Reducing Audit Friction and Driving Accountability with AI in Healthcare – with Yasemin Agatan of AdaptHealth

Welcome everyone to the AI in Health Care and Life Sciences Podcast. I'm Matthew D'Mello, Editorial Director here at Emerge AI Research. Today's guest is Yasmin Agatán, Senior Vice President of Internal Audit at a DAPT Health. Yasmin joins us on today's show to discuss how audit and compliance leaders are responding to a health care landscape or digital systems regulatory expectations and operational demands are rapidly evolving. Together we explore how internal audit teams must move faster, collaborate more effectively with other departments and strive for greater transparency, all while sustaining the rigor required in a highly regulated sector. Yasmin highlights practical shifts, organizations are striving toward including improving data visibility, adopting more structured and explainable processes and strengthening issue tracking and remediation efforts. Just a quick note for our audience that the views expressed by Yasmin on today's program do not reflect that of a DAPT Health or its leadership, but first, are you driving AI transformation at your organization or maybe your guiding critical decisions on AI investments, strategy or deployment. If so, the AI and business podcast wants to hear from you. Each year, Emerge AI Research features hundreds of executive thought leaders. Everyone from the CIO of Goldman Sachs to the head of AI at Raytheon and AI pioneers like Yoshua Benjiro. With nearly a million annual listeners, AI and business is the go-to destination for enterprise leaders navigating real world AI adoption. You don't need to be an engineer or a technical expert to be on the show. If you're involved in AI implementation, decision-making or strategy within your company, this is your opportunity to share your insights with a global audience of your peers. If you believe you can help other leaders move the needle on AIROI, visit Emerge.com and fill out our Thought Leaders submission form. That's Emerge.com and click on be an expert. You can also click on the description of today's show on your preferred podcast platform that's emerj.com/ex1. Again, that's emerj.com/ex1. We look forward to hearing your story. Without further ado, here's our conversation with Yasmin. Yasmin, welcome back to the program. It's a pleasure having you. Thank you so much. Thanks for having me. Absolutely. We've been in talks for a while to have you on these episodes. I'm really delighted that by the time we actually got into the podcast studio and recorded, you had jumped industries, you had jumped spaces. I won't go into all the details, but it's great to have that perspective from across industries and bring it into the healthcare space, especially the MedTech space when we've had other episodes on the show. Of course, talking about this really interesting space from a regulatory standpoint, a compliance standpoint. Last episode, you held our hand and walked us through what it looks like from an audit perspective. Today, we're talking about use cases. Last time you brought up the challenges that we're seeing in the healthcare device spaces. Where are we really seeing the needle move on AI ROI for AI being deployed in audit workflows? Maybe we can start with the healthcare space. But if you're seeing this outside of that industry, of course, we've got your perspective here. Yeah, sounds good. I mean, I am sure there are a lot more use cases than I can think of, but there is one that I'm sure a lot of people in the healthcare industry are thinking about or starting to use or use. And I would say that's probably in the billing and claims audit area for internal audit departments. In this industry, we are dealing with high volumes of billing and claims and patient data. So it's very hard to, it's challenging as we talked about last time to audit everything or audit a good percentage of transactions even because the volume is so high. So that's where AI comes in. And I think not only the volume, but also the coding, right? Healthcare coding. It's very complicated in this area. And AI can really help in the area to ensure accuracy. So one of the areas that I was thinking we would go over today is how AI can analyze basically every billing and claims transaction in real time, flagging, upcoding, duplicate claims, policy violations, suspicious provider behavior. I mean, this is I think really important in healthcare. Yeah. It sounds like from there, I mean, we've had so many folks come on the show and tell us very point blank, you know, for those listeners outside of the industry they should know if they've haven't figured this out from being a patient or paying attention to the news that a lot of hippo regulations and a lot of the way that healthcare laws are designed are designed to facilitate payment. They're not designed to facilitate care. But it sounds like for, there's a wealth of, information across workflows coming from the use cases that we're seeing in billing or that the data that we're extracting from billing can improve care can improve the patient experience. And we can kind of work backwards, maybe, maybe from that problem. Is that, am I reading that right? And what you were saying about billing as a use case? Yeah. Yeah. I mean, I think you're absolutely right. Looking at all the data, you know, the AI is all about like the AI learning, right? The industry, the training of the AI. And I think there's a lot of potential in there to improve the patient experience in the end too, just which is basically the end goal. But with the AI learning, you know, typical, like the diagnosis to treatment mapping, you know, the procedural codes, provider billing patterns, right? To make sure something doesn't get denied because you build it incorrectly. And that impacts the patient's experience in the end. They're not in the healthcare, but also the whole experience of getting your claim paid. So I think, you know, that's where AI can really help not just to ensure accuracy, but also avoid, you know, revenue leakage, fraud risk, and cut down on appeals. Absolutely, which is, which is its own administrative burden really that that's downstream from the system itself being so focused on payment rather than care. But I think we're seeing this writ large, especially in healthcare spaces where, you know, the battleground is now becoming patient care, patient experiences. They're looking over at their colleagues in the financial services space, going, Hey, if banks can figure this out and people are just as, you know, intense about their money as they are about their bodies, that we can maybe learn a thing or two and deploy this in our space. I used to think that that was taboo to even mention out loud. And even 2023, I probably even best if I forget the name of the guest who was on the show, but she came out right out and said it. I was like tap dancing around. She's like, no, you can just say we're looking right at banking and saying, how do we make patient experiences? Like, how do we look at them the same way that they look at customer experiences? Knowing that customers are very different from patients in this respect. I would almost say, I mean, I would almost argue that the healthcare industry sector is even complicated, I think, with all the treatments, voting and bundling and unbundling. I am finding more and more. It's an extremely complicated industry. So for auditors, again, for Mac Day, I, it's really helpful. But it's a very easy to implement, which I'm sure we'll talk about, but it's easier said than done. There are a lot of challenges, I think, to get it up in money. Right. We've had folks come on the show talk about how deep the compliance issues go, especially around privacy, especially with that tug of war. This is elsewhere outside of audits, but just in that tug of war, if we have really good patient data, we can help everybody and we can develop cures for all kinds of diseases. But there are all these laws on the books. And of course, if patient privacy is so important to the patient. Yeah, but that does impact all audits to your point because, you know, as we're setting up these systems, right, we'll say to help us do the audits, we have to consider the security and privacy concerns very, very carefully, you know. If I'm setting up, I'm using a tool to go into our billing system to extract claim data. I need to make sure that, if I'm using a vendor, especially that that vendor has all the security and privacy considerations in place, that the data is encrypted in transit and at rest. I mean, the restriction who on their side is accessing that data, there's a lot of compliance considerations for auditors as well. Not just auditing compliance, but for us to be in compliance. Right. And let's double click in there a little bit more, just given your diverse background in terms of industries and that perspective that you're bringing to the healthcare and med tech space to deploy AI, especially around this billing use case. Tell us a little bit just about those challenges, how they differentiate. And I know you were just saying a moment ago that privacy is a huge part of it, but how it differentiates from your experience in financial services. Yeah, I mean, I think I think the healthcare industry, as far as, you know, I, in my experience, first of all, I think one thing it's has grown a lot with acquisitions, right? So, I mean, not only that though, but data integration in general is, you know, a big challenge, I think, to make sure that, you know, all the data in the claim systems, I mean, they might be fragmented, customized based on, if you have acquired other entities, what their systems were, how they were doing their claims. But even if you're not, the volume is so huge, and the data integration to make sure the AI tool connection is working right, it's not slowing down the system, and it's, and that the secure, right, is I think a big challenge, and it's something that in front of a lot of there's need to work with IT support and security chains very, very closely on, and you know, if there's a middleware needed to make sure that that middleware is monitored, and it's secure, so that's, I think data integration is one. - Yeah. - The other thing is the poor data quality, right? I mean, hopefully the systems are in good shape, but I would, you know, there today, I'll probably, probably a lot of companies deal with some inconsistencies in billing codes, or sometimes missing diagnosis data, outdated provider information, or incomplete billing codes. I mean, the data cleanup, I think, is really important, but it could be such a big task, right? Again, if you're dealing with a lot of acquisitions, to do that, the other thing I think is that it could be an issue is like the lack of labeling, for AI to learn, you know, there needs to be some labeling involved, right? But if you don't have good examples of what you mean by, let's say, let's say incorrect billing codes, if you can identify and label those, if you need to come up with, you know, almost, you know, synthetic cases that you put in there for AI to learn, it could just, you know, the labeling can become a complicated and a challenge. And of course, like we discussed, the privacy and security, it's always on my mind, you know, when we're implementing these things, because the last thing you want is to turn a lot up to cause a compliance issue, but we need to make sure that data retention policy we're in, you know, in compliance with them, obviously, HIPAA and other regulations is just a lot to think about. - Yeah, absolutely. I'm also curious, you know, where what you think that maybe we'll ask this question, I was gonna say, maybe save it for later, but maybe it's appropriate for billing, just because you're working in this, and then same mindset, relatively as financial services, at least where you're looking at balance sheets and the starting from the place of prices and, you know, costs as the kind of gateway to asking the next questions about what does this mean, and what does it mean for the patient? What, how do we want to use this data towards particular goals? But especially in this, in this billing case, what do you think that the healthcare industry might be able to learn from the financial industry, especially now that that really does not seem at all taboo anymore, even for my sensitivity just a few years ago. I mean, I think maybe I'm just gonna go back to the data planliness, you know, to make sure that the data, I mean, in the financial services industry, data, data, they're usually required fields, and their accuracy is, you know, always monitored and it's consistent. I think the consistency and monitoring of the data inputs, which is, but of course, very, very hard because there's, you know, all the incredible volumes of transactions going through. But I think that, you know, with AI, right, in, you know, what is it, garbage and garbage out? Like, you just gotta make sure that that accuracy is in place upfront. Of course. Otherwise, it's just because I think chasing your tail, you know, to get that done without it, it's, you know, you're gonna end up with full positives. - Yep. - In AI, and it's just gonna, you know, I can take away from the credibility of the results. Outside of billing, where else are we seeing in use cases where AI is really driving the conversation around developments in healthcare? - I mean, I think the one other area, this is definitely for, I mean, healthcare, but any company too, is, it's similar to billing, but it's in the ERP systems as well, for internal audit departments. You know, we are, again, as you know, making selections of accounts, their transactions and auditing them. But with AI, if we do connect it to the ERP systems, you know, we can identify, sometimes, you know, obviously the billing is done in the ERP systems, any billing issues, but also duplicate payments, policy violations, procurement, right? I mean, we're in healthcare or in home medical equipment, we're buying a lot of equipment and inventory, you know, make sure that the transactions going through their procurement systems are appropriate, making sure there are two duplicates or inappropriate transactions. I think that's another area that not just for healthcare, but any company, journal entry postings. - Right. - If you can identify the rules for the AI to learn, there's so much benefit for AI to just, you know, real time monitor. And I'm not gonna talk about this too much because I'm no expert in it, but with the AI agents, especially, - Interesting. - I can see the future where the AI agent is going through and doing the continuous transaction monitoring, putting the exceptions into a report for you, you know, sending it out to people, scheduling meetings, and where you discuss the exceptions, and then in the end, writing a report over it. So my God, you know, the. (laughs) - Yeah. - There's a lot of opportunity. - Yeah, it sounds like with both use cases that the simple AI deployment will get you most of the way there. The deterministic deployments, really simple capabilities, machine learning, predictive analytics will get you most of the way to solving a lot of these problems as an organization. Just pulling from your last example and billing, it's easy to see, you know, especially where you were discussing data integration or the different kinds of data that's going to be involved in translating what the bill looks like from a healthcare provider standpoint, versus what the insurance company is interested in, versus, you know, what the hospital is interested in. Translating that for all those different stakeholders, that to me sounds like you're a generative or a more probabilistic deployment that's going to solve being that Rosetta Stone for everybody that needs to touch billing in a different way, even outside of the audit space. It, you bring up agents here. Where else are you seeing agents, especially for the audit side in healthcare, at least the opportunity for agents to be deployed in such a way where it's an advantage that comes from the technology itself. - Yeah, I mean, I haven't used agents a lot, but one area I have used, I'm just dipping my feet into that area, but one area I have used is risk assessments. For, you know, to, once we've done the, you know, discussions, let's say with various individuals, as in turn allotted across various individuals, for the agent to put together a risk register. - Right. - From our discussions to identify the mitigating controls in place, putting a report together and then scheduling meetings, you know, one quarter from now with those individuals for us to chat with them again and update. So the one area I've used is risk assessment, just as an enterprise level. But I know from my research, one area I can expand on is risks with the claims data. Also, where are the potential risk areas, where there could be errors in the claims, or potential anomalies, and the agents can start identifying those on an ongoing basis. But I haven't done this yet. - Yeah. - I'm not identifying, hooding, it's, I'm considering it. - Now, I'll pause here because I'm gonna go on a little bit of a phishing expedition, and you did mention with those last two, you don't have the most familiarity over it, but you engaged my curiosity. But really interested to know where you've seen a genetic, it looks like these two spaces with, you know, risk, with risk assessment, and then risk in claims data. Those are still back office tasks that we're not really directly, directly touching the patient, and this is hardly a genetic surgery, so to speak. We may never get there, just from a comfort ethics standpoint, or even a marketing standpoint, but this still sounds like you're gonna need humans in the loop to really develop these systems and making sure that there's an accurate reading, risk assessment ain't nothing, as should all AI deployments be in that, it's gotta be, you gotta deploy it on something important and at the heart of the business, but you can't afford to be wrong about the stuff. Just curious, you know, what human in the loop is involved in ensuring that, what you're seeing from the agents there is good. - The human in moment is, of course, as audit, I think, like for example, if it's not audit, we're just talking about general areas, let's say automated claim processing. I think there's room for a genetic AI to obviously help a lot there, but for an audit, if you're, saying something is an exception, I do believe that the human immobility is necessary there to confirm that it's an exception, that it's not a false negative and then you know, you're you know, having people spend time on it. So I do think human immobility from an internal audit perspective is going to come when a gender AI reports to you an exception, right, that it has identified during its continuous monitoring. And then, you know, the again, human is maybe going to schedule the meeting, but then human intervention will be there at the meeting to, you know, meet with the individuals involved and determine next steps. But I feel like the team of AI agents is going to do a lot of the work for you to identify the exceptions to keep on, you know, I add on it as you're doing as internal audit as a human, you're doing other things, right. And in the back work, that team is working for you down to scheduling meetings for you to discuss the exceptions with individuals on the periodic basis. Take taking a larger step back that focusing on where you need to cut costs through exceptions, through systems where we know this is a mess, we know there's waste here, we know we're leaving money on the table, so to speak. But if we just use a very advanced technology that can take that microscope and find us all the places that we're leaving money on the table, that that's worth the investment. So I'm getting a really strong lesson here, just in terms of, you know, even AI deployment writ large, not just to gentick, but the point remains that if you're a great place to deploy these, these technologies is into find it exceptions where we know that traditionally in manual workflows, we're kind of playing it as better safe than sorry, whereas we don't have to be, well, we have to be as safe as possible going forward, but we can deploy technology to make exceptions in cases that can help bring down costs and not endanger patients, especially where we have humans in that loop. But I think where you're drawing, you know, that specific emphasis on that you're looking for exceptions, I think that's a really great beach head for folks to take home across industries from listening to this episode. I just want to pause here really quickly, because we've gone over a lot of use cases so far. Financial services tends to be, you know, much closer to the tech space, much faster to adoption. I know we were bringing kind of what could, you know, healthcare learn from financial services as an industry about deploying these technologies, but just any advice from from your larger background for healthcare leaders, deploying technologies in this in these spaces to really make these use cases a success. I mean, I think the what I have found is that implementing these use cases requires a lot of people's input and a lot of individuals experience. Everybody adds their own area of expertise. Like I mentioned, even for internal audit implement and AI tool, it needs to deal with the IT teams for integration. It needs to deal with data scientists for building the right models. It needs to be more with the users to train the AI tool or feedback loops. So there's so many people involved and what I would say, I think that the first thing is to start out with a pilot not to be too overwhelmed with how much is it going to cost or is it going to be successful? I would say start with a pilot on a subset of transactions. In more of everyone early on, use the feedback loops to improve AI accuracy over time. And I believe one of the most important things is to use a tool with a strong explainability feature. So that when the results come, especially when we're talking about exceptions, yep, the tool shows where the accepts you loss. So you're not as a human now redoing the work to compare what where's the exception? I don't see it. Right. But that the tool gives you that. And just speaking to the beginning of your point, you know, I think I think a lot of listeners executive struggle with I'm supposed to choose something important to the business, but low risk. And I'm supposed to choose something where, you know, we can do it very small, but it's at the heart of our business goals. And I think the formula I've heard from other guests is do 10% of an extremely important problem very well. And just in terms of siphoning off, you know, a project small enough to see results, but important enough that it's just not a shiny toy. And I think that that kind of ratio of you don't need to solve the whole problem, you need to solve 10% of it way faster than you ever have before that's got to be the goal. And it or that's got to be the sort of the success you're looking for. I know you were going to chime in there. Is that something you've seen out there? Well, I'm sorry, I didn't mean to. Oh, you're fine. I was saying one there first and I do think like you mentioned low risk more than low risk, I would focus on high value. Yeah, value outcome because it need I believe it needs to be an important area like you mentioned. You don't have to do the whole area, but pick a good subset of an important area that others can see the value, not just an internal audit, but as a company. And it's tough, especially because it's about having that exceptions mindset of you've always thought this was the status quo. This is the way life is that this isn't just a problem, but this is a problem that that's built into the way things are done. But if you truly understand the problem from the lens of data or what the tools are really capable of, you don't have to, you can find the exceptions in this audit process where you always thought this was, you know, impossible for for human beings lay down. So just moving everybody to that that mindset as well, that data mindset of the kinds of problems that can be solved. It's a tough challenge. Yeah. And the exceptions again, I want to please clear, except in result in a value being identity, right? You can identify an exception that you didn't build something or you didn't, there's an unfulfilled obligation. So which could help by fulfilling it, you're helping the patient and by fulfilling it, you're helping the company recognize revenue. Yeah. So you're not leaving revenue on the table. So I don't want it to be thoughtless, just find exceptions to put an audit reports, but let's find the exceptions to add value to the end user and to the company. Right, right. I think I think the reason that that healthcare and financial services are looking so much at each other's homework is because they're running into the same goal, which is, you know, actually the whole battlefield is for banks, customer experience for the healthcare space, the patient experience. That's where the money is to be made, but also really the value of the industry and what good it's bringing to all of us. And if we have technology that's reinforcing those incentives, that is a very silver lining to the much, much uglier headlines I see about AI all the time in the news. And there's plenty of that silver lining to go around. Yes, man, thanks for coming on the show this week and showing us where that is. Thank you very much. Rapping up today's episode, I think there were a number of critical takeaways. Here are three for healthcare and enterprise leaders to take from our conversation today with Yasmin Agitan, senior vice president of internal audit at Adapt Health. First, audit and compliance teams must adopt to a rapidly evolving landscape. Digital systems, regulatory requirements and operational complexity are accelerating and leaders need to prioritize speed, cross department collaboration and transparency. Second, organizations should treat improvements in data visibility structured workflows and explainable processes is ongoing goals. Progress in these areas can help reduce audit cycle friction, clarify ownership and strengthen operational oversight. Finally, AI and data tools are most valuable when applied to identify exceptions, anomalies or inefficiencies in existing processes, leveraging these insights can drive measurable improvements in accountability, decision making and overall enterprise performance. Interested in putting your AI product in front of household names in the Fortune 500 connect directly with enterprise leaders at market leading companies emerge can position your brand where enterprise decision makers turn for insight research and guidance visit emerge.com slash sponsor for more information. Again, that's eme rj.com slash S P O N S O R. I'm your host at least for today. Matthew Damello editorial director here at Emerge AI research on behalf of Daniel Fajella our CEO and head of research as well as the rest of the team here at Emerge. Thanks so much for joining us today and we'll catch you next time. [Music]