Editor’s Note: This is the sixth in our series on Chief AI Officers in Healthcare. Other recent CAIO profiles include Dennis Chornenky at UC Davis Health, Dr. Karandeep Singh at UC San Diego Health, Alda Mizaku at Children’s National Hospital, Dr. Zafar Chaudry at Seattle Children’s and Mouneer Odeh at Cedars-Sinai.
Sameer Sethi, senior vice president and chief AI and insights officer at Edison, New Jersey-based Hackensack Meridian Health, is a seasoned leader and expert in healthcare data and analytics with a track record of enabling use of data and analytical techniques to drive digital transformation.
Before ascending to that new role – one that’s growing in popularity at health systems nationwide – Sethi has had long experience at the intersection of healthcare and technology, improving quality, increasing access and lowering the cost of care delivery. Previously he worked at Mount Sinai Health System, McKinsey and Bon Secours Mercy Health.
Sethi and his team currently are focused on accelerating the use of artificial intelligence and machine learning to deliver high-quality, affordable, more accessible and more efficient healthcare at Hackensack Meridian Health. Healthcare IT News spoke with him recently to discuss what it takes to be a Chief AI Officer and to talk about the efforts he and his team are undertaking right now.
Q. How did Hackensack Meridian Health approach you to become its Chief AI Officer, and what were they looking for? Who would you report to?
A. We are three years into our AI journey. When I came in in 2022, the intent was to build a data and analytics foundational layer that would start to cater to the needs of the network in various different ways – robotics, process automation, AI.
Back then, it was machine learning AI and less generative AI, or almost no generative AI. Agentic AI is fairly new. As organizations started to see the very quick maturity of data and insights enablement, that is where they then said we need to have a focus on AI – and someone needs to focus as an expert on AI.
A big motivator was our CEO, Bob Garrett, speaking of AI very frequently, more so than I think any other health system CEO. Last year he was the opening keynote speaker at HIMSS24, talking about AI.
This year he opened the executive forum for HIMSS25. He’s a frequent speaker on AI. It was our board’s intention as they approached me. The board wanted me to take AI more prime time than we have in the past. The board felt we have invested appropriately in the foundational layer, and now we are ready to start building AI, deploying AI. That’s how they approached me.
I agreed to do it because I saw a top-to-bottom approach to this, which was very comforting around not just talk – they would actually do things. Then I saw a bottom-up as well, which is what I was building, which is a capability to deliver AI. There’s no AI without data. I saw the ingredients were all there, and that got me really excited.
I helped define this role, as well, because it was new. They were looking for somebody who actually would be pushing stuff into production. AI in general, unfortunately, for a lot of health systems, has been an academic exercise. They build models, they focus on accuracy, which is very important, but very little actually ends up in production.
I saw an article last week that said more than 90% of models don’t make it into production. We’re completely the other way. We don’t build a model if it’s not going to make it into production. While we’re building the AI models, we’re also focusing on adoption and human factor engineering and how people are going to use it.
There’s a lot of thought that goes into this. They were looking for somebody who can bring all that together, which isn’t just a modern development piece. Somebody who understands healthcare, has worked in healthcare, knows what success looks like, is very in touch with the business and the business problem versus being a techie. It’s about somebody who truly sits at the intersection of technology and business.
As far as reporting is concerned, I have a dotted line into the CEO, and then I have a direct line into the CDIO. This reporting and this role came into formal existence in November of last year. But it’s not new. This is two to three years in the making.
People don’t realize this, and it’s sometimes overlooked – AI is a technology, it’s a software product. There’s good software and there’s bad software. I think it is important for somebody in this role – to be the right fit – to know what’s good and bad, what’s scalable, whether it’s the right fit – a sense of what ROI should look like.
This isn’t about the shiny toy. If you go after that, you will get some initial wins, and then eventually the person in this role will crash and burn.
Q. What in your background makes you a good fit to be a Chief AI Officer? And what skills should anyone looking to become a Chief AI Officer have?
A. I am not the IT person who has lived the life of maintaining infrastructure. Instead, I’m an insights guy. I work with business, the hospital business all my life, to solve business problems, not to give them software. I’ve built intelligence that gets leaders to get to the root cause of why certain things are happening. That keeps me at the intersection of IT and business.
It’s somebody who gets and appreciates the problem and is willing to get their hands dirty to understand the problem, experience the problem. Because what happens to the folks who are too tech-heavy is they’ll provide a system that doesn’t work for the business. If you’re too inclined on the business side, then you don’t know the limitations of the technology or can’t appreciate that.
Business units reach out to me. I can’t count anymore how many times the conversation starts with, “I want AI.” My pushback to them is, “No, let’s not talk about you getting AI. Let’s talk about a problem that you’re looking to solve. Then I will help you answer whether it’s AI or whether it’s some non-AI software.” It could be they’re not using a current capability; or if it’s change management, that has to be brought in to use something that already exists.
Q. Please describe the AI part of your job at Hackensack Meridian Health. Just in broad terms, what is expected of you? And then in more specific terms, what is a typical day for you like?
A. Sometimes it’s just a simple dashboard versus AI. I think the AI part of my job, or I should say a good portion of my job, is to help the organization look at a problem and see whether AI actually is needed or not. That’s the realization I bring to people. I explain to them that here’s a problem you have, and yes, AI can help with that, or AI is overkill. Because AI costs money.
The translation of the problem to a system and where AI fits or doesn’t fit is a good portion of my job.
A typical day for me is scanning the market for the problems I am seeing the business needs to solve for. That’s a big portion of my job, which includes a build-versus-buy angle. Not everything has to be purchased, and not everything should be purchased; not everything should be built, either.
I have a software development team that reports to me that writes software, not just the AI software, general software, too. The conversation always is, should we build or buy? Build gets you exactly what you want, but it is a lift. Buy gets you something a lot sooner, but it might not be exactly what you want.
So, what’s available on the market, what can we build, how fast does the organization need it, does it match the needs? Creating a process by which we are mulling over the requirements and then looking at what’s on the market versus what we can build is a big part of my job.
I round constantly, by the way. I am consistently rounding on business and asking, What is it you want? I need to keep myself, my head, into where their head is and look at the problems they have. I sometimes say this is an operational issue and can be solved and technology is not going to do a whole lot, or, let’s go all in and build this or buy this.
Q. Please talk at a high level about where and how Hackensack Meridian is using artificial intelligence right now.
A. So we have six buckets that things must fit in. They aren’t domains, but areas where we need to solve for from a technology-enabled perspective and AI-enabled perspective.
The first one is creating personalized and equitable experiences. This is both for patients and our workforce. The second is streamlining administrative and clinical efficiencies. There’s a lot of generative AI and machine learning; there are a lot of opportunities around streamlining certain processes by which we can bring efficiency to our workforce.
The third bucket is burnout alleviation. This is about looking at mundane tasks and not just bringing efficiencies but also figuring out where technology can help people do their jobs in a way they don’t burn out. A good example is writing clinical notes or consumption of clinical notes. What we’re trying to solve for is physicians not writing clinical notes.
The fourth bucket, which is the highest one, is disease prevention. This really moves the needle, us detecting disease earlier then intervening a lot quicker, bringing the right kinds of therapies so there are better outcomes. That’s been our big use case since we started this program three years ago.
The fifth is precision treatment. We haven’t done a whole lot here, but we’ve done pieces of it, which is bringing in the concept of precision medicine, but not putting it into precision treatment yet.
The sixth bucket is research and innovation. We have a whole team that is called CDI, Center of Disease and Innovation. We are creating tools for researchers to do research faster and better. Those are the six areas.
Q. More specifically, please describe a couple of AI projects you are proud of and that are working well for your organization – and some outcomes you’re seeing.
A. One is most near and dear to us and should be most near and dear to all health systems. It’s in our bucket of disease detection and disease prevention. We had a capability. This was almost two and a half years ago. This was one of the first AI use cases we put into production, where we came up with a risk of mortality.
The reason we did that is because we were seeing that just by virtue of process and the way clinicians work, which is fixing the patient, we weren’t moving patients into things like palliative care or hospice when it’s appropriate. We created a score which would score a patient on their chances of mortality within the next six months to one year.
As a result of this, it’s a nudge to a clinician with the right attributes, and we use almost 100 attributes to come to that score. And what that allows a clinician to do is say, Okay, maybe now it is time for us to start thinking about end-of-life care. That’s been pretty moving for us as a health system.
And while that’s a small example, the reason that’s important is it has motivated a lot of other disease detection and prediction for us. We have moved into CKD detection, for example, chronic kidney disease. We are now working on chronic asthma, and we continue developing.
But I think the reason I’m so proud of it is because the team was able to convince the organization to say, The earlier we detect, good things start to happen.
For example, I have a family member, I lost my mother-in-law a few years ago, and she could have benefited from this end-of-life care. But unfortunately, by the time that showed up, they told us we should have been here six months ago. And for all this time, she was going through therapies, and maybe that could have been avoided.
So, we actually show it. Somebody that was in and out of the hospital five times without this capability, and then it was death in a hospital, inpatient. It isn’t preferred by the family and most patients. If you plugged in this capability, it was one hospital visit that actually created the ability to detect mortality. And then induction of end-of-life care, and then instead of a person dying in a month, the person would die in five months.
But the person would be at home and with their families. That’s what motivated us to use AI for disease protection. This applies in some lighter areas. If you think about CKD, chronic kidney disease, this late-stage CKD is detected, unfortunately, a lot later than it’s supposed to or would be preferred. Now, if you start to detect using AI models, you then start to intervene and say, These are the therapies that are required, and let’s start that.
What that starts to do is readmission drops, people have a better quality of life, cost of treating a patient for CKD drops as well, because now we are managing that condition a lot better. So that has triggered a large amount of use cases and ideas around how we can early detect the value of that. So that’s a disease prediction.
Now let’s talk about the operational improvement side. We have almost 180 robotic process automations in production today. What that’s doing is it’s saving hours out of people’s day so they can focus on the top of their licenses.
Could be physicians, could be people at the desk doing clinical work, could be finance or HR. What we’re doing is we’re using RPA plus AI to read documents that come in the email, make sense of those documents, and then trigger actions that a person would do. It could be as simple as receiving an order in email and then going into our CRM system typing those things in.
That was done by a person. People get sick, people leave, you have to train people. But once you build automation for this, then it’s a repeatable process and it just runs on its own. That’s been pretty important for us, and creating an awareness in the organization that a lot of things we do today, mostly simpler things that don’t require a lot of thought, but not a lot of decision making, can be automated through AI and RPA.
For a 5-minute video of bonus content not found in this story, click here. In the video, Sameer Sethi shares tips for IT executives looking to become a Chief AI Officer for a hospital or health system and offers his views on the best ways a Chief AI Officer can work together with his or her peers in the healthcare C-suite to ensure AI gets done right.
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.
WATCH NOW: How can an IT exec become a Chief AI Officer – and then work with the C-suite