Episode 6: Visionary Voices with Sameer Sethi

Larry Kaiser:
Hello and welcome to this episode of Visionary Voices, Optimum Healthcare IT’s conversational thought leadership podcast that hosts digital and technology healthcare leaders to discuss IT innovation and transformation. I’m your host, Larry Kaiser, Chief Marketing Officer at Optimum Healthcare IT.

In this episode, I’m excited to welcome a seasoned leader and expert in the healthcare data and analytics space, Sameer Sethi, the Associate Vice President and Chief AI Officer at Hackensack Meridian Health. Today, we’re going to take a deep dive into Agentic AI in healthcare. Leading today’s conversation is Optimum Healthcare IT’s Chief Strategy Officer, Rick Shepardson.

Welcome to the show Sameer. And Rick, hope you guys are doing well today.

Rick:
Thank you, Larry.

Larry Kaiser:
Alright, so as we do with all of our podcasts, we’re gonna start off with our fun little segment called Bandwidth Banter, which we like to call fun icebreakers for IT professionals. So, have a little fun first and then we’ll really dive into the Agentic AI conversation here. So Sameer, I’m gonna start with you. As someone who is in technology and AI analytics, are you a Mac, or are you a PC guy?

Sameer Sethi:
I am a Mac guy. I’ve been for quite some time. As a matter of fact, I am PC-handicapped. It’s been so long, I struggle with using a PC. As you all may know, the functionality and the capabilities are quite different. Using a PC is a challenge. I really enjoy the integration between various devices and the capability that Mac does really well.

Larry Kaiser:
I’m the same way. The first time I got a Mac for work, I pressed the X button on the mail and it just minimizes it. But I went back to a PC for a period at work, and I pressed the X button. I spent the entire day trying to figure out why I wasn’t getting any emails. And I realized, well, duh, the X on the Windows machine closes it. The X on the Mac minimizes it. Typical work environment. Rick, how about yourself, Mac or PC?

Rick:
I’m a PC guy. We use Microsoft for all sorts of functions, and from a business perspective, I see the centralization, the platform effect there. In my personal world, I’m an Android guy. I like the flexibility. I was originally a Palm guy back in the day, right? And I love the configurability of that OS. I like that within the Android ecosystem, and I don’t know. Now I just use that and I try to build on Microsoft and everything they do from a business perspective. So I kind of separate out my two lives. I am a little jealous of people like Sameer, who have bought into the full Apple ecosystem and have your integrated experience around you. I think that’s a huge differentiator too, but I haven’t made the jump.

Larry Kaiser:
There’s nothing wrong with that, though. There are people in the world that will look down upon you for using an Android. I’m sure you’ve seen all the fun videos out there and so forth, and people asking you, oh, you’re the green bubble guy.

Rick:
I am not included on my wife’s family chat because of this.

Larry Kaiser:
It’s always a fun conversation when it comes up for sure. Alright, so let’s dip our toes into pop culture a little bit. Sameer, Office or Friends, which TV show do you go with there?

Sameer Sethi:
Office for sure. Friends is awesome, but Office for me relates a lot of things. I tell people that my life is relatable to what I’ve seen in Seinfeld, for example. From a work perspective, what I’ve enjoyed in Office is the workplace interactions, the humor. I was one of the first people that ran back into the office when things opened up. Back then, I was at Bon Secours Mercy Health. Two of us were the only people that showed up at the building. It brings back great memories. I think the way things were. So I’m a big rerun Office person.

Larry Kaiser:
Yeah, I can watch The Office reruns over and over again. Rick, I’m not sure if you’re much into pop culture when it comes to Office or Friends, but do you lean one way or the other?

Rick:
If I had to pick between the two, I’ll take The Office. But I’m not much of a sitcom guy, to be honest with you. I have a guilty pleasure show, though, that I will share. Maybe it’ll be lumped into my Android-ness, but I enjoy The Curse of Oak Island. There is something about just the—it’s definitely a guilty pleasure. They’ve probably got about an hour or two of original content per season, but they actually do have 11 seasons now. There are more episodes of The Curse of Oak Island than The Office, which is somewhat incredible. They are really using all means possible to rule out the fact that there’s probably no gold left on that island. I give them great resiliency scores. I give them great credit for working on innovation areas as well as they try to use whatever technologies they can to find or triangulate gold. But I don’t think there’s any gold left on the island. But it’s something that is mindless and enjoyable. And my day on Tuesdays is Curse of Oak Island.

Larry Kaiser:
There we go, Curse of Oak Island it is. Alright, well, I appreciate you guys playing along. This is one of my favorite segments to do, simply because I think it kind of loosens the room a little bit and gets you guys ready for the conversation on Agentic AI. So Rick, the mic is yours. I’m really looking forward to this conversation. So Rick, have at it.

Larry Kaiser:
Let’s share some good information with our listeners.

Sameer Sethi:
Thanks, Larry.

Rick:
Thanks, Larry, appreciate it. So, you know, Sameer, you are a leader in this space of AI. I’ve read no more than three or four articles about you that have come out, probably in the last couple of months, that you’ve been contributing to. There’s a lot that’s happening in AI today. Some of it’s AI, some of it’s automation. In some ways, we’ve been doing AI for decades now, in some ways, as we advance intelligence that are nonhuman.

But I think that each of these different disciplines of AI have overlaps, have considerations in how they work together, but also how they’re different. So maybe, as we kick off here, why don’t you give us your definition of Agentic AI and how you see it within the broader AI and automation ecosystem.

Sameer Sethi:
Yeah. So, you know, when we talk about AI or as we have talked about AI as it evolved, it’s always been about, for me at least, about actionable insights. Now it could be because, as a result of artificial intelligence, and sometimes it’s not artificial intelligence. It’s just insights that derive from data that was collected. And what happened versus what’s going to happen, right?

So I think, you know, to me, the reason I mention that is, for me, Agentic AI is moving from actionable insight to actioning insights. And there’s a pretty solid and distinctive distinction there, because what Agentic AI is doing is unlike in the past when we were providing these AI-driven insights to people, right? What we’re doing is now—Agentic is a combination of providing those insights to a non-human at times, you know, and obviously rightfully including human in the loop when needed, especially in healthcare. But it’s about giving that insight or handing that insight to a non-human technology that does then take action.

So it’s—and that’s where I change it from AI being an actionable insight if done so correctly to Agentic being an actioning insight, which is—it’s actually doing something with that insight. It could be by virtue of using something that’s digital. It could be something that’s by virtue of using robotics process automation, right? It could be by virtue of, you know, generative AI summarizing something as a result of that. But it’s really the actioning piece and doing something with those insights. And that’s done by a non-human technology.

Rick:
I like what you’ve done there, going from artificial intelligence to actionable insights to actioning insights. You’re still sticking in the AI world. You’re creating your own labels. I love it.

So I think the actioning aspect—there’s, I think there’s some levels of trust, levels of confidence that is required before you can allow these actions to occur without a human in the loop, right? Maybe if you have a human in the loop in this world, maybe you can tee up or set up a queue of validation on the actions that are going to be performed. Maybe it kind of creates the human as an agent in this Agentic ecosystem at that point.

Do you think about it that way? And I guess how many—from what you’re implementing or working on today—to what extent do you have any of these agents, nonhuman agents actually actioning versus having the human as that, you know, last checkpoint?

Sameer Sethi:
Yeah. So, you know, this conversation—it’s a great question, and it comes up quite often. We as an organization think about it all the time. You know, I’ll start off by saying my comparison to that need is very similar to what happened when data and analytics picked up a lot and started to become prominent in the decision-making space. And I still remember having the conversations of how do you make sure that the data that’s ending up at the end of the—you know, to a user, right? Those insights are correct.

And the way we solved that was that we said obviously there has to be human in the loop doing some verification. But you have to build QA checks into the ETL, right? And those QA checks would start to confirm that the data that’s coming through, or the insights that’s coming through to a dashboard, you know, for example, is accurate or not, right? So we would have limit checks. We would have fill rates, you know. So we would profile that data and then QC, or quality check that data. And that didn’t solve for everything, but it at least removed concerns. That was 60, 70, 80% of the concerns. So if something from the left wasn’t transmitting well to the right, the QC checks in the middle would check it.

I think the agent space, or the agent concept, is in a similar space where I think humans are required, you know, to yes, validate a few things, but also do those QC checks as well. And that’s what we are incorporating in at least what we are building today. These are early days for us, right? But what we are doing is using humans to do some validation. And through those learnings, we will build that automation as well—as far as the QA automation—into these actions that these bots will do or these Agentic AI tools will do. And I think over time that will reduce, right? As we build confidence with what they are doing. I think the human in the loop will meaningfully be there, but it will be there lesser and lesser.

I think that’s one portion of this. Other is, I’d also want to just point out that, you know, the things that all of us need to work on today for Agentic AI are low-risk use cases, right? This is still a very early new technology for all of us, right? So we are focused on building Agentic workflows that wouldn’t—would have low risk or doesn’t cause a whole lot of harm, right? And that’s important. Because we learn from this quite a bit, right? We’ll see what’s tolerable and what’s not tolerable, what this technology should do, can do, right? And I think over time we’ll require—I mean, the intention is to require—when you build the Agentic workflow out, that we have less human interaction.

Rick:
Yeah, I think that’s the old adage of, to your point, sort of the ML-driven pipelines that existed, and the human has to minimally validate—yes, no. And over time, you get enough quality stewardship, validation, that you can automate further, right? That’s really the similar path that sounds like you’re on.

As we now look at the Agentic nature of taking those actions—the next step, I think. I want to hit on what you said, and maybe twofold here. We’ll hit them both, but we’ll start back. So when you’re looking at some of these, we’ll say, lower-risk opportunities, I know there’s a mix of things that are available from maybe an operational or a financial perspective. You know, clinical starts to get a little bit more challenging or risky in some ways. But are there, we’ll say, governance frameworks or evaluation methods that you use to quantify risk or to identify areas that are more feasible, or enabled for your organization?

Sameer Sethi:
Yeah. So we at Hackensack started this AI journey almost three years ago now. Governance was a big aspect of that journey. Our journey around governance has evolved. I tell people that what we do from a governance perspective today is nothing like what we did six months ago, and it’s going to be very different from what it is today to what we’ll see six months from now.

Our first stab at governance was looking at and doing a technical evaluation. It then evolved to inviting almost 13 domains to evaluate an AI capability, which includes agents as well. So today, we have 13-plus domains that ask questions of the AI capability, which includes technical and non-technical. So it’s no longer just about the technical team evaluating AI and looking for how things like drift would be caught in a model. It’s also about compliance looking at it and saying, “The output, the use of this capability—does that make sense for the organization, and does it adhere to certain standards?” HR is looking at it and saying, “What is the impact to workforce, and how do we make sure that is managed well?”

Moving into the healthcare side, we’re looking at if the output of it is clinical documentation. We are trying to figure out if the output of AI should be documented as a clinical record, or should it—right? So it’s very comprehensive. Today that evaluation is almost 118 questions—call it that—that is asked of the AI capability. The output of that evaluation is actually, put simply, a risk score.

The way we do this is we break this up into 0 to 3. 0 to 1 is low risk, 1 to 2 is medium risk, and 2 to 3 is high risk. High-risk use cases—actually medium to high risk, which is the 1.5 to 3—not just an acknowledgement, but actually a signature. So what we’re doing is we’re doing these evaluations—technical and non-technical evaluations—pushing this over to the business, with the leadership involved in them, and saying, “Okay, this is the use case. This has an ROI, and this is the risk of doing this.” And you, now, business user and your leadership, have to take a decision to inherit the risk of using this capability or not. And that starts to create that level of transparency around what this is and what this isn’t, what it can do and what it cannot do, and what potential harm it may cause—or none of that.

I think that’s helped us quite a bit. Our governance process is super tight. Today, we have over 37 AI applications in production. Some of them are built by us internally, some of them we buy, but they’re all treated equally. We have them go through the whole 118-point check, provide a risk score as part of the governance process. There’s that ROI conversation—should we do it, should we not do it? Is there a soft advantage to it versus a hard cash value advantage to it? All those conversations happen. We bring all that together to the Governance Committee, which is at two levels, and that helps them take a decision of whether we do this or we don’t do this.

Rick:
So you’re taking the business of healthcare to another layer, right? A lot of times when organizations or technology companies are selling into the business or they’re trying to bring a solution to the business, it’s focused on revenue cycle denial management and that department, or it’s focused on maybe schedule optimization for patient access. But what your governance structure sounds like it’s doing is involving all of the—you said the domains—all of the different departments or operational units in your organization to evaluate, almost like a PESTLE analysis for your organization, to understand what the broader impacts are going to be, not just for any given business unit or department. It’s what’s the organizational impact.

So do you find you get a lot more change management? Do you get more change management, adoption, alignment within your organization through that model? And you said transparency, but do you also drive more collaboration or shared understanding across your org at the same time?

Sameer Sethi:
Yeah, we do. I think collaboration is one, because while there may be a direct impact to a certain domain when it’s in the application, there is an indirect impact to other domains as well. So I think creating that transparency and the collaboration process as a part of this helps us quite a bit. It also helps us from a perspective of—if we have something similar, there’s no reason to do this again. So we tend to catch things that we would otherwise not catch as a result of that collaboration. So it’s been very helpful.

And by the way, I think you know this—initially, when we thought about and designed this, the biggest concern that came from these leaders was, are we slowing down the process of innovation? So what was important was for us to standardize. And that’s what I meant when I said that all these AI applications are treated equally. Initially, when we reached out to do this, we thought, let’s not have a non-formal, low standard. We were evaluating these things, and that wasn’t scaling. So what we did is we standardized the process of scoring and risk scoring—evaluating that capability. And that helped us really do this fast.

So today, our governance process generally takes three weeks. In three weeks, we are generally at a place—unless some take longer if a deeper analysis is required—but generally, we say within three weeks it should have gone through vetting: the safety, the security, all these 13 domains, the ROI conversation, the finance looking at it. All that has to happen within that time. And that all came because we standardized the process. We had the organization agree on a standard way of evaluating these things.

Rick:
Great. Are there any of the domains that are maybe more challenging to get through? Or the questions end up having more variation?

Sameer Sethi:
Yes. There are certain domains that are still, I think, in a bit of a—I’ll call it a limbo—around what is the right way of looking at these things. As you may know, there’s lack of regulation—I’ll call it that—around AI today. So we all are learning from each other. There are certain folks, certain domains, that are not clear about what is required to evaluate an AI capability. Technically, we’re very sound. We know what it is that needs to be evaluated. But there are lots of conversations.

One example that I’ll use is data use beyond purpose of service. Today, organizations who we are buying these capabilities from are looking for data to train their models. There is a concept of de-identifying the data and then using it for whatever. But the concerns that we have are—is the concept of de-identification changed with the level of compute power today? That’s available in the market—does Safe Harbor hold anymore when it comes down to de-identification?

So how comfortable are we with telling organizations that if you de-identify data appropriately and call it Safe Harbor here, that it’s good and you can use it? Because if the data gets re-identified, there’s a responsibility that comes back to us. A patient can come in and say, “We didn’t agree to allow you to use my data for commercial training purposes.” That’s an area—that’s just one—that is still under development. And we’re a bit nervous about it. Data use beyond purpose of service, or data use in general, is an evolving area for us.

Rick:
Yeah, that’s a great example. Outside of the healthcare domain, there’s been a lot of, we’ll say, data use—agreements—or not—that have been in place. And you see the artifacts of the data in the responses that we get from a ChatGPT in the generative AI space, for instance. It came out of Reddit, or it came out of the Wall Street Journal. And, you know, was it authorized or not?

And it’s one thing to do that in a commercial sense, in a way—for the consumer, for any individual consumer. But we have a lot of protections through HIPAA or Safe Harbor, or other areas that you identified there, and the sensitivity becomes so much more important with our personal health data. And I think that—if we take that sensitivity, that personalization, a step further—what do you see in terms of providers, physicians, and/or patients, in terms of authorization of use or trust in some of this data or enablement of you—let’s not even talk about any of your vendors—but I guess enablement of you to refine your models or your production? Do they recognize the benefit of an individual case of information feeding back into the machine and creating benefit for the whole?

Sameer Sethi:
Yeah, look, I— It’s a lot to explain to a patient, right? And I don’t think the people that are talking to these patients are ready to have those conversations, right? And I’ll tell you why. You know, let’s use ambient AI as an example here. Right? So today there are various vendors providing this capability where, when you go in, they turn this capability on, and that generates a clinical note, right? That listens to the conversation. The patients have questions around this, right? But who’s in front of them to answer that question? It’s the physician, right? It’s the nurse.

These people aren’t trained to explain what that is and why that is. So it’s definitely an opportunity for us. When we have done this, you know, I think—and as long as we manage that conversation well, I think patients are coming out of this and saying, “That’s cool. That works.”

Right? “This is good for me. This is good for you.” I mean, I know—I’ll give an example—when we are explaining AI capabilities to clinicians and therefore training them to explain this to physicians, to patients, you know, we’re telling them that this helps them focus on the patient.

Right? Because now they’re not spending time typing a clinical note or reading a clinical note, right? And therefore, they’re looking into the patient’s eyes and having a one-to-one, in-person conversation, sharing the screen. And AI is helping with that. Those sorts of conversations are what convinces them, right, to say, “AI is good. You know, this is really helping me. This is helping the clinician.”

This is improving outcomes. But again, I think it’s early days. Like, we’ll have to—we’ll have to teach our patients the value of this. We’ll have to teach our clinicians to explain the value of this to the patients.

Rick:
Yeah, you know, my PCP has been using ambient AI for about a year or two now, and I’ve had some, you know, knee issues and the like. And I—honestly, I love it. I’m very pleased that he’s using that. And he does, to your point, spend more time with me, right? And so I’m all for it personally, but I’m probably not—I’m probably an outlier, I guess. Yeah, go ahead.

Sameer Sethi:
Yeah, I was gonna say, you know, it’s amazing. Just wanna maybe, you know— Just moving to a different direction here, but you know, it’s amazing how the whole ambient space has focused on generating clinical notes, which is, I think, very important.

What Hackensack has done is—we’re actually looking at the whole process of, you know, what does a patient experience look like?

And what we have realized is, while generating a clinical note using AI is really important, what people haven’t focused on is the consumption of that information. So what we are saying is, we’ll work on generating using AI to generate, but we’ll use AI to even consume that information. So we actually help the clinician, you know, using various technologies to summarize what is in a patient chart.

And that’s been overlooked. You know, since you brought it up, Rick, those are the opportunities that are overlooked. People are too focused on generating this information when, in fact, there is so much data out there—about a patient. You know, a physician needs time to read that, right? And they don’t have that time. So we are using AI at Hackensack to actually summarize the information meaningfully by specialty, right? So a physician can walk into an appointment with a patient and be ready—right, versus read that clinical note during that conversation.

Rick:
Yeah, I love that. And then I think there’s opportunities to take that downstream too. Obviously, as the note is produced, why aren’t you then using it to maybe recommend the diagnoses, referrals, the charges? How do you get further downstream to any optimization of clinical documentation improvement or coding? And you know, how—when we look at that, now that’s happening—you have a new primary source of information—as the ambient and extraction. If you have a quality of it, and then you end up having so much more opportunity for agentic workflows, then, to take that data and kind of move that along, right? So I guess— Let’s kind of bring it back to agentic AI here for a minute and tell me about maybe one of your favorite use cases that you’re actually implementing right now for agentic AI. And to what extent is it—we’ll say agent-to-agent on a technology side versus agent-to-human, agent-in-the-loop, and they’re taking action. Can you—wanna maybe share a little bit about one of your favorite workflows?

Sameer Sethi:
Yeah, I’ll talk about what’s on top of my mind, which is under development. We are looking to see if we can use agents for claims denials, and that is not to help—not deny a claim in general, but it’s about when a denial actually comes to us, how do we process it?

So if you think about, you know, when we find out from the insurance company that a claim is denied, you know, they provide us with reasons why.

Generally, a human looks at that information and then starts to create what’s called an appeals letter, right? So this is the appeals process. This isn’t the denial process per se, but it starts with that denial. But we have to create an appeal that goes back to the payers or insurance companies.

That process requires a lot of steps by a human, right? And what we’ve seen is that some of those steps can be actually done by agents. So what we do here in this case is, we have AI read and consume that letter—make sense of it, and then start to establish what’s missing. So is it a pre-authorization that we didn’t add as part of the claim submission? Is it a modifier that’s actually missing—right? And—if that’s the case, we either add the modifier meaningfully, and it uses a rule-based engine to actually do that, or it goes and looks for the pre-authorization, if it exists, and attaches it. It then generates the appeals letter—by using generative AI—right? It then sends it to a human, who then reads it and says, “Yep, this looks right.” And then everything is done—right. So—

And today a human does this whole process. And what I just described to you was a mixture of us using generative AI to read the letter—or should I say, document understanding to read the letter. So I know people use document understanding to read and consume the letter, then use a rule-based engine to say that “if-then” sort of a concept there. If this is missing, add this. If that’s missing, add that. Right?

Then use generative AI to actually generate a letter—and then send it to a human using some RPA capabilities. And then eventually it shuts out. So what we have done here is the agent—the agentic use case here—is an orchestration of various different technologies that are coming together.

And now we have an agent that is actually doing this appeal generation process.

Rick:
That’s fantastic. I mean, traditionally, you’re talking about four or five different roles—clinicians and others that are involved in that appeals process and the writing. And that’s a fantastic example. Thank you for sharing that. And best of luck in getting that implemented in production here upcoming. You’re going to make a lot of your insurance vendor—insurance payers—plans very, very pleased with your response times.

Sameer Sethi:
Yeah. And it helps with, you know, us, you know—getting the appeals letter out, which sometimes takes days—doing this in minutes or seconds, right? I think it helps with the cash flow—helps with a lot of different things as well, right? And such use cases exist today, by the way, across the healthcare continuum, right? It’s not a Hackensack thing. It’s across all industries. But I think that’s what we’re looking for—we’re looking for where machines can actually deliver impact, right? And then hopefully involve humans where—and absolutely where—necessary.

Rick:
Yeah, I was a bit tongue-in-cheek there on the payer side. I’m sure they’re not going to be very pleased with your response times. But I think that your finance group is going to be very happy with the impacts. And I think patients—getting your advocating on their behalf, and getting paid so that those appeals don’t end up in a patient balance at some point in the future, too, right? There’s a lot of opportunity there to have a real impact.

Sameer Sethi:
You know, another use case that I want to mention here as well, which is—you know, Waterline agent is also—you know, there’s a big deficiency across all health systems to check in on patients post-discharge. Right? So we’re very keen on—we actually see an opportunity for a non-human to help with that. It’s a very big—very big patient population that gets—and I think the opportunity to call them—build a chatbot that can have a meaningful conversation, create actions—I mean, you know, which could be, “Should I call you back and remind you at 5 o’clock every day that you need to take your medicine?” We’re starting to build capabilities to offer that service. Those things don’t exist today. Right? Documenting that information and putting it in our EMR, right? Which a human was doing and typing it. So that’s freeing up capacity for our folks to take care of the, you know, the harder work, right? And also, it’s actually doing the work that we were not able to get to.

Right? So I think doing post-discharge calls using chatbots plus RPA, plus a few other capabilities to generate capability to document everything, you know, and have that interaction in some form or—during that conversation, identify need to involve the human in that conversation. Right? I think all that is very heavy and high impact for us.

Rick:
That’s fantastic that you get into some, you know, maybe early disease identification, disease prevention. You’re going to reduce readmissions there. I think that whole area has been criminally underfunded. You know, and, you know, transitions of care management tried to get there, chronic care management tried to get there, and it’s not going to get there from just a human workforce. You’re going to have issues there. So I think doing things with an agent, or automatically—or, you know, so that you can have the right loop with the patient and refine it for individual preferences as well. Personalized medicine comes into play here. Like, there are so many opportunities. And I think we’re just starting to scratch the surface.

Sameer Sethi:
Yeah. And, you know, it’s amazing how far that tech has gone, right? Especially now with generative AI into the mix, right? So, you know, when I had done this for the first time, everything was intent-based. And the issue with that experience—the patient experience, the user experience—was that as soon as you went outside the intent, it couldn’t manage that conversation. But now we are plugging in large language models. So at least our approach at Hackensack is not to use a large language model to have the conversation, but have a combination of intent plus large language model capability to have that conversation, right? So it is very intentional in asking a question and expecting an answer, but as soon as something deviates from that, a large language model capability comes in and helps manage that conversation.

In addition to that, on the side, we are running—we’re running a sentiment analysis. So if you start to see or sense some tension in that conversation, our capability starts to say, “Do you want to talk to a human instead?” Right? So the point I was trying to make is—think about just a couple of years ago, where it was only intent-based. It had no sentiment analysis built into it. It had no large language model capability built into it.

Now, all of a sudden, this technology is offering a much better experience and much better conversation. And that’s why we, as an organization, are putting our bets into this. And we’re saying, this tech is ready now to have those conversations. And the people who are receiving these calls would actually appreciate it. Now granted, there will be folks that will not be happy with it. They’d rather speak to a human—and we offer that—so that sentiment analysis can be detected and offer them that very quickly. Right? So I think it’s really amazing. You know, another feature that we have incorporated, by the way, in that technology is language detection. Right? So what we realize is that people start with a certain language. So, you know, you go in and say, “For English press one, for Spanish press two,” right? We don’t do that anymore. Instead, what we do is that if you, in our experience, start with English and then switch over to a different language, it will detect that language and start to have that same exact conversation without a break—in that language.

Right? So those are things that this technology is now capable of doing, which it didn’t have before.

Rick:
No, that’s a great—that’s a great example. That’s really interesting. And it seems so logical in some ways, but also something that is so challenging to do in our older, older technology world. So thank you for sharing those great examples. You know, I’m gonna get you out of here. I know we’re—we’re almost at time. But—you obviously have been very intentional about driving the change that you’re—seen or see opportunity for, and you probably have had to conduct your own sentiment analysis on the various stakeholders that you work with, so that you can help drive that change. I guess—maybe—what do you recommend, or how can others who want to follow in your footsteps or implement agentic AI—how can they drive change? How can they taste test the sentiment of their organization and drive an impact for their patients?

Sameer Sethi:
So I think a few things come to mind. One is—you have to be relentless, right? It’s hard to explain this technology and what it can or what it can’t do, right? You have to have the right partners around you, right? Who can help you—who have this technology or can help you build it, right? And then I think the most important is buy-in—right? You know, we—you know, I am fortunate enough to be sponsored—you know, by our CEO directly, which has been very impactful, right? You know—Bob Garrett, our CEO, has been behind—technology enablement and AI enablement since I’ve been here at least, right? And he opened HIMSS two years ago as a keynote speaker and talked about AI and automation. This year—this past year at HIMSS—he did that again, you know, for the executive group.

And that’s a reflection of how the organization thinks about automation and AI, and now agentic. And I think that has helped us quite a bit, right? Because it starts from the top—I mean, it starts from the top and then moves into the bottom. And then, I think internally, we’re also doing this as well, where we are creating various education programs for people to understand what this technology is. So going back to my first point—it’s explaining it, right? So I think, in summary, this is my advice to folks—that have a—you know, have a top-down and bottom-up approach to this, right? Hit it from all angles. Be relentless about—explaining what this is. And, you know, it’s going to be hard.

But I think you need to just keep at it. Keep the partners right next to you, because, you know, I realize almost every day I have to almost check myself to say that I work for a hospital system and not a product company. Right? So I have to involve the partners—right—to actually help with this technology. So, you know, I think it’s going to take a village—right? Between us, our leaders, and technology partners—to come bring all this together and deliver impact.

Rick:
Wonderful. Thank you for—and I’ll just say thank you for your partnership. You’ve been a fantastic partner, and I’m sure you’re a fantastic partner for your organizations as well. So thank you.

Sameer Sethi:
Thank you, Rick. We appreciate this.

Larry Kaiser:
Yeah, guys, this was a really interesting, educational conversation. It’s great to hear how Sameer, as a leader of an organization, is really taking the bull by the horns and really being innovative—and really being a leader in healthcare with all these new technologies that are coming out around AI—when historically, you know, healthcare organizations, or healthcare as a whole, have been slow at adopting technology. It’s great to hear all this really cool stuff you’re doing around that. So—so yeah, really appreciate you joining our podcast here on Visionary Voices. Any last thoughts, guys, before we sign off?

Rick:
When can I get more time to talk to Sameer? I mean, sign me up anytime.

Larry Kaiser:
Fantastic.

Sameer Sethi:
Yeah, this has been a great conversation, folks. Thank you for including me. And Rick, anytime—this is what I do, and this is what I love.

Larry Kaiser:
Awesome. Well, that concludes our discussion today on Visionary Voices. I want to again thank my guests—Sameer Sethi from Hackensack and Rick Shepardson—for joining us today and really having a great conversation around agentic AI and AI as a whole. And to our listeners, thanks for tuning into this episode of Visionary Voices powered by Optimum Healthcare IT. And remember that the future of healthcare IT starts with your vision. We’ll catch you next time on Visionary Voices.