Lonny Northrup has spent his career at the intersection of technology, data, and healthcare transformation long before AI became a boardroom imperative. Today, as an AI Governance Strategist at Inova Health, he is helping one of the nation’s largest regional health systems navigate both the promise and the risks of generative AI. In this conversation, Lonny shares candid perspectives on what healthcare executives still misunderstand about AI, where the technology is already delivering measurable value, and why governance, not hype, may determine who succeeds in the next era of healthcare innovation.
Inova Health is a nonprofit healthcare system serving Northern Virginia and the greater Washington, D.C. region. Founded in 1956, Inova operates five hospitals, including its flagship Inova Fairfax Hospital, as well as a network of primary care, specialty care, urgent care, and outpatient facilities, and employs more than 25,000 people.
Q & A With Lonny
Judy Kirby: Lonny, thank you so much for speaking with me today. Your career has moved across manufacturing, aerospace and defense, enterprise architecture, healthcare data, medical informatics, advanced analytics, digital therapies, and now AI governance. How has that path shaped the way you think about technology’s role in healthcare?

Lonny Northrup: For me, the focus is always on achieving results. Across all the industries I’ve worked in, I’ve pursued impactful outcomes. Technology keeps evolving, and now with AI, we’re seeing significant new capabilities and cost shifts that are delivering real impacts.
Judy: You were working on big data roadmaps to reduce healthcare costs and improve outcomes before many healthcare organizations were even using that language. What did you see early that others did not?
Lonny: Again, I think the key goes back to outcomes. Big data started to make it possible to have access to information that could be analyzed, but the quality of the data was, and will always be, an issue. A lot of what we did in that timeframe was demonstrating that the data quality wasn’t sufficient.
Judy: Have we gotten there yet with data quality?
Lonny: No. But with focus, it can be done. We used to struggle with something called an ensemble model. So, if you wanted to take into account, instead of two variables, 10 or 15, but every time you added variables, the quality of the outcome was degraded. Now with the advent of large language models, instead of handfuls of data dimensions, you can get thousands, hundreds of thousands, even millions, tens of millions. The data doesn’t need to be as curated, but you still need something to compare it against to get ground truth. That ability to deal with lots of dimensions of data in an intelligent way is really important. When you combine that with reinforcement learning, you can get really high-quality results without having to be quite as rigorous on how the data is being curated.
Judy: At Intermountain, you were part of data and analytics work tied to significant improvements in outcomes and costs. What examples of that work come to mind?
Lonny: Intermountain had curated EMR data since the 1970s. We had a centralized electronic data warehouse that had normalized, curated, and cleaned all that data. So we had a huge advantage. One of the most significant examples was a reduction in sepsis mortality. At a time when sepsis mortality in most health systems was in the 30 to 40% range, our sepsis mortality rates were in the low 20s. But through a very deliberate approach to improve the process to find what worked well and what didn’t, and using what I would describe as a reinforcement learning approach, we drove our sepsis mortality rates down below 7% over a period of about six years.
Judy: What do healthcare executives most often misunderstand about AI governance?
Lonny: I think it’s kind of a mystery to most people right now. We should have been, and should be now, governing software the way we’re talking about governing AI. But the reality is that software isn’t governed as well as it should be. For instance, an EMR is not an ROI type of thing. It’s a cost of doing business, so it didn’t get the attention that it should have. AI is getting more rigor applied to it, and it’s forcing us to actually do things we should have been doing with software for years. I think governance is important to make sure that behaviors are aligned with the mission and with the culture of the organization.
I’m also a huge advocate of logging and tracking so that you can go back afterwards to confirm, “Yes, this was good,” “No, this was bad.” We shouldn’t suddenly be concerned about ethics just now. Were the decisions we were making before not in the best interest of the patient? If you are systematically denying valid claims, you’ve got an ethics problem. Whether you use a traditional mathematical algorithm or you use AI, there is no difference. So, what you do with AI needs to reflect your mission and be true to patients and be driven by helping patients get better care.
Judy: So, when AI governance is working well inside of health systems, what decisions are being made differently?
Lonny: Some organizations are still behind on setting up a governance structure. Here at Inova Healthcare, we have a really good AI ethics and oversight committee. Sharp Healthcare, where I was previously, does as well. You just need to have the right people in the room so that they’re aware of what’s going on and there’s agreement about things being used appropriately. There are several things that have worked quite well in healthcare with AI.
One of those is ambient scribes for clinical visits. The AI is listening to the conversation, it’s transcribing the conversation, and it’s summarizing and formatting the notes so that they’re ready to go into the medical record. But first, the doctor reviews the notes that the AI has created before she commits them to the medical record. At the end of the day, the doctor owns it.
Judy: Ambient listening has been one of the most viable early AI wins. Where else do you see realistic opportunities for AI to create meaningful clinical, operational, or financial impact?
Lonny: There are areas that I just call everyday AI. Using it for email communication, using it for meeting minutes and meeting summaries, creating documents, especially draft documents of policies, and training materials. AI is becoming a better answer engine. We’re accustomed to searching for things, and Google says it found 7,942 documents and put them in order of relevance.
Whereas AI says, “Here’s the answer to your question. And here’s the three documents, down to paragraph level, where the answer came from.” That scenario is very valuable in a lot of use cases, particularly in customer service. The thing that’s just starting to emerge is agentic AI, where instead of just answering questions and generating content, the AI is actually pushing buttons and executing things in your systems. So, it’s navigating Epic, it’s navigating Workday, it’s navigating Oracle. That’s just starting to be explored, but it has become capable faster than I expected. And again, it’s going to be really important to have human oversight.
One really good use case we had at Sharp was for our call center. When patients call in, voice calls use AI and basically navigate menu selections audibly. So instead of pressing 1 for this, pressing 2 for that, we had 67 different things that you could end up doing by answering questions from the system. It is a great experience. We had about 44% of calls (the average is 20,000 a month) lead to self-service closure without a live agent.
Judy: How should health systems decide when to build AI capabilities internally, when to rely on vendor-embedded tools?
Lonny: It’s just a new version of an old problem, build-versus-buy. Should a healthcare system be in the business of developing software? Maybe, if you’re big enough, you may be able to turn that into revenue streams as well. I think we’ll see a lot more homegrown solutions using AI because you can develop things dramatically faster than before. It just happens to be one of the things that AI is extremely good at, whether it’s vibe coding or lights-out coding. But still, you don’t do anything without people, and when you have good people with good tools, they are able to do things significantly faster.
Judy: With the financial pressures that healthcare organizations are under, do you think AI can genuinely help? Also, where do you think expectations may be too high?
Lonny: You have to test it. And if it works, great! Scale it. If you test it and it doesn’t work at the atomic level, it’s not going to work at the system level. Here is a really high-value example that comes up quite a bit. You do a patient survey or an employee survey, and one of the questions has a free-format text response. When you have tens of thousands of responses, someone has to go through all of them and see what people were saying. AI is exceptionally good at reviewing all the responses and coming up with the top five categories of responses and counting them. It would take a person four or five, six weeks to do that. AI will get an initial answer back in minutes. Anytime you’re dealing with free-format text responses to surveys, you should pull AI in. It will probably be helpful.
Judy: Earlier, you alluded to the rev cycle. Health systems are using AI to get more claims through. Insurance companies are using AI to deny more claims. How does that work? Are we getting into a sort of Dr. Seuss “better batter” or “better bread-butter battle?
Lonny: The battle of AI. Basically, you’ve got AI systems competing against each other, but it’s just speeding things up. Hopefully, you get the resolution on a denial of a claim faster by having AI working on both sides. It should create a more efficient process that’s more adherent to the rules of the road. People who systematically deny valid claims should go to jail for that and pay lots of penalties.
But if you’re trying to do things efficiently, it should just increase the throughput. We see that radiologists can be effective and have a significantly higher throughput. In the area of claims and rev cycle, it’s the same thing. It increases the speed and the quality. It can flag things where there’s insufficient data, then go collect that information. I’ve seen it cut the time needed to track down documentation in half. It’s not an “us versus them.” It’s, “Let’s get this done faster.”
Judy: Interesting. What advice would you give to healthcare executives who know they need to move faster on AI but are worried about the risk?
Lonny: At this point, I think there are enough proven cases that you can be safe just by following the early adopters. Everyone should be doing ambient scribes in clinical visits. Same with rev cycle, claims, and denials. Epic has been very good with their AI roadmap, so you can lean into what the big vendors are providing and see what’s working and what isn’t. So, I think there, it really comes down to seeing what others are doing and being a fast follower of things that are working.
Judy: People are concerned about losing their jobs to AI. Companies are doing massive layoffs and saying it’s because of AI, but there hasn’t been validation of that. How do you stem fears that jobs are going to go away or change substantially?
Lonny: If you’re not learning how to use AI to do your job more efficiently, you should be. If your job can benefit significantly from AI and you’re not learning how to use it, you should be concerned about the future of your job. The first wave of layoffs I saw was companies that were going to do them anyway and blaming it on AI. Others thought, “We probably don’t need as many people here, so we’re going to let some people go.” I’ve seen many of those companies hire those people back, and I think there will be more of that. AI is good at doing tasks, but it can’t really do the job. You need humans doing the job, with AI helping to do the tasks more efficiently.
Judy: Some people say AI is more hype than reality. IBM Watson was supposed to revolutionize healthcare, and it didn’t. Can you talk a little bit about that and why AI won’t go the same direction?
Lonny: Yeah. Watson played on “Jeopardy!” almost 14 years ago. And at that point, many of us thought, “Wow, this AI is real, it’s going to change healthcare!” One of the greatest accomplishments of my career was preventing healthcare systems from buying IBM Watson because it was just too expensive, and it wasn’t good enough. But generative AI is very cheap, and it is good enough, and it’s getting better at a rapid pace.
There’s a lot of talk that we’re in a bubble right now that resembles the dot-com bubble. But the reality is that there are real things happening, and it’s not a bubble, and it’s only going to get better. For instance, the combination of quantum computing and AI makes each other better. And as one improves, it improves the other one in a virtuous loop. So to me, this time, it is real.
My daughter and son-in-law have had an internet business for several years. They spent three years and $25,000 developing this app, and it never did what they wanted it to do. And long story short, they tried vibe coding the app themselves. In three weeks, including a learning curve, they developed an app that was better than the first one. That’s just a real anecdotal story, but a lot of people fall into that category with apps. I see some big systems looking at refactoring the entire code base to look for efficiencies and to get rid of antiquated and old technology underneath. So, this is not hype. There’s certainly more than enough hype around it, but it’s also extremely real and extremely cost-effective.
Judy: You have also done work focused on helping people live healthier, longer, and more fulfilling lives, especially around retirement. How has that shaped your view of where healthcare should be headed?
Lonny: For those who are not familiar with it, there’s a very rigorous scientific approach to better health called functional medicine, but it doesn’t get reimbursed. And here’s a problem with our current healthcare system. I use an AI through a company called Function Health, and, for three years, I have gone in for exhaustive lab tests. Previously, these would have cost thousands of dollars, but have come down to hundreds of dollars, and they’ve added AI.
So now I can ask, “What’s the trend of my A1C? What are the top two things that I should be focusing on to improve my health?” When the lab tests alone are not sufficient, I can feed in more information, like blood pressure and sleep data. Now the AI knows more about me than any doctor ever will. These AI-driven advances can help us optimize our health rather than just sit around waiting for something to break.
Judy: So, in your opinion, it sounds like the future of healthcare will be driven by AI.
Lonny: In the past, I was a real skeptic of AI, especially in the IBM Watson days. Today, I think that generative AI is going to be more impactful than personal computers, more impactful than the internet, and smartphones, and all those things combined. This is going to be a revolution.










