies can keep innovating. And he notes the ground has shifted underneath the whole question: "how many businesses got started within, like, a couple of weeks after people jumped on Opus and started writing some stuff?" The threshold to getting to market is far lower than it's ever been.

[00:37:01] The take-shape-make split that governs what Inova buys versus builds — three buckets: technology you take and turn on with no changes; technology you shape, where Epic sits ("you certainly sell me something, but I'm going to configure it in a way that jives with our business"); and make, starting from ground zero running internal DevOps with data pipelines, MCP, and the rest. Today it's roughly 20% take, 60% shape, 20% make. If he changed anything, he'd move to 70% shape and 10% make, because there are a lot of good vendors building this stuff.

[00:37:59] What happened inside Inova's walls in a single month, and why he calls it an amazing thing — two physicians and one person from finance each came to him saying they'd written software they thought would benefit the organization and asked him to check whether it was safe. His answer was that he's the wrong person, but the AI and data governance structure exists for exactly that. Rather than resist, Inova is embracing it: build a structure around citizen developers and around agents with reputable, explainable data sources, then wrap the whole thing with observability. His reasoning is practical — he'd take software engineers embedded on the front lines all day long, and since that isn't possible for hospitals across the country, frontline people saying "here's a problem, and I was able to create something that sort of could solve this problem now. Can we take that and make it industrial-strength?" is the next best thing.

[00:39:44] Why cheaper prototyping changes the economics of being wrong — "I have probably had more blunders than I care to count. I'm just very happy that my win rate is significantly higher than my failure rate." He's comfortable buying some things that turn out to be poopers because the wins are bigger, and clear about the discipline that follows: don't throw good money after bad trying to save something that's sinking — move on, adjust, and pivot.

[00:40:53] Why he frames governance as a way to preserve trust rather than reduce risk — "we are in the trust business," and if you can develop technologies that are trustworthy you're ahead of the game. He says plainly he can't be the person governing all of it because he's frankly not smart enough; instead the data and AI governance structure is predominantly business constituents across the enterprise, organized as big committee and small committees (clinical safety, ethics, legal, business, finance). Big committee answers one question — is it safe and is it trustworthy — and the smaller question of whether it suits the business rolls up underneath. "Things get embraced by trust, not by the amount of risk you carve out of something." He admits Inova probably over-governs, and thinks that's the right call for now in a space this fast-moving: "we are more cautious than we are free at this point."

[00:43:47] How Inova scales AI oversight without scaling headcount at the same rate — with generative AI output being variable rather than deterministic, monitoring is handled digitally through a partnership with an organization called Signal One, which monitors AI for utilization, drift, and accuracy and works closely with Inova's governance. His constraint is economic: "I can't make the cost of the machine doing it twice as expensive as the cost of humans doing it." That doesn't mean no eyes on — each vertical has assigned AI product managers responsible for recovering monitoring results, reviewing with stakeholders, and confirming that what was predicted is actually happening. The structure is mandatory for any AI feature or function entering production.

[00:45:35] The demographic collision he says is less than 10 years out, with the numbers behind it — counting from 2023, the US population aged 50 to 90 was set to grow from 45 million to 70 million people, roughly a 40% increase, and those are precisely the ages at which people use the healthcare system most heavily. At the same time fewer people are entering the profession, with a national shortage of 88,000 physicians projected by 2035. Between retirement pace, aging demand, and labor shortages, the healthcare system will have to find a way to meet that demand with the resources it has.

[00:47:05] Why he believes technology is an imperative rather than an option, and where the capacity will actually come from — providers seeing patients is already maximized, and he questions how much more human burden can be placed on people already at high risk of burnout. The opening is the enormous amount of human middleware: instead of the coding group sending something back to a provider after the fact, tell the provider up front that a care gap exists and they're saying the right things to close it. His example is a recommendation surfaced in the moment — you're about to diagnose Bob with COPD, but his plan doesn't cover it until his oxygen saturation is a couple of points lower, so here are other recommended treatments as the condition progresses — still the provider's decision, because nobody walks into an appointment planning to read their health plan's clinical criteria. He expects public-private partnerships, incentive structures to draw people into healthcare, and policy work alongside it, and closes on inaction being the bigger risk: "the risk of not stepping in and not being creative and not trying, even if it doesn't work out the way we hope to, is far safer than not doing anything... I do not think it's going to be a wait and see game anymore."

‍

‍Fast 5 Lightning Round:

1. What is your favorite book or book you’ve gifted the most?
   The First 90 Days by Michael Watkins
2. If you could instantly master any skill, what would it be?
   "Piano"
3. Would you rather have Super strength, super speed, or the ability to read people’s minds?
   "Super speed."
4. What is something in healthcare you believe others might find insane?
   "I think that the consistency of the, quote-unquote, 'standard of care' is far more variable than people think."
5. What is the last movie or TV show you saw, and what did you think of it?
   "Yellowstone. It was awesome."

The Digital Patient has been recognized as Feedspot's #1 Patient Engagement Podcast of 2025. Thank you to our listeners for making this happen!