AI in RCM: What Healthcare Leaders Should Know

Introduction to the Webinar

This is Claire Wallace of Becker's Healthcare. Thank you so much for joining today's webinar, "AI in RCM: What Healthcare Leaders Should Know." With that, I'm so pleased to introduce today's speakers: Jonathan Wiik, Michael Waluk, and Pinaki Ghosh Ray. Thank you, John, Michael, and Pinaki, for being here today. I'll go ahead and turn the floor over to you to get us started.

Thank you, Claire. Thank you, first and foremost, to Pinaki and Michael for joining me. Welcome to my world. I think I'm on a webinar at least once a week or every other week, and I'm so excited to talk about what we're about to discuss today. We're gonna dive right in, if that's okay.

So, I really want to just define AI for you today. I am not the AI expert—that's Mr. Wallach there. He'll get into it in a little bit during his section.

But I think I remember the days when I was in revenue cycle—and Pinaki remembers this too, I bet—where you heard these two letters and immediately replaced them with the letters B.S. You were like, "What does AI really mean?" You gotta talk about whether it's RPA, whether it's machine learning, what the use cases are, and where we are. I like this diagram.

FinThrive put this together to talk about this broader category. Think of AI almost as a library, and then you've gotta check out these books within it—or a website, if you will—and you have to go to these different tabs within it. As you enter tabs and go into different windows, it gets more and more finite with the level of functionality. I would argue this Venn diagram also has some interdependencies as well.

Chapter
The Journey of AI in Healthcare

You can't have natural language processing without machine learning, for example. A lot of it is going to be foundational on those datasets that are there to start your journey. And we are on an AI journey. I think it's starting to finally get a lot of momentum in healthcare.

It's been around for decades, but I'm starting to see, especially on the finance side—it got a little bit more of a bump on the clinical side for obvious reasons—but we're seeing use cases. Pinaki and I are gonna walk through those with you today, just in terms of billing and patient access and some of the analytics and things that are there, and really help define what it is that we're doing in this space.

AI can make up funny stuff. Are these chihuahuas, or are these muffins?

It doesn't know the difference, and you have to know the knucklehead things that it is. It's a computer. It's going to take inputs and create assumptions. It's getting better at it, but as you're listening to Michael and Pinaki today, I want you to think about the application and make sure that you've got validation and monitoring and those types of things in place as well.

Because I think the genie's out of the bottle. I was at HIMSS a couple of weeks ago, and it's exciting and also terrifying at the same time. You've gotta really harness this in a way that makes sense to where your teams understand what it's doing for you and also some of the dangers and risks that are there. We don't have a lot of talk today about governance, but that's something I think we'll probably do in some sessions down the road to talk to hospital systems and get some perspectives from Michael, who's been doing this for a while, in terms of what organizations are doing to manage this.

Chapter
Committee Oversight in AI Implementation

Like, what does that look like from a committee standpoint?

Are there policies in place that say when you can and cannot use it? How much of it can you use? Is it exclusively used? What are your copyright and use permissions that are out there? Those types of things are all considerations as you go on this journey.

Let's talk about healthcare for a minute. The Journal of American Medicine—one of my favorite articles—it's getting a little bit dated now, but I really love it because it has these stratifications, like clay, that are talking about waste. They said there's about $800 billion of waste. Congress, interestingly enough, just passed a bill in the House and the Senate that's gonna get reconciled that wants to cut $880 billion of waste out of healthcare. That number is darn close. I know it's off by $80 billion to this one, and this is an older study, so maybe it's amortized up. So as much as you think about Washington, D.C., I think they did a little bit of their homework.

The layers—I'm not gonna get into a super huge amount of detail here; you guys will get a copy of this deck—but it talks about failure, care delivery, coordination, and low-value care. The stuff in blue really is, you know, the art and science of medicine, I like to call it. The stuff that's in orange really is the administrative part. We're gonna talk about that today in quite granular detail. At least JAMA feels—and they did the study with Humana, if I recall.

Chapter
AI's Role in Simplifying Processes

In a joint effort, a lot of that administrative complexity can disappear. Now, the only thing I don't agree with in here is they've absolutely eliminated it. Right? It's gone.

I don't know that AI is going to do that, but it absolutely should make things more simple for your organization. You're still gonna have levels of fraud and abuse, I would argue, and a little bit of waste. Things are still not gonna be quite priced appropriately, but the administrative benefits that come out of automating intelligently through AI, especially in your revenue cycle, are gonna be apparent. I'm not just saying this either.

This is a survey from Acosta, and they're saying, "Hey, how do you enhance your revenue management? What do you do in your revenue cycle?" And they have, you know, kind of a pretty equal or equitable pizza here, I would call it.

Find a strategic partner—that could be a FinThrive, right? Adopt new technology. We're gonna talk about that today.

Eliminate redundant systems. You'll hear FinThrive talking about the power of the platform, the platform effect, you know, reducing that redundancy. There's a cybersecurity influence there as well. The more nodes that you have out there, the more risk and exposure you have.

Most hospitals I talk to have 25 to 30 systems. Pinaki, I think, at north of 16, he's got a goal. He's gotten that down quite a bit. He's got a goal to get that down into the teens or high teens this year at Prime.

And what I love about this slide is that almost all of those solutions surround tech and finding a partner. About 80% of them, and we're gonna talk in a lot of detail today about that.

Chapter
Financial Impact of AI in Organizations

Artificial intelligence, per this study from McKinsey, has a $360 billion (with a "B") influence or possibility to organizations.

Excellent study from McKinsey. Happy to pass it along after this talk. It talks through specific use cases that I've highlighted there in green: coding, prior auth, denials management, CDI. Super huge, super important application.

Analytics is another area that I think is lacking within the revenue cycle. We all have slicer-dicer in Epic if we're using that. We might be using QlikView or Power BI, other tools. Those, frankly, can't connect some of the modularity and some of the insights that can happen from an eligibility response as it relates to the loading of a bill, as it relates to those dollars being posted, and the insurance follow-up and some of the denials and underpayments that are happening. Very, very important to have that. 750 hours of RCM's time annually can be saved by looking in the right places instead of spinning wheels.

I've talked enough. I'm gonna give you to someone way smarter than I am, Mr. Pinaki Ghosh Ray from Prime Healthcare. He's gonna walk through who Prime is. If you haven't heard of them, I'm sure you have, but if you haven't. And then he's gonna share some things that are going on there, and then he and I are gonna kinda tag team on the RCM stuff. Pinaki, thanks for being here today.

Chapter
Pinaki Ghosh Ray's Perspective

Thanks, Jonathan.

Good morning and good afternoon to all of you in different parts of it. Disclaimers—couple of disclaimers. I definitely don't profess that I'm smarter than Jonathan.

He writes with John Wiik, and you can understand why I don't wanna be smarter than him.

The second part is really around the fact that the next few minutes that I'll take from your conversation, I/we don't claim to be super users of AI right now.

Right? We are in a space of cognitive error where this is one of the tools that is out there in the market, which we have to absolutely understand a ton better.

A lot of people are getting to understand this a little bit more than others, and we wanna stay ahead of the curve as much as we can.

Some part of my conversation, as I inform you about Prime Healthcare, is gonna be pointed about who Prime Healthcare is, what we are doing in a specific use case around AI. And then I will partner with John to sort of help you understand what else is going on from a general perspective, what my fraternity of other providers is doing, or trying to sort of get into the space of AI that we talk about.

Who are we? What is Prime Healthcare? Prime Healthcare—and I can, I'm passionate about where I work and passionate about what this company stands for. And as you can see from a slight blowout, Prime—and I've been here for the last good part of the decade—is really around saving hospitals, saving jobs, and saving lives. So that's primarily who we are from a Prime Healthcare perspective.

Chapter
Mission and Values of Prime Healthcare

You can read through the missions and the values aspect of it. I'm not gonna regurgitate it, but the real value that we add as a provider network is to our patients and the communities that we serve. The differentiator that Prime brings to the table is that we are physician-led.

John, if you wanna just slide down.

The aspect of being led by physicians shows up in our ability to get recognition from the market and our peers from the perspective of the multiple different aspects that we get recognized for. The State of California has recognized us, and we are the second most recognized from a general perspective. Seven of our hospitals have received Patient Safety Excellence Awards.

We are obviously a top health system from a Fortune perspective.

We are in the "100 Top Hospitals," and multiple of our hospitals are included in that. Healthgrades has also been generous in recognizing our ability to perform clinically as well. If you move to the following slides, this just goes to show where we are at. We are across 15 different states in the United States and have about 50-plus hospitals, which the following slide is gonna really walk you through. We are also broken up between two parts, from a general perspective: foundation hospitals and for-profit hospitals. So that's something unique about Prime Healthcare. It allows us to be in markets where it's more attuned to serving the communities, and there are obviously aspects around for-profit as well.

Chapter
Strategic Focus on AI in Operations

If you go to the core of our conversation—hopefully, we talked enough about Prime Healthcare, and I'm open to taking questions toward the later part of it.

From the standpoint of our overall strategy—and AI definitely features in it—we have focused more on ensuring that our physician-founded and physician-led organization runs in that same model of acquiring any new hospitals with a perspective of driving standardized work across anything that we bring in. Obviously, that goes from a clinical aspect of it, but it also extends into the revenue cycle standpoint and into anything that we are doing from an overall artificial intelligence perspective.

So, running to the core of it, what we have done—and we'll use a specific example over here—is really our investments into technology and specifically into artificial intelligence. One example of this is our partnership with STEER Health. That's something that has been ongoing for the last two years now. We have launched, in partnership with STEER Health, a platform called "GetCareNow," which is designed to address any new and evolving healthcare needs for our customers.

What does it do? It focuses on creating easy, convenient, and affordable patient care, which is really more patient-centric.

Chapter
Real-Time Feedback in Patient Experience

Our strategy is to utilize state-of-the-art technology to improve our access to care, enabling real-time feedback on patient experience.

The specific aspect around artificial intelligence really gets into the core of it, where we start looking at GetCareNow, where the patients can actually start communicating with us. And that's where our journey to support the patient's consumer experience gets into the aspect of artificial intelligence-enabled conversational texting, live customer navigation, and patient-centered 24/7 virtual care. So the strategic aim of GetCareNow, in addition to getting better health outcomes and a lower cost of care, is to ensure patients feel known and valued whenever they engage with our Prime Healthcare network.

With that being said, I think we’ve covered most of it. You can see it's more than 10,000-plus licensed beds, serving across 600-plus communities.

We have written off $12 billion to $14 billion in charitable contributions.

Chapter
Capital Investments in New Entities

And, whenever we are acquiring any new entities, we put in capital investments to ensure the community and the patients are served better.

Cool. Well, thanks, Pinaki. That's awesome. I, you know, Prime's a behemoth out in the West.

I think I appreciate the not-for-profit/for-profit hybrid there. It's pretty unique. The investments are there. Pinaki and I are gonna tag team on this next section here and really talk about AI and revenue cycles specifically.

We're gonna go deep into some use cases—not super deep, but just kinda bounce back and forth and banter a little bit as we're walking through these next sets of slides. You know, these are challenges that I'm sure everybody on the call has seen.

I had, you know, 15% to 20% turnover at my house, Wade. Pinaki, probably a better leader than I was. You know, I think that's just the nature of the business too, especially in those entry-level roles. There's been a lot more, I would argue, compression in terms of price banding or average hourly rate banding in most communities now.

Healthcare used to be one of the more, probably favorable employers to work for. And as fast food and retail and some of the other markets came up, we compete for those now, and that results in turnover. And cash is something that's always stuck. It's never moving as quickly as we would like in our organizations.

Chapter
Understanding Claim Edits

And claim edits certainly are something that I think everyone measures and touches that I talk to, and understanding where those ICDNs and DCNs and CORC and ROC codes and posts and lockboxes and all of those touches happen. Automation can help with all of those things. I think we outsource some of this. We handle it in-house. We try to do what we can as an organization as we move forward.

These are use cases across the organizations that Clarivate or HBI measures.

And what I found impressive is this is kinda like a nice piece of art. There's not really any one big case that folks are using it in. That big pocket there was payment posting, that second tranche. The one there in the middle, insurance verification, that makes sense. And as you get over to the right, you start seeing some things in posting cash as well.

There's a large swath of use cases as you're moving across. This is just another laundry list of different things that we're using AI for—things that we encounter with our customers that Prime is using as well across eligibility workflows. I think eligibility and claims represent large areas of automation opportunity that certainly are the two, I would argue, cargo ships that pass into the port of revenue cycle more than anything else—eligibility and claims. Pinaki, do you see or have ideas about automation opportunities at Prime that fall inside or outside of this list?

Chapter
Evaluating Automation Opportunities

Or are there some things that you wanna highlight as we go through the next slides? Or are you still kind of embarking on this journey, as you mentioned?

So, as I said, I think, from a general perspective, us as a network are evaluating multiple different options within this aspect. Right?

In general, we're not generally talking from a larger sort of provider network friends that we have, and we're talking about it.

Most of these things, if not all of them, are being targeted and sort of reviewed. I'll hit on one specific area, which is really around the fact that if you look at the challenges around the insurance found avenue. Right? So you have aspects around insurance verification that is going on.

It's gonna be completely contingent on how our staff—any staff from a patient access perspective—goes in there and starts reviewing it and then goes back and sort of updates it. You would lead into situations where they're doing some kind of quick reg, somebody's coming back and catching up later, and so on and so forth. Having opportunities for AI, which is coming in and the ability to sort of update information on queues, especially when you don't have a laundry list of payers, which is infinite. We have a finite number of payers that are out there from a patient's perspective, especially if you know you're in a specific neighborhood, you would know exactly what's out there.

Chapter
Revenue Recovery Strategies

So leveraging AI becomes a lot more opportune from a general perspective in that specific area. But then again, it doesn't cut out the other items over here as well. Back to you, John.

Thank you, sir. I like saying it's your revenue. Go get it. It's your chance. You could do things for free as an organization, and you heard about the billions of dollars that Prime donates back to its community.

I think if Prime had the choice, it would have preferred to have had some sort of funding mechanism than having a write-off. Most hospitals are. You absolutely are there as an asset, as a steward to the community, but it's best to be paid for the services that you provide. It's a non-sustainable model over time to have 100% charity care. And in the survey in the hospitals I talked to—just survey in a straw poll manner—there are all kinds of levels of opportunity in terms of automating and looking at where things are at.

The value drivers for this are the big three: where is the problem, is there low-hanging fruit—or sometimes I like to call it high-hanging fruit, where there are some things where you're getting from good to better to best. How much time are you saving? And then, is this scalable? I've seen lots of things bounce off the atmosphere as folks are doing this automation, and they're not applying it in a way that is changing or moving the needle within their revenue cycle.

Chapter
Foundational Processes in Revenue Cycle Management

Pinaki and I are gonna walk through three or four of these. We'll see how we're doing on time. As I mentioned, you'll get the slide deck here. But these are foundational, I would call blocking-and-tackling type processes that occur within most revenue cycles. Michael has a couple of examples of some of the innovations that we're doing, so stick around toward the end of this where he's getting into some denial workflows and appeal workflows that are there, and also understanding just contract management and what that looks like, which really drives a lot of these other things.

When we talk about the front end right now, these are four major processes that every organization pretty much has to do to get financial clearance on the front end from their intake standpoint. Eligibility, obviously: Do you or do you not have insurance? Do you have multiple insurance policies? What's primacy?

Insurance discovery is being used there quite a bit. The MBI lookup is still very real. Are you looking at the identification number and able to match that up with the cards that are in front of you? And then, are there information request letters? Do you have coordination of benefit issues?

The notice of admission (NOA), the consents—those types of things are all things that can absolutely drive challenges in time and automation in your organization.

Are there things in patient access that Prime is looking at, Pinaki, that fall into this or maybe another category?

So I think we're using, as you had covered, some of these aspects of it. Right? So having information around a person's presumptive eligibility aspects of it—that's something that we're definitely looking at. Looking at avenues of keeping on improving our insurance discovery process—that's integral at any point in time. Right? I don’t need to profess how much of a differentiator it is for a person or a claim if you are a self-pay claim versus if you actually have an insurance aspect of it. So that's also there.

Chapter
Evolving Prior Authorization Processes

Work around prior authorization is also evolving, and we have to keep evolving with it. So that's constantly moving toward the goalpost kind of a situation.

Great. Back to the back end here.

Pre-billing scrubs, I think, are pretty common in the market. Late charge management—the OR used to drive me up the freaking wall, for lack of a better word, with our implants, our stents, any of our implantables, some of the infusion drugs. We had them built in purchasing. We paid for the dang thing, but we did not have the exact invoice and the J code or the C codes attached, and that really jacked up, for lack of a better word, our DNFB.

And there's a ton of AI out there that can match things based on the surgeon, the people in the OR theater, the patient that's there, their age—those types of things to help create a checklist and things that are there. Populations on type X7 bills and claims allow bills to get charged better with an independent claim number. Resolution and claim edits as well within the work queues. From the people I've talked to in Epic, they're kinda sick and tired of work queues. There are too many work queues right now.

They're trying to consolidate those down. Again, big benefits here in terms of wasted time, staff focus, reduced billing costs, expedited cash—those types of things. Pinaki, thoughts on the back end?

I think there's a lot to be done in the general perspective, from a scope of what AI can come back and do for us. I'm extremely hopeful around more items coming on board, but it's too early to call it right now.

More stuff from the billing office—third-party liability is a big deal.

Faxing to Medicare, which I believe they still do. I was just out at a conference—a revenue cycle conference out in San Antonio—and they were talking about how they asked the room, "How many people still fax?" And everybody raised their hand. And then they said, "Is it less than 10% of your business?" And almost all the hands came down except for a couple, which is great. So that tells you there's still paper going back and forth between hospitals and, I would argue, payment vendors or post-acute care, those types of things, and the payer.

If that's happening, that absolutely is something you should talk to FinThrive about from a document management standpoint. There shouldn't be fax machines that exist anymore, to be frank, and there's ways to make that an electronic submission. Also, just from a record and indexing standpoint, there are companies that we partner with and others that help make that process just a lot more efficient. The same goes for the Fujitsu scanners that exist on the front end.

We didn't talk about those, but there are just ways to remove that manual step, and it's also just more accurate. I'll skip on down since you already talked about the back end a little bit, Pinaki. Let’s get into cash.

Chapter
Claims Workflows and EHR Integration

Let's talk about accounts receivable and really claims workflows more in the back. But EHR workflow integration from claim statusing, leveraging that 276/277 transaction, managing those 835s as they're coming back as you send out your claims—are you matching them up with the ERA, the EFTs, the remittances?

Those medical record denial solutions—one of my favorite use cases for AI, specifically RPA—is to put a claims hold on all payers that require medical documentation attachments with the claim. Go into HIM, go script that, pull out the associated SOAP note, H&P, discharge summaries, write the letter, put it on top of the claim, then send the whole package out instead of playing tennis and it’s 40-love three weeks down the road. You might make your DNFB go fat for 72 to 96 hours, but you're gonna save yourself a week of time down the road by automating that upfront. And there are some very payer-specific rules that impact that. And then we talked about patient request letters. But are there things in the cash realm, Pinaki, that come to mind, or are you still kinda dipping your toe in the water here as well?

This is the "we’re dipping the toe" arena. This is cash. Right?

You can't mess with that.

That's alright. No worries. These are some examples of some use cases that FinThrive has as well. Delivering out-of-the-box intelligent automation—I love that term. But we've measured it in terms of FTEs and ROI. And I think as you're evaluating technology, be it from FinThrive or anybody else—and I know Pinaki does this as well—you absolutely should at least get what's in those green ovals known, defined, set as a goal, monitored, reported back on, and adjusted as you go forward.

Chapter
Measuring Improvement with Automation

I think if you're using automation, intelligent automation, or any form of AI—be it machine learning, RPA, or some of the generative AI that Michael's gonna talk about—it should make you better. It shouldn’t make you worse. And what is "better"? Keep it that simple.

How are we measuring "better"? We didn’t have this before, and we seemed to be doing okay. If we have it now, what is the delta? What is the incremental improvement that we’re doing better?

There are some use cases here that walk through that. Simple things like, "My team was able to touch 12 more claims a day." Great. The next time there’s a position open, if your productivity benchmark is 60 claims a day, and you put this in five different claim workflows, you may not have to fill that position. So you saved an FTE in terms of claims follow-up touches because you’ve automated that work through workflow and RPA.

It can help to optimize when used properly. And, Pinaki, I’m gonna lean on you here a little bit because I know that you’ve seen things go bad in your career.

It can drive tremendous cost savings. We’ve shown you that today. It absolutely can maximize your workflows, and it can make manual, repetitive tasks go away. But it’s important to have an enterprise-level mindset. Prime is a very large, 51-facility organization, almost nationwide—coast to coast for sure—and has not deployed AI in a way that you would think in terms of the size of the organization, and that’s intentional. Can you double-click on that for me, Pinaki, and help me understand? You guys are very critical of the tech that’s coming into your organization. Why is that?

Chapter
Prioritizing AI Implementation

So, a few things that come to mind. Right? From a general perspective, we have to be absolutely sure about what we are bringing into our organization.

The second effect is really from a prioritization perspective.

Our core competency is clinical. Right? We stand out from a clinical perspective, and we want to make sure that anything we are bringing in absolutely resonates with that same physician-led environment about saving lives, saving jobs, and saving hospitals. So as we go through this new rush to deploy AI all across the board, we are being very diligent about what we should be looking at, what we should be identifying, and running with it.

As much as all of us on the phone who are probably also utilizing some shape or form—you’re using RPA. Right? So please be sure that RPA is separate. AI is separate.

Right? So RPA—absolutely, everybody’s using it. AI is more coming in from a general perspective of GenAI and others. So the way that we are thinking through it is we are being deliberate about making sure that there are no hallucinations-led items that come through and sort of mess with your overall operations.

And that’s part of the reason we are being deliberate about where we are implementing it or will implement it and making sure we are staying on top of it.

Excellent. Excellent. If you take anything, take that 60 seconds that Pinaki just said and apply that at your organization because I think that’s huge to understand. One, everything’s centered on the patient, where you’re at, and then, yeah, let’s buttress that with the other things that we need to do in finance to make that frictionless and touchless where we can.

We’re gonna focus it there. I like this. For any Flavor Flav fans in the audience, I had to throw him out there. I actually saw him in Vegas a couple of years ago.

He showed up with Florida.

Don’t believe the hype—Digital Underground.

You know, don’t believe the hype with a lot of these things. Dovetailing off of what Pinaki had said, AI absolutely isn’t a set-it-and-forget-it. You have to have monitoring, reporting. I heard Pinaki say the word "diligence" to make sure that, you know, you’re reticent about not having hallucinations disrupt your operations. There’s this fear that it’s gonna replace humans. And, you know, Jeff Woods is a person that spoke to Pinaki and me at a conference we were at a couple of weeks ago.

He said something that I think we’ve all heard, but I’ll repeat it just for folks that may not have. You know, AI is not gonna replace humans, but it’s absolutely gonna replace humans who don’t know how to use AI. People that know how to use AI are gonna replace humans that don’t. So it’s an upskill.

It is not something to ignore, gang. I think Pinaki and I and Michael are gonna really hit that home in the summary. This is not something you can say, "Yeah, that’s neat. I’m gonna wait for everybody else to figure it out over the next three to five years." You’ll never catch up. AI is going at such an astronomical pace. The generative AI things that are happening.

The deep-seek modules that are out there—and I’ll let Michael kinda get into the details. The tech, frankly, is going at a speed that is absolutely skyrocketing.

Emotion—I’ve actually dealt with AI quite frequently, and you can build emotion into it. I think it’s there. There is this level that you can’t trust it, and you heard Pinaki talk about hallucinations and making sure you have validators or bots for the bots, I think, is important. That removes some of the fragility or brittleness of it. You can’t automate everything.

Prime’s an example of that. Prime doesn’t have a huge level of penetration of AI at their organization on the finance side. But as a physician-led organization, they absolutely are putting it in the clinical side, starting there, figuring out what can and can’t be automated, and building a lot of those policy procedures in place to understand. And then I’ll skip down to, you know, the enemy of good is perfect.

Right? I think you gotta start small. I’m a lean guy. That incremental improvement of moving that ball up the ramp and sliding that wedge underneath it over time incrementally is huge.

Do you have any hype or other insights on this slide, Pinaki, or do you want me to boogie to the next one? Or—okay. We’ll go to the next one. I think you’re still on mute.

I didn’t hear you.

Big Yoda fan too. This is a checklist. I’m not gonna get into a super huge amount of detail with these things, but, you know, you can’t "kind of" do AI. You either need to do it or not.

Chapter
Checklist for Successful AI Deployment

And I don’t really think you need to deploy it holistically across your whole organization. You just saw from Prime that that’s a great example of where they’re really double-clicking on the clinical side and understanding their patient-centered care workflows, their physician workflows, their Epic environments, their instances of Epic, and really standardizing there. Then they’ll expand into some of the things that we talked about in the front, middle, and AR areas of revenue cycle. Stability is one I’ll focus on—the third one down.

You do not want a volatile process. That’s gasoline on a bonfire, I like to call it. If you’ve got a bonfire, you need to put the fire out before you start trying to make it better. AI is an accelerant.

It absolutely is an accelerant to things. You wanna make a good process better, not a bad one worse, if that makes sense. And can you measure it? We talked about that, especially in terms of hours or dollars or maybe turnaround time or just, frankly, throughput.

Pinaki and I are gonna be quiet for a little bit, and we’re gonna let Michael talk about what FinThrive’s up to from an AI standpoint.

Michael, lay it on us.

Thank you for coming. Oh, yeah. Thanks a lot. I appreciate it. Yeah. So I don’t think anyone will be surprised that AI adoption and investment across healthcare is increasing because it’s growing everywhere.

Chapter
AI Adoption Trends in Healthcare

Clinical use has been at the forefront, but RCM use cases are evolving. We’ll see significant investment in the next few years. That’s true at FinThrive too, where we’re focusing on those that reduce costs, increase our customers’ revenue, and streamline their work. So on the next slide, I’ll talk about our AI evolution at FinThrive.

This one’s a little different. Okay. Today, we have actually over 50 use cases for AI to prioritize and work through. This slide lists a few of them, and we started by developing machine learning models that predict key metrics like cash flow and identify claims that are most likely to be paid, and then we started clustering denials and underpayments so that they could be appealed in bulk.

The next step in our evolution was using document intelligence to extract data from photos of insurance cards and licenses, improving accuracy and efficiency for our patients.

And that was great.

So when LLMs and GenAI arrived, we formed a cross-functional team that meets regularly to discuss AI initiatives, progress, new technologies, and ideas. Oh, and I should mention that we’re starting a client group as well, focusing on AI and security, and we welcome anyone that wants to participate. A community like that can help you get started with your internal best practice groups.

One of our group decisions—yeah, one of our group decisions—was to roll out GenAI tools for every colleague to use, whether they’re developers or customer service reps, and this helped our organization understand what AI is capable of. And it’s generated ideas from every group in the company, so that’s how we got a lot of our use cases. Everybody’s thinking AI-first and understanding what it can do, and our use cases are piling up. The next slide has a few more of them.

Chapter
Leveraging Existing Infrastructure for AI

We’ve been building ML models for years and have a strong data platform that helps us move fast. But you don’t need all that to get started. Generative AI can offer you quick wins with much less infrastructure.

You could start with a use case that doesn’t require a major workflow change in your organization, so you can pilot it easily and then show some value early. And I’ll run through how we got started. Like, our first GenAI win was in prior authorization.

We used a large language model to compare policies and generate rules that were previously entered manually. Now we’re automating that policy monitoring and injecting rules directly into our Auth Manager app.

Our first chatbot was built for our internal support reps so they can quickly find solutions for customers’ questions or problems.

And NextAI was used to auto-reply to patient messages in our Access Coordinator bot. So you could say that that is speaking back to a patient, but it’s filtered through. And if it can’t answer the question, it ends up in the inbox of our end users. But this reduces the inbox clutter so staff can focus on high-value interactions.

And I just want to point out that this is an example of AI not being about replacing people. It’s about letting people focus on higher-value work. So it’s not to be feared. It takes a long time to develop these processes one by one, and you could say almost everybody’s job could be replaced by AI, but there’s so much work involved, and it’s quite expensive, so you’re safe for a little while.

Chapter
Future Chatbot Integrations in Analytics Tools

Soon, we’ll offer a chatbot in our analytics tools, and this kinda kicks off a series of projects. We’re allowing customers to chat with their data.

We’re making insights more accessible, whether it be around claims, payer trends, financials, or any other aspect of your business. We’re at the point now where we can start putting all these types of tools together, as we’ll see on the next slide.

Yeah. So here’s a good example. It’s an example of integrating these tools using agentic AI. It looks complicated, I know, but depending on your environment, it may be orchestrated within a single tool.

In the first step, we’ve got tons of denial appeals with known outcomes, so we’re training a model to suggest the appeal strategy that’s most likely to work based on what’s worked in the past. You know? So next, agentic AI has a lot of different definitions, I know we hear that all the time, but here, we’re just using it to execute a set of tasks based on the suggested strategy.

So in this example here, we have medical necessity, and the first part of that strategy is we need to retrieve the medical record from an EMR or EHR, and we use RPA.

Then we can generate an appeals letter justifying why it’s necessary, and that’s something that we can improve as we go on, tackling different types of codes and different types of cases. And lastly, we can use the RPA to appeal the denial with that generated letter in the payer portal. But we can also use voice RPA to call the payer, and then the same can be used to check the denial status too. And a feedback loop’s important to improve the process.

Chapter
Feedback Loops for Continuous Improvement

Each appeal’s result will help us revise our machine learning model, update the strategies, and improve the letter generation—all that stuff. Okay. So here is another example, and I figured people are wondering, "Where do I start?" I know I put a lot of it together on that last slide, and it was pretty fast.

But basically, you pick one problem—something that’s repetitive, high-volume, and annoying—and that’s your best candidate for AI. For us, contract loading is a good example of such a problem because, depending on your application, it could take days to enter the contract details into your system. We load thousands of them every year. So how do you approach it?

You could, if you’re small, give a chatbot a simple contract and a claim, and it could actually price it correctly. But when you have hundreds of complex contracts, you need a system. So one step to get started, you might use an LLM to compare the current contract with an update, and it could generate a checklist of changes to streamline the work. You could go further by splitting that work into sections—sections of a contract or service types that would appear in your checklist. And an LLM could be taught to understand each section. They take it one by one. Like in the last example, we could use agentic AI to walk through this checklist and generate data into your system for each section of the contract.

Chapter
Testing AI Solutions for Accuracy

You could test the result by pricing known claims. And then, again, a feedback loop’s important because it would help you continually improve your mispriced codes.

So there are many ways to approach the problem, but it’ll depend on your environment and skill sets. And I think the important thing is to get started, as Jonathan says. Yeah. I hope you found it helpful.

Great. Michael, that was awesome. Yeah. So I think, you know, it reiterated some of the things that Pinaki and I talked about there too, as well as, you know, dipping your toe in, making sure that you’re doing your diligence, you’re not creating hallucinations, and you’re starting small.

Prime’s example is they’ve started and really gone deep on the clinical side, and then they’re dipping their toe into some of these other workflows. Those are two very common workflows, Michael. Thank you for sharing them with us. Everybody’s got denials, even Pinaki.

And everybody, I think, has contracts to load. Those are two very repetitive, stable workflows to start from as we go.

GenAI is here. We’re gonna talk through each one of these, and we’ll probably go through them just a little quickly, just in the interest of time. But, Michael, I’ll probably lean on you a little bit as we go through here. But why—you mentioned this—you know, it’s expensive to install, but on the human side, and you talked about high-impact areas. But let’s talk about, like, one and three a little bit. I think you can kinda couple those together. Maybe one, three, and four, really.

Chapter
Understanding the Importance of AI Governance

What is the human element of GenAI? It is an investment. Why is it important? You know, this replacing humans is a fear, but why is it important to think about these things, specifically one, two, and four, as we’re walking through AI in our organizations?

Well, I think GenAI is a new thing.

And why it’s different is it’s nondeterministic.

The answer isn’t like one plus one is two. It can change. It can hallucinate. You have to put guardrails in place.

We have what we call a trust and quality framework that has some guardrails on the input and the output from GenAI, and then also we’re always constantly evaluating that. Sometimes it’s another LLM testing the answer of the first LLM, that sort of thing, and we monitor over time because, just like the machine learning models, we have to figure out when we need to retrain them based on, you know, data changes over time. Payers are paying at different times or they’re denying different types of claims, that sort of thing. So it’s something that has to be continually evaluated, and people have to understand that difference.

It’s new. It’s something very new, and it takes people a while to understand that this answer may not be correct. You know, it’s doing its best, and it’s supposed to help you along, but we still need the human to be there to agree or not agree. For example, prior authorization.

We generate an answer based on the payer policy, and people have—it’s their own free will—whether they go and get that prior auth or not. And we track whether they did or not and whether they were right or not based on the denial of the claim.

Right.

Chapter
Key Areas of Focus for AI in Healthcare

Pinaki, I might lean on you here. You know, there are these four areas that I think are very important. This is a great LinkedIn article, by the way. I put the link down there below. As Michael mentioned, GenAI being relatively new—I’m using air quotes—this gentleman, Ben Carroll, did a great job on the first slide and I think the second slide in terms of the text and content.

But, you know, GenAI in terms of coding, billing, denials, patient financial experience, revenue cycle optimization, or platform—is there a favorite in there, Pinaki, that you would probably double-click on or go in? Or should you try to inflate all four tires on the car at the same time so you don’t go in circles? Or what are your thoughts on these areas?

I think, given the fact that you had mentioned something around usability, timing, and aspects to another slide that you had, that will be really dependent on how people utilize some of it. Personally, I focus on patient financial experience. Yeah. There’s a lot of work to be done, and there are a bunch of different companies who are utilizing AI to come back and present it to us from a provider network perspective.

So it’s gonna be a buy-versus-build conversation if you’re a provider network. That’s the first thing that I would say. The second thing that I would go into is really this predictive analytics for denials management.

Yeah. As a provider network, we have always had denials ever since we existed from a general perspective.

And it’s gonna be there. Utilizing GenAI and its power from a perspective of utilizing predictive analytics is gonna be key—about how we look at it from a service department that the denials are coming in, the CARC codes, obviously, that exist as we have these conversations with our peer groups as well.

I would say, I think I saw a question since I have the floor for answering it. I’m sure we’re running out of time. That’s alright. The case studies that we have in AI for RCM are gonna be really dependent on who you’re talking to. Right?

Yeah.

So most of these provider networks that I’m talking to, most of my brethren from a fraternity of other facilities, we are looking at it and going, are we utilizing it ourselves? Are we gonna build it ourselves? Are we gonna utilize it from somebody else building it? So that’s a constant result there. As far as we are concerned, we have obviously held on the case that we talked about, and it’s already in play.

Chapter
Adapting AI Strategies in Provider Networks

We have to sort of see how much of it we are utilizing from a general perspective of doing it ourselves versus purchasing it from others. And as more provider networks come into play, it will depend on exactly who the provider network is, whether they wanna build it and be more technology-driven or purchase it after it’s completely vetted out.

That’s a great answer. Yeah. I think you’re gonna hear this buy-build discussion happening more, whether you, you know, partner. I’ve heard it’s very rare it would be one or the other, Pinaki, I think, to where you’re all buy or all build.

I’ve seen mostly hybrid environments. You’ve got someone getting the bots to you or the large language models to you from an infrastructure standpoint, but you might have your own in-house engineers that monitor that on a go-forward basis. I was just at the revenue cycle conference in San Antonio. I’ll forward you the deck and anyone else that would like to see it.

It was an excellent talk about AI and revenue cycle and how you deploy it and whether you buy or build. And as Michael mentioned, it’s an expensive tool. And as it should be—it’s saving you millions, if not hundreds of millions of dollars over time. So it’s a wise investment, but I think you need expertise and you need internal controls.

But having too much of one or the other, you can start to get into a little bit of trouble as you go forward.

Chapter
Transitioning to Q&A Session

I think, in the interest of time, we'll probably just jump to questions if that's okay. And, Claire, if you're still out there, Pinaki, Michael, and I are eager to answer whatever questions you might have in the remaining eleven minutes or so. Michael and Pinaki, thank you so much for your insights. But have at us—what questions do you have for us?

Yes, thank you, John, Michael, and Pinaki, for a wonderful discussion. I'm going to go ahead and hop into some questions in the Q&A box. Audience, as a reminder, feel free to submit your questions via the box you see on your screen. Let's go ahead and get started with our first audience question.

Someone would like to know what opportunities are available for those interested in AI governance.

I'll start with what I heard at the conference, and then, Pinaki, you probably have other governance committees at Prime, I imagine, that have representatives.

One of our FinThrive advisory board members, from a very large organization in the Southeast, talked about his AI governance, and he's a CFO. And, Pinaki, you were in the room when Mike was talking, but he mentioned having the Chief Technological Officer and the CIO there first and foremost. I mean, this is a bright, shiny object—a toy, if you will—that is birthed or genesis to the organization. He also had the CMO there from a patient safety and medical care standpoint.

The CFO, obviously, is writing the checks, he or she. Then they had department heads involved, and they came up with policies and procedures. You don't want so much administrative burden out there to manage it, but you want to make sure voices are heard and risks are assessed in a way where you understand, for example, when a hallucination is present. How much of this is patient-facing or not, I think, is a good lens to look through.

How much of it are you going to allow the end user to control versus oversight by a clinician or an extender when you're in a care area? On the revenue cycle side, those might be dollar thresholds. If there are million-dollar-plus claims, is that something you want AI to handle entirely, or should there be more appellate rights that the system can't see? But, Pinaki, do you have thoughts on governance? I think that's an excellent question.

I think it's a good question, and I think it's going to evolve as we move forward. But, yes, for sure, there are certain parties who absolutely need a seat at the governance committee table.

Technology leaders like the CIO, CTO, CDO, and CISOs would obviously be involved, given the delicate environment. Plus, at the same time, the CMOs and the Chief Experience Officer could be involved if you're doing things around patient-facing aspects. The Chief Financial Officers would obviously be involved from the perspective of how the revenue cycle is impacted or how all of it ties together.

Great. I think, Claire, we can go to the next one.

Chapter
Addressing Risks Associated with AI

Yeah. Our next audience question is wondering if you could address some of the risks of AI as well.

Pinaki, do you want to start with that? I think you probably had some good points you were discussing with Prime in terms of dipping your toe in. What are some risks you see?

To me, anything you don't know is like walking into a dark room and trying to figure out how big the elephant is and which part of the elephant you're touching. Let's just be honest about it, right? I don't profess to be someone who's well-versed and knows AI inside and out. But here's the reality of it:

There are going to be risks involved, both personally and from a general perspective. We would stay away from anything that directly impacts patient care. However, there are aspects that are easier to implement without impacting patients, such as on the revenue cycle side—recognition of information and collation of it. Then it extends to information about communication with peers. That's how I would break it down. Anything with a larger impact would fall into a leadership bucket.

Just to share, and I don't think the question is directly related to the risk aspect, but I actively advise my team to use any AI tools available. They're using Copilot right now to conserve time on transcription-related outputs, minutes of meetings, and follow-throughs.

So, it's not a one-size-fits-all approach. Start with the smallest item, utilize it, and then move on to the next use case, gradually increasing usage.

Michael, do you have thoughts on risk? As an expert in the space, I think starting small is probably a good idea. You don't want to boil the ocean and create uncontrolled chaos from an AI standpoint. But are there other risks folks should consider? I think that's a slide I'll include in the next deck we do on AI risks. What are your thoughts

Chapter
The Importance of Human Oversight in AI

Yeah, that's really what I was saying before about GenAI and hallucinations.

Even with human review, AI can still increase efficiency. You don't have to manually enter data or sift through multiple documents to come up with a proposed solution or suggestion. Most of the time, in our use cases, AI is accurate, so you can quickly review it. However, a human should always be there for anything that could lead to trouble.

For example, it's okay to interpret a contract and test pricing before releasing it. But when answering customer questions or writing denial appeal letters, human review is essential.

Sounds great. Claire, we'll take the next one, I think.

Yeah. Someone else would like to know how much effort it will take to interface AI with payers and reimbursement.

I'll start with that one, Pinaki, and then you can agree or disagree with me.

I think it depends on how well you're connected today. For example, I know Epic is working with payer platforms. FinThrive has deep analytics in terms of payer connections and integration with Assurance Discover. Michael showed some workflows we're innovating on for the future.

And trust me, there's a long list of other things behind that. I think it depends on how well you're connected today. You also don't want payers to connect on everything.

I've heard horror stories, as Pinaki mentioned, about giving payers complete access to medical records for their members. They can reverse-engineer logic for denials, utilization management, or medical necessity based on care utilization trends. So, less is more. Focus on areas like eligibility, payment, denials, underpayments, medical necessity determinations, and prior authorizations. These are the "tennis balls" constantly being volleyed back and forth.

Whoever asked that question, feel free to email me. We'll put our emails up in a minute. There are about seventeen pieces of data that both the provider and payer touch once for the same patient during every encounter. Seventeen times—it was measured.

That's in my book as well. These include eligibility, insurance plan, policy number, discharge date, CPT-4 codes, HCPCS, and so on. Pinaki, I don't know if you have deep integration with payers at Prime, but I would start with where you're at today. It will be difficult if you have nothing. Be cautious about what you're connecting and focus on the areas I mentioned. Do you have anything to add?I

think it's a broad question. I would just add to your comments. It really depends on who's connecting and their appetite for connection and investment. It's a build-versus-buy discussion.

You might want to break it into smaller pieces. It's broader and requires more investment and time to create something dependent on someone else. Breaking it into smaller chunks and working with it is probably the best approach. That's my personal opinion. It also depends on the information you already have coming in through your clearinghouse and how you're leveraging it before moving into the AI space.

Chapter
Navigating Claim Denials in Healthcare

Yes. I believe we have time for one more question. Someone in our audience says, "We receive so many claim denials from our Medicare Advantage insurance companies. It's simply hard for our staff to address the incredible volume of denied claims. Can you please comment on the situation?"

I'll start with that one. I get this a lot. I know Pinaki probably does too. Here's the deal, guys: denials are a four-letter word now.

At every conference I attend, I hear "denials" as much as I hear "AI." And I'm starting to roll my eyes. I'll get on my soapbox for a minute.No

matter how much you complain, denials are part of doing business with a payer. Their job is to manage care. They will inspect every claim and charge and pay according to the contract you signed. Payers are not inappropriately denying claims for the most part. They're denying based on the information provided at the time. Overturn rates are typically 80-85%, usually because the information wasn't there, the payer couldn't see it, or there was a disagreement about medical necessity.

The only way to reduce denials is through the contract. Use analytics and AI to understand which provisions in the contract are causing denials. For example, definitions of medical necessity, inpatient care, or mutually exclusive language. Adjusting provisions like authorization timeframes can make a big difference.

Pinaki, do you have anything to add?

I think it's a broad question. I would just add that it's about recognizing payer trends. For example, a bulk appeals solution could help send touchless appeals, ensuring claims are paid. It's about recognizing payer behavior and finding the best strategy.

You bet. Well, Pinaki, Michael, and I want to keep the conversation going. This is part of a series, so there will be opportunities to participate in part two of the AI discussion on denials. I'm glad that question came up last.

It's a good segue. There's a QR code here where you can engage with us. Pinaki and Michael were gracious enough to share their emails. Mine is there as well—watch out for the two i's and a k in my name. But you're welcome to email any one of us—or all three of us—if you have additional questions. We'll review those questions after this session is over.

Chapter
Closing Remarks and Future Engagements

I want to thank you all for joining us for this hour today to discuss AI of today and tomorrow. Pinaki and Michael, thank you both so much for sharing your time and insights with me. Your contributions were invaluable.

To everyone else, have a great day, and thank you so much for being part of this discussion.

Thanks again!

Yes, I’d also like to thank John, Michael, and Pinaki one more time for such an excellent presentation. And a big thank you to FinThrive for sponsoring today’s webinar. Thank you all for joining us, and we hope everyone has a wonderful rest of their day!