PegaWorld 2026: Applied Agentic AI: How enGen is Orchestrating Next-Gen Care for Payors

Well, we're about to go through is a pretty interesting application of AI in a Healthcare and one of the biggest areas in health care, I think, which is around care management. I think this is where the real future is in all these things. And one of our, I think leading people doing this is happens to be engine and what we've been doing there for years, um, in this particular area, I believe is leading some of the, the path that a lot of us are going to be taking, especially in how we use the genetic AI, how we apply AI. And you know what I call an applied agentic AI, which is more than just thinking of an agent that does magic stuff all over the place, but it's more about how do we use agents within the workflows, with how we apply it in a repeatable, predictable manner?

You've heard those terms, but it's really great when you start seeing it. That rubber hit the road. And I think that's what we're about to go through here. So without further ado, I want to introduce Rick Rakowski from enGen, who's actually leading this project. So unlike me, who's just the kind of emcee kind of character up here, this is really going to get some good insight as to what they're doing, what their plans are. And with that, Rick, I'm going to turn it over to you to get started. And then we'll go from there. All right. I have the first slide. You don't know. I'll take that. But thanks Robert. Appreciate that. I think I'm on the first slide but I don't see it. So it's on this screen but not these screens. How do we switch it. There you go. Okay. Thank you very much again.

Thanks, Robert. Thanks, everyone. It's been a great couple of days. So I think we're the only thing standing between watching chess. Master play chess. Chess, I believe, is. Yeah, right. Okay, so my name is Rick Krakowski. I am the clinical product director at engine. Engine is a software tech company. We're a wholly owned subsidiary of Highmark Health, which is the third largest not for profit Blue's plan in the United States. And we've been working together with Pega for years. And I'm going to talk a little bit about our journey and how we're using Pega to transform how we fundamentally administer care management and to do that, our product that we've created and we've been working on for about five years now is predictable. And that's what we're going to consider our intelligent care orchestration layer. So we're going to talk a little bit about our transformation.

Okay. Before I do that, we're going to talk here. I'm just going to cover off on what I call the five storms that are converging. Okay. This is what health care is experiencing. And this is what's forcing us to have to rethink how we manage and deliver care. Demographic shifts. From a clinical perspective, we're seeing aging populations and a rise in chronic illness. Chronic illness is going to require some deep, complex longitudinal care. Okay. To do something like that, you need to have tremendous amounts of data and a really well put together workflow or a series of workflows, which we like to call interoperability. Workforce shortages. We've got clinician burnout. There's not enough of them. And I think what we're finding as as we have the rise of chronic illness, we have CMS ratcheting down SLAs and utilization management space is that some of the systems we had some time ago had too much manual effort in it.

You don't want your clinician like avoiding their clinical acumen because they're busy swivel chairing or, you know, trying to understand what's going on and doing admin knitting processes together themselves. Rising costs. That's not a surprise to anybody. Providers, members, they're all feeling the pinch. And health plans are right in the middle. Regulatory pressures I mentioned CMS and the SLA is being ratcheted down. Also, I think right now with the interoperability play CMS CMS0057. F is really a challenge for a lot of plans, and I think if plans haven't started or they're struggling now, it's going to be very difficult to meet that mandate and then customer expectations. All ages now use their phone. Everybody it seems can order from Amazon or Instacart. And a lot of them use it for banking as well. So these are all coming together in an environment. And this is where we feel like predictable has been positioned to manage in our transformation.

So we moved from a traditional care management can't keep up approach, like with what's going on today, fragmented data across systems. You can't get a whole picture of your member. I mentioned the manual workflows, the context switching, or I'm sorry, the moving between systems. And then there's absolutely not a lot of there was not a lot of insight from data AI, Agentic or Other. That would be between member touchpoints. That connected journey was not very connected, Um, but it wasn't the people. As you can see, people really want to try hard. They really want to help. It was more of the architecture. And that's where we're going to talk about what we've done with predictable. Okay. So predictable predominantly is made up of utilization management and case and disease management. On the prior auth side, anybody that's had any experience with that, it's been cumbersome for years.

Um, one of the main reasons CMS interoperability is out there is to make a change, reduce that burden on providers. Um, what we seek to do is bring together unified data like clinical data and medical policy, and then use something like AI to render a decision or a recommendation to a medical director or a nurse, help them along the process. Today. That's just, that wasn't something that was possible five years ago and wasn't happening. But it doesn't stop there. That's just inside. Um, that's just being able to render a decision. What if the type of authorization it was would dictate maybe an outreach and case management starting it there. It also has to flow downstream and the claim needs to pay automatically. What if the claim fails because there wasn't an off submitted? Should it start? A retrospective and predictable answer to that is yes.

You need to orchestrate across systems. So that's important. So we're trying to unify data interviews, intelligence interoperability into the workflow across the entire care journey. Okay. This wasn't, I don't want to say this is like a, just a, I guess an engine problem or predictable problem. Predictable was the solution to the, the legacy system. This is industry wide. You can't be transactional anymore. You can't have fragmented data. AI is just not going to support that. And you're not going to be able to support your clinicians in the workflow. So what we did was we aligned on four strategies that we felt need to be pushed and Shifted for us to actually make a difference and become transformational in the care space. Transactional orchestras have already said that, but that's how you're using data across the entire ecosystem to drive actions. Multiple actions. Sounds a lot like Agentic.

AI could be used there, right? Rigid to modular plug and play. You have to be compliant. Fire compliant doesn't mean everything needs to be fire. Fire is not really that easy to handle in my estimation. But you need to bring all the data that you can into the process. Manual to AI powered OK. We're talking about using AI, traditional and AI agentic to force actions to inform clinical decisions, to inform care decisions, eliminate eliminate steps you don't want a clinician doing in a workflow automation and task intelligence like that. And then you want to move to a consumer centric. We talked a little bit about what's that care journey. How do you tailor a care plan to them? How do you get to know your member better than they know themselves, so you can help them in that care journey. All of this is what's going to help combat those storms.

I mentioned earlier, driving down costs, doing more with less. We need to be able to scale. The traditional way of doing things doesn't scale. You need to have an orchestrator. Platform. Okay, so we had we call them our five Platform strengths. This is what we did to address, um, the strategies that are listed across the bottom here. Um, and this having a diverse data connectivity, we talked a little bit about that, but that means you need to interact with EMR, EHRs. It could be coming through a member portal. Any portal could be a file lab data. It could be coming from connected devices, which is usually unstructured, right? You need to assemble this and have this data on the ready across your entire care ecosystem. You need to be able to look at that whole person and understand what you're dealing with. External integrations. Again, plug and play.

That's not just with different vendors or delegates. You need to get to your clinical program vendors. You need to get to AI models because it's not always a one size fits all, and you need to be able to get all the potential different types of data. Like we said, interoperability, you have to be, um, bringing workflows together, okay? You have to support a better experience, a seamless experience for the member and clinicians, AI powered, we're talking about using that to drive down costs, bring up quality, eliminate waste on the, um, side and just a better journey for the case manager. And we're going to talk a lot about that with the example I gave you at the end of what we're doing. And lastly, clinician centric configurable workflows because not every client we have is the same populations vary. What you need to do, how you manage them vary.

And you got to make sure that's configurable and not requiring code. We also offer a bypass to go along with what we have from a system perspective. So we can run not only the system side, but the operational side as well. So when you bring all this together, again, it's an intentional push to unify data right into the workflow, give that intelligence, that insight to help your clinicians driving down admin and bringing up the experience for the member. Okay. So the so what of all this? Across the personas. The so what of this? Is the member the patient themselves across prior authorizations. Now you're talking about giving them better access for care, right? Historically, again, we talk about how tough that prior auth process can be. You can't impede them getting a service because the system can't handle it because of the constraints. So with what we're doing, this is going to improve that.

They're going to have access to care and it's going to be timely. On the case management side, it's about outreach and coordinating lesser handoffs, a connected experience, helping them along their path, not asking the same question three times if they call customer service and talk to somebody there versus a clinician from a payer perspective, you're helping them make better decisions, more informed decisions, taking the admin out of the process right along the workflow. And then from a provider clinician experience, you're allowing them to do what they do best and focus on their patient. Our member. We can't we can't have them buried in paperwork trying to figure out how they're jumping out of an EMR to go over to a portal to submit an authorization for somebody to only find out they didn't give the right information. And you go back and forth and things like that.

So there's a lot of value there in bringing all this together. So when you look across the bottom, you've got the data lake and the data information on the right, you've got all the ecosystem and other systems on the left here with predictable in the a minute. It's orchestrating that care across the whole continuum of care. This is how you're going to capture some of those gaps. Eliminate gaps in care, right? That insight served up, and it's available for the clinician to work with the member. Okay. All right. So I've mentioned AI as we've gone through this. So I'm just going to review some of the areas that we've, um, where we've implemented because it's anywhere from ideation to scale and production. So I'm going to cover that a little bit. Now that we've talked about what our transformation was because you needed to get, you know, transform the data and the workflows to really make this useful.

Okay. So AI driven care plan and engagement. This is everything to do with case management. So we've come up with a process where we're going to help generate care plans. Today. We do it based off of an assessment, a primary assessment. That's a mandated type of action you have to do when what you're doing, care planning, and that's going to generate a care plan. I'll show you that later, but we'll show you how AI is informing that care plan to be even more personalized and tailored to the member care plan. Pre population is in ideation. That's actually going to be used with our ambient listening tool, which we'll talk about. That's the third bullet there. I don't want to give too much away because I'm going to be that as part of the example I'm going to give later. But we want to use that to pre-populate the care plan and any other data we have.

That's going to be the personalization aspect of things. But don't forget, it drives down the admin. Two on the clinical insight and decision support. That's all about every piece of data we know about the member having a full summary of information about them and being able to, as a, um, clinician or a case clinician, being able to search on them, ask questions, kind of like a rag model, be able to go through that and chat with it to understand and prepare or make better decisions. The AI assisted utilization management. Take a minute and talk about these two OK. Both of these, the um quality agent and then the recommendation support are both done at the time a nurse or medical director actually render determination. So in an authorization process, they're going to select approve or deny after looking at all the clinical data, right. What the um quality agent does is at that point in time when they select approve or deny, it's going to gather up all the information on the authorization, any other clinical information that might need demographic data.

And it's going to understand what was the sentiment of the information entered by the medical director. And does it align with what was selected? Sounds silly. There's been mistakes made where people select the wrong, the choice they meant to deny, they hit approve, things like that. But it goes further than just that accident it's talking about. Did they write the right rationale? Is everything there? Then it's going to validate the two the tool against what the nurse or medical director said. And if it mismatches, it's going to go for secondary review automatically in the workflow. And you have somebody else look at it. At that point, you're either going to retrain the tool because in the beginning, how do we know which one's right? So you have to do some of that research. You know, this isn't like fast. It doesn't just happen overnight, but this actually is in production and it is working.

And we're starting to scale that, that process. So that's, that's brought a lot of quality. It's giving you more accuracy in your decision making. Now the second bullet there, the, um, authorization recommendation is more of what I was talking about before, or at least alluded to where we're taking all the clinical information, all the data around the authorization, comparing it with AI to the medical policy, and rendering a potential decision to the medical director. In this case, it's running in parallel, doing it independent from the medical director, and then we compare them. If they mismatch, we move it to the to the secondary queue for review. Now in both cases, all the information on the decision made by the AI model or agent is shared with the recommendation. So it's not like, hey, we approve and you don't know why. So it's very easy to do the research, but this one can be very dangerous if you don't have human on the loop when you have, um, recommendations going out the door.

It's heavily regulated in healthcare, right? So we have to be very careful about what we do. Not to mention remember, might not get the access to care that they deserve. Excuse me, but a good example of how we can implement this in a responsible way, because we'll talk about AI governance in a minute, is you have these two running side by side, but once you have that where you want it and the accuracy of the agent is to where it needs to be, you can now slip that in front of the medical director. No longer do they have to do all the legwork to get that information together. They have all that information. They can confirm it, make a decision. You can continue to train the model if need be, and at some point it may actually hit a success rate high enough that you wouldn't even stop it.

But you have to be pretty sure about that. And I would never do that. Ever do that for a denial. It would only be default to approve. So you maybe give up a percentage of approvals that may not, um, maybe wouldn't have happened if you had a provider, but then you also have to take a look at what's the rate of appeal on that. If it's that high, you know, often it's approved. What's the appeal rate and what's the cost? Is it even worth stopping anymore? So once you have that reliability, then you can take the human out of the process altogether. So you can see the value in that, right? As we have to do more and more and more taking that work off their plate. And then AI governance and performance is absolutely it's across all of this. Like I said, you have to have metrics, excuse me, you have to have metrics.

And we actually have a dashboard that has to go out to, and you have to show the progress you're making. It has to be very transparent, very auditable. In every case, we have carte blanche to use Pega fine set of agents and AI in the sandbox to do anything we want, but as soon as you want a business case number one, you got to work with your operational team, right? They're leaders because you want to build that trust. You have to find a very good use case to use and then show ways to prove it and pilot it. And again, human in the loop is a good way to do that. But then all the tracking has to be there from a governance perspective. All right. So. We made two big bets as we were going through this. And it kind of evolved. I want to say back in 19, we were thinking about AI as as much as we are now.

Um, but AI and clinical data integration and ingestion. So the bottom left quadrant there, that's kind of everything we talked about bringing all that data in, right? so it can help with automation, decision support, tailoring care plans and so on. That's great. I'm not going to rehash much more on that, but that that was not something that happens in 4 or 6 months. This is over time. You have to have a data repository, data lake to manage this. Bumping back up, I already talked about the auth quality AI. I won't rehash that some more. And then on the care management side, the care planning, we're going to go to that example in just a minute. But we have four different ways where in the example I'm going to give you, there's four different AI interventions happening to make that entire end to end journey possible.

All right. So on the right side of screen, I was going to let you talk a little bit about the suite of products. I don't know how you sat here this long without saying anything. Um. I could have criticized me a little bit. It's been very difficult. I was going to let you talk a little bit about that. I mean, we recently went to the Cloud OK to gain access to all the AI agents that were now being able. It's accelerating our build. Okay. It's right in the workflow. It's been a really good experience for us. And we've actually built and piloted things very, very quickly. And then Constellation, we're still talking about that. We're planning. We're discussing Constellation. But aside from improving the workflow and experience, Constellation is going to also have a, I'm going to think, a big impact on the users, whether it's clinician on the utilization management side or case and disease management side.

So we're talking about that and we're excited to look into that option. But across the. I was going to let you talk about the tools. Yeah. So what we're talking about here, you've heard all week here at PegaWorld is pretty much where our platform is going and how we're utilizing these, these components that we've built over the decades. Of course, now, so what Rick's talking about is basically a series of solutions from the core of this one, which is the platform, the Pega Platform, the operating environment, and then customer service for healthcare and care management are basically the two pieces that are doing this primarily. Um, so these are the pieces that have all of these genetic capabilities built within it. Um, you know, the more and more we use these things. So they're taking advantage of that. The belief is here, and one of the beauties of this particular presentation to me is the fact that this is really an applied way.

This is the way you start with a genetic AI. You don't start with the bigger agents and start working down. You start building in these spots where the agents start really adding value immediately. And as the workflows evolve, these agents are available to do it. So it starts out with the care management components. It also underneath brings in other components as well, such as Process AI, CDH, which Rick showed is with their Myhub solution is what they have. That's basically CDH that interacts in this world that we could talk about later, but it really does help tie this whole orchestrated layer together because in care management, you can't separate the patient member from the provider, from the, the plan. That's just not the way the world works. So what we're doing here is showing how you start building that future from an active care management environment.

In this case, utilization management also falls under care management space as well. They're sort of tied at the hip. And so these are the main components that we're talking about from the Pega perspective. Our platform and all of its components, along with customer service, care management and CDH are the main pieces. We also have other solutions out there as well that aren't directly applicable to this particular use case, but it also expands it out because guess what? This flows into claims, this flows into appeals and grievances. This flows over the overall arching thing. So this is one piece of that journey. But I wanted to point out that these are components that we use. They're not custom built. You know what Rick and team are doing are building the intellectual property on top of these basic components. Basically what we're doing here. So a good case of how this is being done.

And so as Rick said, most of you are building these hyperscaler type data components. You know, whether we call them Patient three 60s or, you know, in this case, that's a member listening service. They call that piece of it. And so they're bringing all these things together so they can analyze data, they can look at that, and then they can use predictable myhub and Other Pega components to orchestrate the experience and actions from that data insights that they're gaining. So this is the way the whole system start working together, this bigger architect sort of situation that we believe, hopefully, that this is the way a genetic AI in these sort of use cases are going to be done. In fact, I was watching television this morning because I'm on the East Coast time and I get up early. And I had heard an expert that was talking about AI, and he's been talking about this for two years, about he was kind of negative against it.

And now he's realizing that, well, you know what? I think we had this all wrong. AI for personal use is just incredibly excellent. But we got a long way to go on the business side. And I was sort of hollering at the television, no, we're there, but you don't know about it yet. So I think this is a good indicator of where we're going, what we're doing. And thanks for the opportunity to talk. I appreciate it. As you were. So good as you were. Yeah. So, I mean, you can see data is the lifeblood of all this. The AI doesn't work without data. So genetic or otherwise. So all right. So this is the example. This is real world example. This is what we're working on. Now some of this stuff as I said is piloted and not necessarily completely in production. But the transformation we made is what's allowing this connected care journey for our friend Steve and pretend our member.

And in this particular case, Steven has a cardiac event that lands him in the hospital. And we're going to talk about one piece of data that's going to trigger this entire thing. And that's going to be a discharge that's entered into the EMR. So hopefully as I go through this, you can see all the connectivity, all the interoperability, the AI and everything we've talked about with data kind of flowing through here. And initially even in this first event where that EMR passes through in near real time, we're going to find out that Steven's being discharged, and that's going to set off 3 or 4 events, first of which is it's going to update the authorization with the discharge date. It's going to set a transient transition of care outreach in the case management application, automatically assigning it to a case manager who's most skilled and has room on their calendar in the first 24 to 48 hours.

Because in case management engagements, everything, if you can't get to that individual in the first 24 to 48 hours, it tends to drop off in connectivity, right? But why stop there? We know that Steven likes to get messages on his phone through his My Health Plan app, and we'll send a nudge. Not right away because he's probably still in the hospital, but we're going to let him know, hey, look, for us, we're going to be reaching out. We want to help you. Here's how. So that's four different things that happen in different systems. There's a portal or a handheld app. I'm sorry, your website, your health plan app. Um, we also started in case management, utilization management, all this stuff working together just from one piece of data. So Tom is the case manager who comes into work and picks this up and he immediately opens it up because the urgency is high enough that, um, it bumps it up the list because again, the 24 to 48 hours, he validates all the information, the discharge, what the discharge reason is.

Um, information on the authorization is a single click to look at that. And then on the right side, there's not like middle, right, middle, there's non-clinical information where allowed to reach out and all that stuff. That's great. The one thing I forgot to mention, that data, that data trigger up front, also kicked off a full summary of everything that's going on with Stephen. So we know everything we need to know about Stephen and insights that are generated from that gaps in care. That's in the bottom right corner there. And that's what Tom's going to do to research before he does the outreach or the phone call to Stephen. Tom's going to look at that and understand everything he needs to do now. Predictable wasn't didn't have data all over the place, but in predictor member 360 view is about 10 to 12 tab pages that he that Tom would have to, you know, work through and bring the data together, make sure he understands what's going on.

So it was a little bit of a time drain. It's not, you know, not nothing, but it's, it's definitely something when you can have it all in front of you. And then as he goes through the process, he's doing his research, asking questions. He wants to know what's going on with types of medications. Is Stephen taking. And then it gives that information so he can this is very simple example, but he now is armed with all the data. He needs to understand what the discussion is going to be like when he calls or outreaches to Stephen. Now in the future, we're looking at doing outreach identically to. But in this particular case, we're not going to go there. So this is really the first piece of AI that's being used here, the summarization process. And we find out a few things about Stephen going in that he's got um, just he's got, he's armed with information to talk to Stephen now.

So he starts the outbound call. So he calls Stephen armed with all the information and he's going to start the call screening here. He does the typical validation of who Stephen is, make sure he's talking to the right person. And he goes through the normal critical information. You know, you're being recorded HIPAA, everybody loves HIPAA, and then we're using a genetic or ambient listening during this call to enhance the experience. So what ambient listening is going to do? It's going to allow Tom, as a clinician to not have to document any information. Right. It's going to be just a conversation focused with Steven on all the information he's armed with. I'm not going to have to write anything down. This tool is also going to serve up real time insights based on how the discussions going. So a quick example would be usually in the screening you try to get to know the individual, especially if it's the first screening.

So he's talking to Steven and Steven lets them know that, you know, he's happy, sad, he's feeling a little bit better, but he's glad he's home. His granddaughter is there because her mom had to take a job out of town, and she wanted to finish high school, so she's staying with him. But you get this idea and understanding that, you know, I like when we go out to lunch every two weeks, it's right after my doctor's appointment or whatever it might be. The tool picks up on the fact that she's probably going to leave after high school. It's getting close. Maybe we should reach out. Maybe you should ask, what's his capability to drive right? Or maybe we might have to do something to help them. So it starts looking at these barriers to care, like transportation. Or it might even be a food barrier too.

You don't know what else is going on, but it's prompting the clinician to actually ask the right questions. Once you get through all that, it records all that and it's going to bring that into a very nice summary with a lot of detail and a lot of suggestions. We'll get to that in a minute. But the second half of this is you then have to go through that primary assessment, which is long. Thus, the reason we want to now pre-populate and verify versus answer every every one of these questions. Right. But in the end, it's going to bear out that Stephen may have a little depression. Um, and that he's, um, coronary artery disease, he gets migraines, hypertension, things like that. Not that I'm going down the path of the actual health part of this, but I want you to understand that's what's going on here.

So the ambient listening tool then generates that summary with detail on sdoh detail on self care that you can monitor in the next call. It's going to give you a list of things that you should follow up with on the next call. And then when this call is done and they hung up, you have all that information. And at that point, again, clinician in the loop. I like to say Tom gets the last say, right. Does he agree with everything? It's there. He can change it. You can't delete it. Right. There's an audit trail for everything. But again, the human helps to understand that the ambient listening did a great job, but we have to make sure it's accurate. All right. So after we hung up and we're done with this, the very. That's the second piece of AI technology. The third and fourth are both going to surround the care plan itself.

So what you see here is coming out of the call, answering the assessments. You're going to see four problems here on this particular care plan. The first two are actually coming from the assessment itself the answers. And it's driven that way. The last two, you can see the little purple AI star over there are generated by our AI tool. So we're using a genetic AI to go back and look at historically and whatever information we have out there for like cohorts, similar medical conditions, what's worked in the past, where do we have success on past care plans? And it's bringing in those problems to this care plan. So we're trying to inform it, get more information out there. That's just opening up the care plan. The the second piece of AI we use is we just had that full phone call, and we have all the information from the ambient tool.

We're not going to take that and validate it against all these problems that are out there. And it's going to let us know if the problem. There's alignment to all the discussion and all the information we have. If it's partial, full, or not at all, and where are the gaps, we might bring in other problems. And in the end, at the end of the day, Tom is the one that's going to finalize this care plan. And we feel like this with these two pieces of AI are getting us to a really, really personalized and tailored plan that's going to help Steven, you know, get the outcomes that he wants. We got to keep him out of the E.R. and keep him healthy. So I'm sorry, I'm backing up a little bit because I didn't have the last picture, and I thought it might be good to show this survey on the bottom or all the way on the right where Tom's at.

Um, that's where the care plan is getting finalized and it drops down to the bottom right. I don't know. I can't find the little pointer here. Um, basically what's happening now is that care plan is being uploaded into a fire resource, which I think in Infinity 26, I think is an automatic type of thing you can use, but it's up there so it can be downloaded and pulled down through the EMR so we can get Doctor Martinez, his PCP, on board with the multidisciplinary care team at the health plan. We want to all be rowing in the same direction. This has been historically difficult to do. Um now, the health plan we work with in Pittsburgh actually owns a hospital system too. So you wouldn't think it would be as hard. But surprisingly, there's some struggles there. But this is going to allow Doctor Martinez to order the services and everything that Steven's going to need because he's coming home from the hospital.

It can also help with things if you need. If it was one of those scenarios where you needed DME ramps, anything in the house for home health, that stuff can all be done in advance and activated relatively quickly. So this is where, to me, this kind of brings the whole journey together. This all started with one event trigger that pushed it all the way through this process. And really the only stopping point is Tom as human in the loop because you want to do that outreach. Now, does that get replaced with a genetic outreach? Because I know people like to talk to, they seem to like to talk to agents more than people sometimes maybe. But in the end, I think you've reduced All the admin. You've taken a lot of the admin away. You've infused data and you can see the interoperability through the whole through the whole process.

And I think this is a massive step in the right direction. It's not done. There's always more to do. But this is exactly the journey we set out to get to make back in 2019 when we just started talking about the strategy. So I think that's it for me. Well that's great. Well. Thanks so much for that. Any questions right now? Hopefully what you're seeing here is really sort of the the beginning work in the Agentic future that we all are facing here and how we're going to go about doing that. It's probably not going to be a giant LLM in the sky that suddenly says, we're going to take over clinical care. It's just not possible to do that. It's going to be a much more gradual set of stepping stones to get there. But the main thing is, I think is going to be the speed because we're showing one example, you know, around a, you know, a transition of care sort of situation, utilization management in front of that.

There are thousands of these sorts of use cases just within the utilization management area alone that we're having to map out. So with this sort of infrastructure in play, expanding the business of health care into these other domains of clinical care is going to be really, absolutely necessary in order for us to achieve some of those future goals. So here we're talking about saving time, saving effort, improving quality that way. The other thing that's happening behind this, though, is some of these efforts are basically solving problems before they escalate into something bigger, especially when you're able to get an authorization turned around in time. You don't have that treatment delay aspects that sometimes factor in to how care is being delivered. You don't have some of the the regulatory problems of not doing the things in the right order, their SLAs and all sorts of things around this particular area that you've got to be on top of.

And so what the system does is now takes that out of the The responsibility of the human and gives the AI the chance to sort of inform the human as to what's going on at that point in time. So they can either take the action or the action has been encoded in a business rule somewhere. So now that they can actually execute it, business rule in some ways is a human in the middle, by the way, it's just the what the human said. We have a question. Yes. Can you hear me okay. Can. Yes. Great. Um, I'm just interested in your thoughts about, um, not just using AI and analytics for like determining the best care for the beneficiary, but also monitoring cost effectiveness. Like, is this the most cost effective treatment long term? Are there other options like with preventative care? So yeah, just curious about that.

You want to answer it. You want me to go for it? We're not doing that at the moment. But I mean I think we've talked. Sorry. No go ahead. We we have talked about um that in terms only to the degree that like we would Analyze. Is it worth doing the care? Is it worth like finishing an authorization? Because we know the cost of us going down a path where it might get appeal, or it gets high rate of appeal, it's going to cost us more. So it's not worth us actually even asking a medical director to look at this, we'll just approve that 3% difference or whatever it is. So from that standpoint, we've done a little bit of research, but I don't know if there's anything else going on in the industry that you're aware of. Well, I think some of this actually happens in different places in your infrastructure.

So I think this MLS handled some of that right. When it starts looking at criteria and where things stratification, it might balance some things. But for the most part it's factored in, um, usually, um, in design time, for sure. It's factored in because you start thinking of those things. Where's the best site of care when they start doing that utilization level. That's really more care management at that level. But that's what's happening. Absolutely. It needs to be there. I mean, it's probably the number one decision you've got to make availability being the first, obviously, but then what is the cost side of care becomes a huge issue in this particular area. Um, also in the case in authorizations it happens as well. I mean medical authorizations, you know, pharmaceutical authorizations are definitely that way. Um, and then this whole step therapy concepts that drive some of those things.

So you want to make sure that all those in order. So it does the financial piece, not just as a projection of financial value or return on investment. It's more about using it within the decision to understand the best choice. Um, obviously if it's a choice of all things being equal, then cost is going to be the biggest thing. If it's availability problem, obviously the one that's most available is going to be the most appropriate. So it's a mix of things, but it does combine all those elements together. And AI does that fairly well. But this has been built in our business rules more than it is AI, because this is less about an agent coming up with a great idea and setting a goal for that. It's more about what are the rules in place and how do I use those to make sure I'm making the right decision?

AI may inherit some of that and make decisions faster, but it's got to be informed by that data that Rick's been talking about, which comes from underneath. So I think yes is the answer. But it's a complicated question. So as you know, it expands. But that's pretty much how we approach it though. We're trying to take more of a realistic picture than having an AI entity come up with its cognitive space and saying, oh, this is what the money is, and this is what it should be. We use it every day, all the time, and I don't trust it for that. So hope that answered the question. How are you doing, sir? Is there another question from the back? You have to take turns. Line them up. How are you testing for and protecting against bias, for example, in generating things like the care plan?

Yeah, we've we've definitely seen some of that. Um, for the, on the example of the, um, agent we've done like prompt engineering, that's more prompt, even though I know prompts like big no, no, like from, from the sessions. But um, yeah, so we're just doing that through testing, you know, repetitive rapid testing. Um, and, and then we continue to just evolve the model. We're not doing anything, I think different than that. And if I got the question right, it was more the drift and the hallucinations from the AI itself. Was that the question? Yeah. So how. Do you mitigate. Bias? She actually was asking about bias. Bias within how we practice medicine, how we diagnose medicine, how we prescribe both for gender and race perspective, which is a risk of using AI to recommend or make decisions in that area. I don't think I got the question entirely, but one of the things that we do.

Know, I was going to say, well, I mean, so again, part of the reason some of that is really hard to get to, um, part of the reason that we keep the human in the loop and we really haven't removed them is just for those types of scenarios. Um, because I mean, every, every doctor has different patterns. I mean, it depends who you're working with. I mean, I think that's the same reason we have, We evaluate gold carding based off. Are you aligned with, you know, our care care strategy or, you know, losing the term here? I was going to say that one of the things that we do at Pega is it's an architectural approach that that we sort of take to it. So when we operate an AI agent or an AI entity in anything, we're sort of constricting its use, right? So we narrow down its sort of cognitive brain, if you will.

So it's very focused. So it actually eliminates or reduces a lot of the drift, a lot of the things that naturally does. So by narrowing it down, you could do a bit of that. But like Rick said, it's mostly a cautious thing. Um, you know, as you know, AI is like a 13 year old in a white room with a crayon. Um, you can do some pretty interesting things. Um, so what we try to do is control that a little bit more. And it's usually architecturally, we do a lot of it. And then you get down to the, the use of it, the training of it, all those things. But it's really, really down to narrowing the use and controlling that. It seems to be the one of the bigger answers to eliminating some of that, but it's still there in some areas. So human in the middle.

And then a lot. Of that's the catch all. But I mean, it's just a lot of testing and reviewing the information and going over it and through it. And with our governance committee and just having the right people involved, that's really all we're doing at this point. So again, most of that's aligned with the quality agent and, um, the, um, recommendation because um, recommendation is in pilot mode, it's not even really, it's more POC to pilot. It's not really even pushed into project. We're working on 25 different policies right now and trying to go one by one and understand where there's bias, where there's hallucination, and then trying to correct the model as best we can. It's time consuming. It's not like you do it in a week, right? All of our data and evidence will be already biased. Yeah. That's right. Materials past evidence that will already have bias.

There's definitely bias in the information we have because it's I mean, historical, right? We've got years and years of data and it hasn't been cleansed or anything like that. So yeah, we have to it. Is a huge issue in health care, for sure. I mean, it's just one of the things that we have. So this is one of the areas that putting data in a workflow is almost magical. Because what it does, it naturally curates the data and says, I'm not looking at the data at whole and that Lord knows where it's connected from or what it's, you know, what biases and things have been applied to it within the workflow. It tends to be much more specific. So we're instead of saying we're going against all the data, we're actually pulling it into a very narrow space, which we just need pieces of that data to make it complete.

And then that naturally narrows some of the bias down. It doesn't eliminate it entirely, but it starts breaking it down into smaller pieces that are easily or more easily and readily identifiable that you can start saying, okay, there's the issue. Cleansing data, though, is a, you know, it's just a forever effort if you're going to do that because it comes in low quality. And then how do you know what's right? I mean, we're the only industry I know that treats everything. That was a hypothesis and an actual diagnosis. The same in the historical record. So it becomes problematic just at that level. I put it in the workflow that you start eliminating some of those things. As we start figuring out what is the piece of data that needs to go here. It's not all those. It may be the most current, but it will start eliminating some of that.

We do see that happening just in the date operation. That's it. Well, thanks so much. Thanks for the questions. Anything else? Any other questions? Oh, look. So no, first of all, great strategy. These are obvious questions, which comes when you have a kind of AI strategy, which is so ambitious. Right? Human in the loop is the answer to it. So I have a very basic question. So you talked about EHR data getting into the whole processing piece of it. So are you well integrated with Epic and Cerner to get the information in from the clinical data? Or it's more like you get the information based on what's there into the whatever the doctor is prescribing as part of the request of the prior arts. Yeah, well, I mean, we get it both ways. Not integrated with Cerner right now, epic, EAP, things like that. So we're getting lots of data from them.

Kara reported in a lot of the hospitals, even competitors use that. So we're actually able to get a lot of that data. So we are ingesting that and we do have, um, but we also get it the other way, etc.. So if we have somebody that's on Cerner and they're probably still using a portal, at least until interoperability hits and they actually start to submit that way. Um, we would get it through the portal. That's the only way in that case. But we have such a huge epic footprint in most of the states, and I should say regions that our clients are in, that we're able to get to a lot of the epic data. So it's real time integrated with epic, or you get it as part of the auth request, which is coming through ability or one of the clearinghouse. Oh I'm sorry.

Okay. So you're asking. Yes, we do get the authorization request. You're asking if we get other data? Yeah. My question was, do you get the complete 360 view of that patient through the EHR information? When you show the architecture, it was. Medical records and all the information that comes with that. Yeah, we do get some of that information because again, the health plan will work with Highmark is actually moving to the EBP model. So we're starting to ingest all that data. It's not all there yet, but we are bringing in large amounts of data around our members now into a data lake, if you will. Actually, it's HD, it's a Google Cloud product right now that we're going to actually move away from soon. But yeah, we are getting some data. It's not just from an authorization standpoint. We also get data from different lab vendors, others that we work with on a case management side like Sword for PT, you know, we get data from them as well.

But yeah, hopefully that answers your question. We do get more than just auth data. I think the clinical documentation rule that just came out is going to be interesting Well, so now you can start aggregating data from them while under that moment of care. So pretty interesting the way this world is going to evolve. I think this is where a lot of the interaction between providers and payers will happen, because payers, providers are just, you know, they're besieged by the time it takes to actually provide an authorization to the peer to peer review, for God's sakes, and all those things that they get caught up in. The more that the systems allow that to be automated. And by automated, I mean, oh, you've got you sent me a, you know, a PA and now I've got to fill it in. You're missing some data here.

So instead of rejecting it or putting it back on your staff, if I can just reach into you, you know, using fire or some method like that, I can actually get the data completed. The genetic stuff would do that. And now I give it to you in a more complete way. We're seeing this happen across the spectrum when it comes to different areas of just managing and putting AI like an, an intake process to clean up things, to just make the workflow work better seems to be great ways to start using it as opposed to, let's just rebuild the world and see if we got it right this time. It's a much better approach. I think you'll see that more and more about how we do these things. It's actually adding to what she asked. You gave an example, Stephen, right. When the clinician called Stephen, you're pulling similar data, right, to going back to what she asked in terms of preventing any bias.

Are you paying more control into your data and labeling controls? How do you ensure that? Because there could be a bias based on demographic, based on what you're. Because you're pulling some similar profiles when you were showing the Stephen use case. So how are you preventing that? Because that's a huge thing in terms of bias and alternation that happens, right? I understand why you said don't feed it extra knowledge base of data. Keep it within guardrails. But in this scenario, there is a chance for bias to happen. So is it that the data and labeling controls, where do you enforce that check? I understand the data supply chain. All that has to be governed and everything else. But have you guys seen any bias in that specific instance. I don't. Know. He's talking about the care plan that we were just talking about. Like all that information.

So as I what I was talking about right now, that's more proof of concept and pilot moving to pilot mode. Yes, we have seen some. We continue to test and evolve that, but we're probably never going to move. Remove Tom from being the individual to actually looks at all that information and makes the determination whether we should stay with it. But it's just continual testing for us. Right. Um, that's what we're doing right now anyway. Yeah. All right. Thanks. Sure. Well, I think we've run out of time. According to the clock. Um, I want to thank everybody for coming. This has been great for us, and hopefully we got some good insights out of it. Um, so with that, you're going to go see the chess tournament. Not much of a tournament.