Sanjog Aul [00:00:02]:
Hello, everyone. Welcome to the CTN Virtual Think Tank. The topic for today is Agentic AI in Healthcare: Creating Measurable Value Across Care, Service, and Digital Operations. My name is Sanjog Aul, and I will be your moderator today. I’m from CIO Talk Network, that is CTN, and thank you so much, Genesys and Cognizant, for making this possible. Thanks again.

Sanjog Aul [00:00:30]:
What we are doing here is, as the topic says, agentic AI in healthcare. It shows a lot of promise for sure, but whether some things are truly working out the way we expect could be a different question. Look at the different areas in which agentic AI could be applied, and whether it is being applied in the best way possible, what we know based on the technology that’s available, the culture, the policy, and the public sentiment. There are so many different aspects coming into being. With all that investment going in and all that effort going in, is the value getting created for the patient, for the members, for the care teams, and for the operations in these enterprises, so that we could measure them, trust them, and feel that this is truly creating value? This is supposed to be an interactive, collaborative, conversational dialogue. That’s how we’re going to run it.

Sanjog Aul [00:01:32]:
We’ve got very experienced leaders for this discussion. I’ve got Sanjay Gupta, Chief Transformation Officer for Novant Health. I’ve got Robbie Freeman, Chief Digital Transformation Officer for Mount Sinai Health System. I’ve got Michelle Crovo, who is the CX and AI Consulting Leader for North America at Cognizant. Then we have Joan Cox, who is Chief Experience Officer for LifeBridge Health. We’ve got Ashok Chennuru, Chief Digital and Data Transformation Officer for Elevance Health. We’ve got Chris Helzerman, who is the Industry Strategy Executive for Healthcare at Genesys. And we have Mouli Venkatesan, who is the Chief Information and Technology Officer at Blue Cross Blue Shield Association. Such a fabulous panel. Let’s very quickly go through the mechanics of how we’re going to run this. I’m going to request each of you to take two minutes and give a quick background about yourself and your organization to set the context, and share the top challenge which holds you back from making the most of agentic AI in your environment.

Sanjog Aul [00:02:40]:
Once we have gone through that, we will come back to the next session, which is a solution discussion, where we take those top challenges and see if we can put our heads together to crack some of the problems where we are hitting snags. With all of that said, I would request one more thing. Given we are going to talk for an hour on agentic AI, I also wanted to see if you could give it a personality. So along with the top challenge that you explain, could you answer this question: if you were to compare AI agents to an animal, what animal would that be and why? With that as the format, let’s start with you, Sanjay.

Sanjay Gupta [00:03:24]:
Sanjog, thank you for inviting me. I’m Sanjay Gupta, Chief Transformation Officer at Novant Health. We are a $12 billion healthcare provider across the Carolinas. I spent two decades before this in financial services, at Bank of America, Ally, and Allstate, before I got into healthcare to have an opportunity to see how I could contribute to making healthcare a little better, and fortunately, Novant Health has that as its core philosophy. Speaking of the challenges in front of us, I see two core aspects. One is that if we as consumers were looking at healthcare, I think it’s safe to say none of us would design it this way. So there are opportunities to make healthcare more convenient and more understandable.

Sanjay Gupta [00:04:11]:
I think fortunately AI has the ability to contribute significantly in that regard. The second aspect is that across healthcare, we are facing significant headwinds which bring with them a cost challenge, and again, I believe AI can contribute significantly to help alleviate some of those headwinds. As far as animals are concerned, Sanjog, it’s really hard to come up with a compelling answer, because the first thing I thought of was a monkey. A monkey doesn’t quite convey fortitude, but it is an animal that is very enterprising in coming up with unique solutions, and I think in some ways AI needs to do that as well. So while the visual is not great, from an intelligence perspective, it works.

Robbie Freeman [00:05:01]:
Thank you, and thanks for the question. I think one of the biggest challenges ahead is how we ensure that we’re using and deploying AI in a way that’s responsible. We know that this technology is not without risk, particularly generative AI. What we don’t want to do, particularly as healthcare delivery organizations, is just turn it on and let it loose out in the wild. We want to make sure that we have the right process pre-deployment to understand performance, safety, accuracy, and security before we let it loose, and then we want to make sure that we have the right processes for ongoing evaluation. The better we can do that, the better we can calibrate both our risk and our decision-making around how these tools can add value. In terms of which animal, the first thing that came to mind for me is a golden retriever, because they’re so eager to please and help, but they also get it wrong.

Robbie Freeman [00:06:03]:
If you ask your golden retriever to go and fetch something, they may bring you back the wrong thing. You may get a shoe or a sock, not the ball that you threw. AI agents also get things wrong, and that’s why we need responsible use of AI. That’s why we need the right level of monitoring. That’s why we need to use agentic AI for the right use cases, because it doesn’t get things right every time.

Michelle Crovo [00:06:36]:
Hi, and thanks for having me again today. I’m the CX and AI Advisory Leader at Cognizant. I sit within our Moment practice, which is a unique part of Cognizant that focuses on end-to-end experience transformation. I spent the majority of my career in service, whether on the operations side or, having started in industry, then in consulting. I spent a lot of time in customer service, so I think through not just the customer experience but the agent side, and the back-office and middle-office teams. I think it extends to everything we’re talking about today. I’ll probably talk more about the problems of actually creating, developing, and implementing AI and where we’ve seen challenges, but I’ll hold that for that section. The first animal I thought of was a raccoon, and I don’t know why.

Michelle Crovo [00:07:22]:
I think it’s because they’re shifty, unpredictable, and scrappy. When we talk about some of the challenges we’ve had, simplistic AI can work. When you expand it, even when you test it, it seems like it works, and then you scale it and start to run into all sorts of different problems. I don’t know why I came up with an unpredictable animal, not a fluffy golden retriever, Robbie.

Sanjog Aul [00:07:50]:
That’s fine. If you were to spell out in short, like a phrase, the heartburn we spoke about, the top challenge you’ve seen, given that you’ve got experience working with so many of them, have you seen a pattern where something pops up as the top one?

Michelle Crovo [00:08:08]:
I think there’s probably a tie between two. The first is picking the right use case, and trying to pick one that isn’t too complex and is low risk to start with, which in healthcare specifically can be really challenging. The other is which platform. There are so many platforms out there today, and they’re all changing and evolving week to week, month to month. Which one do you pick? How do you not get locked into one? Should you be experimenting with multiple? So the heartburn is often where to start. And then, when it’s not working, is it how we built it? Is it the platform we’re using? Trying to figure out where and how to pivot is the other part that we handle a lot.

Joan Cox [00:08:52]:
Hi, everyone. My name is Joan Cox. I am the Chief Experience Officer for LifeBridge Health System, which is a five-hospital system in the greater Baltimore area. We have a large medical group, and then we have a large LifeBridge Health Partners side that has insurance, long-term care, urgent care centers, ASCs, and all kinds of additional business functions. I’ve been in healthcare my entire career. This is my second iteration as an inaugural CXO in a healthcare system, so it’s been quite a journey helping the organization adjust and focus in that sense.

Joan Cox [00:09:34]:
I come from the greater Boston area originally. I spent many years in Boston, not to date myself, but I came up during the age of Romneycare, so we implemented and understood quality-based reimbursement and quality-based metrics at the very beginning of that exploration and those changes in healthcare. My background is in mental health. I always say that other than a couple of months delivering pizzas for Bertucci’s when I got out of college, I’ve been in healthcare my entire career, and it’s been an incredible gift and honor to spend the majority of that time, certainly the last 15 years, focused on patient experience and provider and team member experience. Given that perspective, the greatest challenge we are seeing is that we’re very young at LifeBridge in our AI journey, especially our agentic AI journey. One of the challenges we face is systemness, because we don’t have a very concrete process and a sense of systemness.

Joan Cox [00:10:47]:
We are very independent facilities that have come together. That’s a challenge when you’re looking, as Michelle said, at whatever platform, to make sure you’re finding the right one. The other big challenge we have is making sure that the right people are at the table while those decisions are being made. There are a lot of really great solutions out there that seem like they’re going to solve a lot of our issues, but when you have the people who are actually going to be using it, and, through our PFACs, our Patient and Family Advisory Councils and other internal consultants, the people who are going to be impacted by it, you want to make sure that they’re part of that discussion and conversation. That’s not an easy thing to do in an implementation and exploratory process. Most importantly, keeping the human at the center of all of this, making sure that our patients are the ones who are truly benefiting from it and that that’s our true north in implementation. It can get a little hairy when cost savings and reduction in resources are on the table as well, making sure that our true north is really what we’re following in our mission.

Joan Cox [00:11:57]:
Like Michelle, when I think of agentic AI, the first thing that came to my mind was a Gremlin. I’m dating myself, but the movie Gremlins, right? A cute, adorable, wonderful little thing that everybody wants in their back pocket, but you put the wrong thing in it, you add water, and it turns into a monster. So making sure that, A, we’re putting the right thing, the right information, in, we’re building it properly, but also making sure that we’re not steering it in the wrong direction, because things can get pretty hairy when we do.

Ashok Chennuru [00:12:36]:
Good afternoon, Sanjog. Thank you for having us. Ashok Chennuru, Chief Data, Digital, and AI Transformation Officer for Elevance Health. We are a large Fortune 20 company that serves over 120 million members. As some of my fellow panelists have articulated, the industry is facing both headwinds and tailwinds related to how do you simplify the administrative burden that exists in the system between the payer, providers, and other partners? How do we keep the consumer in the center with focus on experience, and then how do we manage medical costs? From an agentic perspective, the opportunity is how do we leverage the tech to simplify the whole burden that exists, leveraging agentic autonomy to simplify how we pay for claims, how we authorize for care, versus manually exchanging data back and forth. That’s the opportunity we have. From an animal perspective, I would say agentic AI is like training a dog.

Ashok Chennuru [00:14:02]:
The more you train, the better it gets, and you may think that you have achieved perfection, but then another problem, and then you need to probably redo some of that again.

Chris Helzerman [00:14:16]:
Hello, thanks so much. Chris Helzerman with the Genesys healthcare team. I’m an industry strategy executive here at Genesys. Genesys has a pretty sizable focus in healthcare: 70% of US lives are with plans that use Genesys. We’re focused on helping our clients achieve the most value with their solutions. Leaning into the question about the challenges we’re seeing across our customer base, we’re seeing a struggle with where to focus and optimize value, how to accelerate value realization, and which solutions are going to have impact. Intuitively, I see a lot of organizations starting with a focus on operational use cases, and in some instances they realize that simpler solutions can solve for those, and that agentic isn’t necessary.

Chris Helzerman [00:15:15]:
In other areas where we’re seeing more and more focus is in helping our clinicians operate at top of license and using agentic to augment and free them up from administrative burden and allow them to focus on patient care and improve access. I think there’s a lot of hope in that particular space. When you talk about animals, what comes to mind is a mole, because AI is popping up everywhere.

Mouli Venkatesan [00:15:57]:
Great. Thanks, Sanjog, for having me on your show, and great to see all my fellow panelists here. Mouli Venkatesan. As Sanjay said, I lead technology and digital for all of the Blue Cross Blue Shield system as part of the association, and the Blue Cross Blue Shield system touches one in three Americans. 120 million Americans are covered by that. When you drive at scale, one of the most important things in front of us is how do we create solutions at scale that drive one of the major missions for us: how can we make sure that healthcare is affordable in every zip code across the country that we are privileged to serve our members in.

Mouli Venkatesan [00:16:45]:
With that, the challenge in front of us is twofold. One is how we continue to bring down the cost of healthcare from ever-increasing medical costs. Second, healthcare still works on paper and faxes, with a lot of friction between multiple entities, to name a few, payers and providers and other health systems. How can we use agentic AI to have a laser focus on reducing friction for consumers like you and me across the system? It’s a major focus for us and a major challenge, especially when you have to collaborate at scale across a system that touches one in three Americans. That is what is in front of us as a challenge, and we’d love to get inputs and thoughts on how we can go after that. On your question about the animal, my thought was a horse. It’s nimble, it has speed, and it has resilience.

Mouli Venkatesan [00:17:46]:
It has grit, because it’s not just the speed. It can go on mountains, it can go on grit, the terrain can be any rough, you can still go on it. That’s the animal that came to mind.

Sanjog Aul [00:17:59]:
Very nice. I love the fact that each of you use a unique animal. Nobody repeated. That is so awesome. A lot of creativity in the room. With the different challenges that we have, we’re now going to move to the solution discussion session. Let’s take two minutes or less if possible. Sanjay, starting with you.

Sanjog Aul [00:18:17]:
The problem that you mentioned: could you tell us specifically what you tried to crack it? Because that’s your top problem, I’m sure you would have put all possible resources behind it. What happened? And what could this room do to help you get past that, if possible, through their inputs and insights?

Sanjay Gupta [00:18:35]:
Thanks, Sanjog. I’ll start with the customer experience side. As you can imagine, there’s a lot to get done there, but I’ll share two examples of how things are improving. One is that care is required around the clock, because when we have a medical issue, it doesn’t come with a schedule, and yet availability of care is limited to a finite number of hours. Otherwise you end up in urgent care or the emergency room. One of the ways we tackle this is by launching virtual primary care, available 24/7. The idea is that you can now engage with us at your convenience, and care is no longer limited to a certain number of hours.

Sanjay Gupta [00:19:25]:
Where AI comes into play is that when you engage with us for virtual primary care, the intake process is done by AI, which saves our physicians a significant amount of time and allows them to see a lot more patients. That’s how AI is being very helpful, not only in giving access but also in giving our physicians the ability to see more patients. A second example in that vein: if you happen to be in a hospital, I think all of you can relate that a point of frustration is when the physician will come around for their rounds, because you don’t have time to wait all day. We have implemented an AI solution across three of our facilities whereby the physician hits one button when they start their rounding, and the AI informs the patients or their loved ones exactly when that physician will be seeing them. If the physician gets delayed, which is possible during rounding, it updates the time. Again, it’s addressing points of access, but also making care more convenient for our patients. Those are just two examples.

Sanjay Gupta [00:20:36]:
My point is that we have a long way to go and many more use cases. This is just a couple of examples to start the journey. On the headwinds I talked about from an expense perspective, one of the places we started was our AI inbound and outbound calling. As you can imagine, you have two issues there. Leveraging human beings to call can be expensive, but you also have limited hours. This does not have to be a case where, by taking out expense, you make it more inconvenient. In fact, the opposite is true. You can actually make care more convenient.

Sanjay Gupta [00:21:17]:
Our call center is now available round the clock for mammography, for example, where we’ve implemented an AI inbound and outbound calling solution. Not only have we made care more convenient, but our abandonment rate has gone to zero, because AI never sleeps. That is just another use case of what we are implementing, and there are plenty more I can speak to. The challenge we have is more of a cultural one: people, while they recognize how powerful AI can be, are afraid of it as well, because they’re not quite sure how it’s going to play out. So, more to come from a discussion perspective.

Sanjog Aul [00:21:59]:
Sure. If you were to walk away with something actionable, what question would you pose to this group so they could give you some insight and you walk away a little more enlightened?

Sanjay Gupta [00:22:12]:
I would love to probe the cultural aspect, to understand how other people are getting buy-in across the enterprise, so that it’s not just the purview and objective of the AI team in the organization to roll this change out, but the entire enterprise gets excited about the possibilities and embraces them.

Sanjog Aul [00:22:36]:
Any takers on the cultural front, if you’ve pulled some magic on that?

Joan Cox [00:22:40]:
Sure, I’ll jump in. One of the key elements we’ve introduced is getting the end user to the table early on for the discussions, for identifying what the true problem is that we’re trying to solve for them, not just what conceptually we think of from the operational side, and getting them involved in the selection of the product. Most recently, we’ve moved to a new interpreter services system that leverages significant AI in the support. It’s really helpful, not only in making sure we’re choosing the right product, but in the adoption, because they were part of that process. This was their choice. They’re the ones who decided that they wanted to choose it, so there’s kind of membership inherently tied into that process. That’s one of the things we’ve done that has been really successful.

Robbie Freeman [00:23:46]:
I would love to build on that point, because I totally agree with Joan that we need to have the end users embedded and part of the decision-making, and at times you can even co-design with them. From a culture perspective, we’ve also found that peer-to-peer works: if you can have a network of champions or ambassadors who can be out there teaching each other, that works really well. If I step out onto one of our inpatient units in a suit as someone from administration, it doesn’t land as well as when it comes from one of their colleagues. So we have a network of folks, everyone from the front desk to nurses to physicians, who can go out and work with their peers when we are implementing a solution, and that has helped a lot with change management, buy-in, and adoption.

Sanjog Aul [00:24:40]:
Sanjay, the ideas that came to you: I’m sure these are ideas, or a flavor of them, that you might have attempted to adopt. Do you see some things working versus others not working on the ground for you?

Sanjay Gupta [00:24:57]:
I think the challenge is not to get people engaged. When you’re trying to take on impact at scale, you want to make sure that what people are signing up for reflects the full opportunity. If the opportunity is worth $100 million, what you’ll invariably find is people will say, “Well, I’ll sign up for $10 million, because I’m not quite sure how that’s going to play out,” and that has an impact in two ways. One is that you can then only invest for a $10 million opportunity, which is not ideal; you want to invest for the $100 million opportunity. Secondly, you want to run at pace, so that you’re not doing things the traditional way. Prior to AI, you would come up with a three-year implementation plan, and that was okay. Now it’s not okay, because not only is the AI domain changing so quickly, but as I mentioned, the headwinds are strong enough that we don’t have the luxury of waiting.

Sanjay Gupta [00:25:57]:
So that plays out both ways.

Mouli Venkatesan [00:25:59]:
Sanjog, a couple of thoughts on this point, from my previous life as well as the newer organization. I would start with this: AI is not a technology agenda, it’s an organizational agenda, and it has to begin with that. I’m sure Sanjay is thinking the same way. We categorize it into four sections: mindset, heartset, skill sets, and tool sets. More often than not, in every company, and I’m sure there are exceptions, AI starts with the tool sets. I’ve got a phenomenal tool. I found a great tool.

Mouli Venkatesan [00:26:34]:
A great partner is bringing in, I want you all to see it, and so it starts with the tool sets conversation, and then we run into challenges: let’s create a business case for it. I would rather start with the mindset. The mindset shift is: should we do it incrementally and make a process better, or should we completely eliminate the process? The same way that Sanjay talked about in terms of his intake process, where they are trying to look at, is that a value-added process for a human or not? If not, can we eliminate it, or can we automate it differently? That mindset should come with the heartset, which is that healthcare is still personal and has to be human-centered.

Mouli Venkatesan [00:27:13]:
It has to be driven both on the care side and on the plan side. How do we drive with the heartset? Then comes picking your tool sets in terms of technology, which is something that is significantly advanced. Then, is your workforce skilled enough? Usually people focus on the technology workforce. I’m talking about your entire organizational workforce. It should be an AI-knowledgeable workforce, from your operations to your finance to your clinicians to everybody. Then how do you make sure you drive a robust program on that? Otherwise, it becomes a project with a start date and an end date, whereas an AI agenda in an organization should become a movement that does not have a start date and an end date, and that lives beyond a particular person that’s driving it.

Mouli Venkatesan [00:28:08]:
A couple of ways in which we have done this: you have to create organizational advocates. Predominantly, those advocates have to be more across your businesses, not in technology. You’ve got to create those advocates ecosystem, within which are your culture advocates, and incubate that, and create champions within your organizations. That is when you actually transform. Otherwise, you will get incremental value. You can absolutely drive value.

Mouli Venkatesan [00:28:39]:
There’s no doubt about it. The $10 million that Sanjay talked about is still a valuable effort, but you won’t get the transformation that you need.

Sanjog Aul [00:28:45]:
Robbie, let’s listen to what you’ve tried to do with the challenge that you’ve had, and then let’s have the team help you out.

Robbie Freeman [00:28:53]:
On the challenge of ensuring that we’re doing this responsibly, we stood up an enterprise-wide governance structure. We used to have separate digital and AI governance, and we brought that together about two years ago. One of the cross-cutting functions that helps us there is what we call our AI Assurance Lab. We have a dedicated independent entity that reviews the high-risk AI models. Before we implement a use case, we risk-score it to say what the risk of failure and the risk of harm are, and then our data scientists, engineers, and assurance lab do that testing. We feel strongly that people shouldn’t grade their own homework, so the folks who build the model don’t evaluate their own model.

Robbie Freeman [00:29:39]:
This independent team will evaluate the model. About 30% of our portfolio is in-house software development, so the assurance lab will test and evaluate both internally developed AI and AI tools coming from third-party vendors, to make sure we understand whether it performs as advertised, whether there is any risk of harm, and whether there is any bias. We’ll look by self-reported race, by payer, by gender, and by other demographic information to see if there’s any bias, and then make a recommendation back through governance on whether the tool is safe. More broadly, in our Department of AI and Human Health, we’ve published on some of the popular consumer tools out there. We recently had a paper in Nature around ChatGPT Health, and what we saw was that oftentimes ChatGPT Health will over- or under-triage a patient who’s asking healthcare questions, which can be quite concerning, particularly the under-triage. If ChatGPT Health tells you everything’s fine, stay home, but everything’s not fine and maybe you need to be seen, that could be a problem.

Robbie Freeman [00:30:57]:
Or if it tells you to go to your closest ER and you don’t need to be there, that contributes to overcrowding and unsafe conditions. Some of the tools out there that consumers are using are not yet, based on our research, ready for prime time or for real medical decision-making. That’s why I think figuring out how we do this in a way that’s safe and responsible is going to be so important. We don’t want to just keep implementing and scaling without having that infrastructure in place.

Sanjog Aul [00:31:29]:
What questions do you have that could be answered by the experienced folks here, which would help you truly inch forward?

Robbie Freeman [00:31:37]:
I’d be curious how folks are thinking about their infrastructure to monitor AI tools, and whether you feel there’s anything out there that’s mature at this point, either for ongoing monitoring or for broader orchestration of AI tools, or whether you’re just figuring out a monitoring plan at each use case level.

Sanjay Gupta [00:32:12]:
Robbie, I’m not claiming to have all the answers, but I can throw out a few suggestions. I think it is an opportunity where you have to look at use cases differently. Where we have tools that are not touching the patient, we can be more satisfied by a governance group looking from time to time at whether the model is working as designed, and backtesting the model to confirm it’s working right. That’s one example. As you get to the clinical side, the stakes go up. In your example, we’ve actually launched something called Novant Health Advisor, whereby patients can come in and get more sound answers than they could on ChatGPT or name your source on the wide web. The way we are doing that is a little different: we first have a judgment AI tool running behind it, so we have AI checking on the AI on an ongoing basis. Then we have humans testing it from time to time.

Sanjay Gupta [00:33:20]:
Then we have built-in clinical safeguards, so we have explicitly said that if the conversation goes here, please stop and switch it to a human being. I don’t think it’s one-size-fits-all, but there are different approaches you can take.

Chris Helzerman [00:33:34]:
I think this is a really important question. Historically, especially in operations, we would deploy an IVR that was very deterministic and very predictable. Now we’re operating in this area of agentic AI where the models are shifting around underneath and a lot of change is happening. It’s not just about the technology, it is about the people and the process. Here at Genesys, as we launch more and more agentic capability and bring it into the market, we are adding the ability for AI to check AI. That’s going to provide more and more capability to monitor in real time. We also think about this almost as a pool of agents or resources that are available and have to be QA’d and monitored, much in the way that we have historically monitored and QA’d our labor force.

Chris Helzerman [00:34:32]:
The various quality processes that we have in place for our human talent, we believe, are also important to implement for the agentic agents. Wraparound services, dashboards, reporting, recording, and analysis are all really important. We don’t just stand this up and walk away from it, and I think you’re spot on in asking this question.

Michelle Crovo [00:35:07]:
I was going to say that most of the models, tools, or frameworks you choose should have traceability in them, so making sure that you understand how to use that. At Cognizant, we’ve rolled out a broader focus on context engineering, really thinking through how the agentic framework is being used and what the different nodes are. We’re getting technical here, but as you look at how an agentic bot or orchestration is being built, you can build into the model two different agents checking itself. You can do different loops and then see where it failed and how it failed, going back into the tool. All that said, as you work with technologists, we talked about tool sets and skill sets. A lot of the times when AI is failing, it’s not actually the model itself. It’s some kind of condition or guardrail that wasn’t set, or a constraint that wasn’t put in, limitations that weren’t applied, or not having an authoritative system that it’s referencing.

Michelle Crovo [00:36:06]:
So when we think about that, Robbie, it’s more about what is it referencing, what did we tell it to do or not do, what limitations should we put on it, and then making sure that we’re choosing models or platforms that have traceability built in. Typically we’re also working with hyperscalers as you think about scaling, so you should be able to use what the hyperscalers offer you when it comes to observability. I got a little bit technical there, but that’s how we look at it, so that it’s platform agnostic. You can choose a whole bunch of different models, and most of them have it built in. It’s just making sure you know how to use it, and then, when something triggers, how you go take a look at the traceability of it and figure out what happened.

Sanjog Aul [00:36:46]:
Michelle, now back to you about your challenge, which you try to tackle, and the insights, the ahas, the hallelujah moments and doomsday moments, whatever you want to share, and then we see what the group here could help you with.

Michelle Crovo [00:37:04]:
Sure. I posed two: picking the right use case, thinking about complexity and implementing it, and then platforms. I’ve talked a little bit about the platform, so I’ll choose the former. When you think about agentic, where it works well is when you can pick a workflow where you can automate or pull together front, middle, and back office. It’s going to be inherently complex to begin with. As we start to build these use cases and work with our customers, it’s identifying all the different sources of truth and data sources. What are all the different teams we have to work with? What are all the different source systems? How do you do this today? It just takes a really long time to get all the right people together.

Michelle Crovo [00:37:48]:
As we started to build this, even about a year ago, starting to pull it together, trying to build it in one agent and realizing, no, we really need to break this down: a different bot that can focus on just eligibility, a different one that can talk about benefits, a different one that can check prior authorization, a different one that can do XYZ. Great, we have all these different bots that can do one specific, small part of the process. Then we needed to pull together an orchestration agent to be able to figure out, based on the conversation that we’re having, which of these other bots do I need to call on to do the specific thing? The other part was having a really robust intent agent, because a lot of these complex use cases have multiple turns, meaning a person’s talking and they keep changing their mind about what they’re talking about, and so having the technology be able to address that and go back and forth. A lot of the lessons learned were in testing it. It didn’t work as expected. How do we break this down further? What guardrails do we need to update? We probably allowed it to do too much. How do we create some constraints there or do some double checks? That just takes a long time. Working through that just takes time with the right people.

Michelle Crovo [00:39:06]:
As I look at this panel of C-suite leaders, how would you approach making sure we have the right people in the room throughout the process, and that we can get the right, accurate information? Because, like I said before, it’s not usually the technologist or the skill set, and it’s not usually the tool. It’s typically what you told it to do, what was the source of truth, and the requirements, how we designed it. When we don’t do that well, we miss something and then have to rebuild it, and that’s how it gets really hard to go from a pilot or a POC to full launch, because every time you add something new, something else is missed. That’s my question, and I think it’s a good question to ask in general for anybody who’s watching this: how do you think about setting that up for success to begin with, knowing that most of the use cases that are really good and ripe for agentic are going to be complex and involve five or six different functional areas?

Joan Cox [00:40:08]:
I can talk a little bit about, just to expand on that, the challenge there, Michelle. I think a part of that friction as well is we stand up a piece of technology to solve a problem, and we’re really relying on the people who are utilizing it, but also the people that are impacted by it, to almost do that constant QA. And they’re super busy, because they’re actually taking care of patients. Or they’re a patient who’s hit a friction point, and they’ve now said, okay, I’m not going to continue this contact, I’m going to go a different route. In every second of every day in healthcare, workarounds are being built in reaction to things that are not working. What we are trying to do, in the beginning of our journey, is make sure that this is a continual process and a continual conversation, that this isn’t just a solution. It’s actually another work stream, a continual work stream, which is a huge challenge when you’re talking about an environment that, to Sanjay’s point, I’m in Maryland.

Joan Cox [00:41:30]:
I don’t know how many of you are familiar with the Maryland model, but we’re facing significant headwinds in financial challenges and resource management. I don’t have an answer, Michelle. It’s a great question, but I do know that if we don’t get it right and have that real dedication to making sure that we’re constantly doing QA with the people and the patients, then everybody is just going to develop their own workarounds, and we’re going to have patients showing up at the office because they can’t get through properly on the call center, or they’re not getting the answers to their questions that they need, which we’ve seen happen as well. And also really doing the due diligence on the front end, before the build, to understand what those numbers look like and what the full context and size of the problem is before trying to implement the solution.

Sanjay Gupta [00:42:33]:
Thanks, Sanjog. Michelle, I think it’s a very valid question. Things are changing so fast. A year ago we were asking, will AI work? Six months ago, we were asking, AI is going to work, but how much? And now we are asking, okay, can we do more? Things have moved very quickly in a short amount of time, but the important part is that we had a sense of what we wanted to build. We knew the Lego blocks, as we call them, were going to fall into place and build something bigger, but if we had painted the entire big house we wanted to build, people would have given up on us. Instead, we started with, let’s go build the Lego block. In the case of the call center, I started with mammography, not because it was the biggest use case, but because it gave me a chance to prove the concept works really well.

Sanjay Gupta [00:43:20]:
Once people saw how well it was working, the person in pharmacy raised their hand and said, “When can I have this?” The person in physical therapy said, “When can I have this?” So it created a pull, but the call center was just the start of the process. In parallel, we were also building the decision tree logic so patients could schedule with the right provider. Now you have two Lego blocks: one that can schedule, and one that knows who to schedule with. Then you add a third Lego block that understands what intake I can get from them. My point is, yes, these are Lego blocks you’re building, but you don’t have to tell everyone you’re going to build all of them, because you will actually lose people. You’ve got to prove the case as you go, but you’ve got to have a north star of where you’re going.

Sanjog Aul [00:44:05]:
Michelle, if I understood your question correctly, you were saying that it’s a moving target. Every time, you could have a different set of stakeholders who might have the input required for you to do it right. Sanjay, what you just mentioned is more of an adoption cycle: you find some heroes and start with the smallest one. But the way I heard Michelle, even within mammography that you mentioned, you start, and initially two people show up in that meeting, but later on we uncover something and a third person comes and says, “You did not cover my use case.” So this becomes an interesting challenge. Do you have any inputs on that, how do you tackle that, so all stakeholders, if possible, with an evolving requirement, could be brought up front? Is that even practical to envision or imagine? Or is it a leave-it-to-God problem?

Sanjay Gupta [00:45:02]:
Two points there. One is that we never said we’ll take care of every phone call. Our goal was to take out at least 50%, and if you go north of that, great. That leaves room for someone to say, “Yes, we hear you, and we will get to that particular use case as well.” But Michelle, inherently what you’re trading off is what speed you want to go at versus how comprehensive you want to be. You can bring everyone into the room, and you will delay your process substantially, because they will all have different opinions and it will slow down. There’s an appropriate time and place for that. For an AI project we’re looking at that is going to touch our hospital operations and OR, we have done exactly that. We’ve built a cross-functional group.

Sanjay Gupta [00:45:49]:
That entire group was with us when we did the reference checks. The process is going to go a lot slower, but we will have greater adoption. In other cases, we don’t have that luxury and we go in a more nimble manner. Not a perfect answer, but some insight.

Mouli Venkatesan [00:46:05]:
I keep it pretty simple in my head, Sanjog. On this problem statement, I use a concept called problem finders and problem solvers. How many of us have invited four people to a meeting and ended up with 27 people in the meeting? All of us. It’s because people like to go to meetings, and people like to be relevant. In this context, we’ve got to look at who the problem finder is and who the problem solver is, and remove the seven layers of separation in between. That is the organizational challenge all of us have, and will continue to have, not just in healthcare. The more we are grounded in where the problems are and who the problem finders are, and connect the problem solvers to them, the more we can accelerate AI, or anything, at scale. AI can absolutely accelerate scale when we do that.

Mouli Venkatesan [00:47:01]:
The hard question we’ve got to continue to have is: what is your relevance to this conversation? What is it that you’re bringing in? In my career I’ve been in many conversations where I call myself the professional meeting-goer.

Mouli Venkatesan [00:47:18]:
I add zero value to those meetings. I don’t know why I got invited, and I don’t know why I’m there. In many organizations and many companies, everybody’s been there. The hard question we have to ask is how to get the problem finders and the problem solvers together faster. It takes 18 meetings before they talk to each other. That is the problem we have. That is the problem in speed and velocity.

Sanjog Aul [00:47:41]:
Joan, let’s come to your challenge, what you’ve specifically tried, and let’s see if the team can help you get a better answer.

Joan Cox [00:47:48]:
As stated earlier in some of these discussions, one of our biggest challenges is a lack of systemness, and it’s not a failure on the organization. It’s working as designed, as they say. It really is how the organization has been structured, and it’s one of the things that has made it an attractive partner for some of the hospitals we’ve brought on. The challenge that we face is being able to find solutions that we can implement across the system, the same conversation that everybody’s having about how do you get the right people at the table and not have it negatively impact your speed to execution. Quite frankly, Mouli, I think you just solved my biggest issue. That’s a wonderful approach and a really brilliant concept.

Joan Cox [00:48:56]:
I also think our challenge is making sure that we are meeting the needs. I look at this in a little bit of a different lens, in the sense that I’m really frontline connected with our patients and our patient feedback and our patient interactions, as well as our team member engagement and physician engagement. So making sure that we are providing a tool that is going to enhance human connection as opposed to replacing it, and making sure that it is truly doing that. Ambient listening is something we’ve implemented across several of our facilities, certainly in our med practice and our primary care clinics. That’s amazing, and it certainly has helped to facilitate opportunity for that human connection, but it only works well if it’s not a barrier to initiate the actual function. Also, we need to make sure that we’re providing skills training for those who need opportunity around enhancing that human connection, who need the skills to be able to engage in those elements. My biggest ask would be: how do you best evaluate a tool for systems that’s not systemized, essentially? That’s probably our biggest challenge, quite frankly.

Joan Cox [00:50:36]:
Again, we’re very much at the beginning of this process. One facility we have in particular is very advanced from this perspective. We have smart rooms and extensive virtual care taking place. From an agentic standpoint, we are implementing agentic AI in our population health management process, in our post-discharge calls. It’s that first layer. It’s not replacing those conversations, but it’s the first layer of outreach, which has been pretty incredible. Our continued challenge in particular is our community support and the patients we are trying to connect with. Oftentimes, the ones we need to connect with most, from a quality and a reimbursement perspective, are the ones who are the least engaged from a technology perspective, whether it’s the community they live in, their access to technology, or their comfort. The digital divide is a very real challenge.

Joan Cox [00:51:46]:
It certainly should and can be considered a social determinant of health, especially as we grow AI and technology in healthcare delivery. We need to make sure that we’re not leaving behind the people who are impacted by it and least able to engage.

Sanjog Aul [00:52:07]:
So is your question to get a tool which would serve well for a chaotic environment, or rather where you are at the rudimentary stage of adoption of agentic? The historical philosophy says first get your business in order, then you slap tools on it.

Sanjog Aul [00:52:26]:
You are partially there anyway, but you’re saying there is still chaos. Does anyone in the room have a recipe for something like this, or have you seen some other organizations who may be in a similar position that might have done something to crack this? Any ideas? You can take that offline if you don’t want to tell which chaotic environment you did solve. But is something possible? Chris, I’ll come to you, because you are from the solution side, and a lot of people would come to you to say, “Hey, can you guys do miracles?” And you will say, “You know what, get your business in order.” So, would you have a suggestion for Joan?

Chris Helzerman [00:53:09]:
I was thinking, Joan, that piloting different solutions comes to mind. We don’t know what will work or what won’t until we start trying things. Throughout my career, and I totally appreciate the comments about social determinants of health and the challenges of engaging different patient communities, I think a lot of times we have some bias about what will work and what won’t, and I’m always surprised by what does. I’ve been in the room when we’ve said, “Oh, you can’t use texting and digital to engage a Medicare patient or even a Medicaid patient,” and yet, when you think about the Medicaid population as an example, many of them do have access to mobile phones, because they’re provided at substantial discounts or there are incentives for those.

Chris Helzerman [00:54:15]:
Often that’s their only connectivity, because they don’t have home internet. There is a digital divide, but how do we meet these patients where they are? Testing different engagement tactics is really where it’s at, and we learn through rapid tests of change. To me, the thing is, let’s not just say, “Oh, that won’t work.” How might we try some different techniques and see what does work?

Michelle Crovo [00:54:57]:
From the platform side, I tend to start with what problem you are trying to solve with the solution, or what use case, because there are some solutions that can scale enterprise-wide regardless of the other systems that are in place. Being on the functional business side, I would just bring in my technology friends and get them all into a room, for they’re experts on the different systems, and ask them to tell me what’s compatible and what’s not. So I would say starting with a use case or an outcome that can be solved regardless of the tech you have. I do think that does exist. Or there may be platforms that are compatible with 50 or 70% of what you have in place today, so maybe doing a POC or a pilot there and then potentially an alternate solution. As your different locations come together long-term and as contracts expire, I’m sure the goal is to simplify and standardize, but by being able to look at solutions and, as I said at the very top of this call, trying different platforms as they evolve, you may end up doing two different POCs that help you ultimately make a bigger decision across your enterprise.

Michelle Crovo [00:56:12]:
There are solutions that you can deploy for certain use cases that are not necessarily dependent on what you have in place from a legacy perspective. I just wanted to offer that.

Sanjog Aul [00:56:22]:
Let’s move to Ashok. You did mention, and it’s a good problem to have, that you have a very mature environment. You’ve spent a lot of time on it, I’m sure, but you still mentioned certain challenges. Being your top challenge, you and your team must be on it. Where are you with solving them? Are they solvable problems, or are they just the nature of the beast?

Ashok Chennuru [00:56:48]:
I would say definitely solvable, but I’ll give you a simple example. When we were automating complex processes, both internally and end to end, for example authorization, that starts on the provider side to a handshake between the payer and provider and the internal payer processes, one of the challenges we had was, we automated, leveraging AI and GenAI and agentic, but when we operationalized this, we started to see a lot of fallouts, because then it became more relevant that, did we automate the right process? Did we invest enough time in process re-engineering and process simplification, versus trying to take the 15 steps that exist and try to enable AI on top of them? The lesson learned was, get the right process simplification, process experts both internally and with the providers, and simplify that and then enable AI on it and agentic, to drive better value. That’s still evolving, but the key message is, to my fellow panelists, have you come across scenarios like that? How much do you invest in the process aspect before you apply AI? How much do you invest in getting the data right before you apply AI?

Sanjog Aul [00:58:18]:
Have there been cases, Ashok, where, I know some examples in different countries where the telecom wasn’t as well established, so they just said, “Okay, we’re not going to even look at landline, we’ll just go to the mobile.” I’m just trying to draw a parallel here. So do you not have those creative destructive moments where, instead of trying to tinker with the process, you say, “Okay, put that aside, let’s reimagine”?

Ashok Chennuru [00:58:40]:
There are other examples, not in authorization, because it comes with regulatory challenges that may not be the best for complete reimagining. On consumer experience, because of ChatGPT Health, there’s a big complete-reimagine opportunity versus going to an app, so we see a few complete reimagines. But in this particular case, it still is transformational, but not complete, because of the ecosystem complexity, regulatory approvals, and so on. For example, if we are denying care, we need to go to every state Medicaid and get approval that AI is not involved in denying care.

Ashok Chennuru [00:59:26]:
So there are strong regulatory concerns, but that doesn’t mean we won’t overcome them.

Sanjog Aul [00:59:36]:
The way I understand your question is: is there a sweet spot? Is there a line you draw to say we should stop tinkering with the process, or trying to make it perfect, before we start putting technology on it? Is there a balancing act? That’s what you’d like to know. So, anyone in the group, do you have a magic formula for this, or if not, whatever you’ve tried? Go ahead.

Michelle Crovo [01:00:03]:
I could offer “it depends,” which is the traditional consulting answer. Where there are processes, regulations, or things in place that are out of your control, just state those as limitations and reimagine around it. I’ve done that, where we have some things that we’re going to put in this box that are not going to change, but how should we do these other things? When we go into an organization and they’re saying this process is really broken or clunky, trying to replicate it and try to make it better, faster with AI is usually not the right place to start. But in healthcare, in financial services, there are certain industries where there are things that are going to be out of your control. So maybe first having that look and saying, what’s within our control and what’s not? And then in the within bucket, maybe what’s easier to reimagine to start, because then you start to get into the stakeholder discussion of which 20 people do we need in the room to reimagine this and then get approval across the whole organization to actually make the change. So that’s what I would say, in my experience, where I’ve started. Then there might be some bigger ones where this is really bad and we have to reimagine it, and you do that in parallel. This is going to take longer and it has to get done, but how can you parallel path doing multiple things where you do one in a more simplistic way? And then, if it’s not broke, don’t fix it. If generally there aren’t issues with one, it’s about making it maybe faster or more efficient.

Michelle Crovo [01:01:32]:
You don’t have to reimagine everything. I often find going to new organizations, it’s “this is the time to just start over,” and sometimes that may be true, but often there are things you can put in the bucket of “this works pretty well, let’s just leave it as is for right now and iterate on it.” There’s a whole middle bucket of, a lot’s in our control, what can we do differently? And then there’s a whole different bucket of, we can make some meaningful change here, but there are certain things we’re not going to bang our heads against the wall trying to change.

Chris Helzerman [01:02:01]:
As Michelle was talking about some of the challenges in driving change in areas that can be a struggle, I was reflecting on my past life around consumer experience and patient engagement. We would often run into the conversations with the regulatory law department about what was possible and what wasn’t possible, and I’ve learned not to ask the yes or no question. Not “can we do this?” or “is this allowed or not allowed?” but rather, I engage the law department in the problem-solving activity. Here’s the business problem and opportunity, and how might we deliver an experience like this? How do we do this in a way that mitigates or lowers risk for the organization? So not “can I do it?” but “can you help me think of ways to do this in a way that mitigates risk for the organization?” Engaging regulatory law department team members in problem solving can help go a long way in creatively surfacing solutions that meet them where they are and help provide a better ultimate experience for the patients and members.

Sanjog Aul [01:03:29]:
Chris, now back to you about your top challenge, for which you want answers from the group.

Chris Helzerman [01:03:35]:
Earlier I surfaced the challenge of where to focus. I see our clients really enamored and excited about AI, and I’ve seen a number of instances of trepidation: “This feels like a big investment, and I don’t know whether I’m going to get the return on it.” I’ve seen organizations come to us and say that what they want to do is lower costs and administrative burden, which is sensible given the financial situation in healthcare today. But when you click down on it, the real costs are the utilization expenses that are not being managed. I’m thinking about care management in this particular case. An organization came to us and said, “We want to cut down on the amount of money we’re spending on outreach.” The problem wasn’t the amount of money they were spending on outreach. It was that only 50% of the patients who needed to be contacted for care management were actually being reached, and they were leaving tens of millions of dollars of utilization opportunity on the table.

Chris Helzerman [01:04:58]:
So, thinking out loud, how are organizations, and what collectively are people finding, works to help determine where to focus and accelerate momentum and drive success around various different initiatives? What are the techniques that you’re finding that work for helping organizations zero in on those top opportunities?

Sanjay Gupta [01:05:37]:
Chris, from our perspective, we broke the opportunities into three buckets. The first was, what can we do from a patient experience standpoint? The second was, what can we do to advance clinical outcomes? And the third was to drive greater financial impact. By having that financial impact bucket, we were able to make sure that the portfolio generated a collective return, so that people didn’t start questioning why we were spending any time on patient experience or clinical advancement in light of the headwinds I spoke about earlier. The nice thing is that we are advancing on all three. So what we did is, in each of those, we were not trading off: should I make an investment in virtual primary care with something I’m trying to run in revenue cycle? It was like, we are making advancement there. Joan brought up ambient listening on clinical. We did that, but then we added to that the ability to provide help for the inbox, so they’re spending less time, “pajama time,” as we call it. We made an investment on, can AI proactively start reaching patients to provide proactive care for lung cancer screening, or whatever the example is, which we are doing. By having a clear north star for each of those three buckets, and goals for all three, we weren’t having to compete one against the other, and the entire portfolio looked really good.

Sanjay Gupta [01:07:07]:
We could turn around and say we’ve improved patient experience, we’ve improved clinical outcomes, and we’ve driven a very strong bottom-line impact.

Robbie Freeman [01:07:18]:
I’ll add to that, because we have a similar approach to Sanjay. We look at value across quality, experience, and financial impact, and measure that across the portfolio. One thing that’s been helpful is that we let anyone in the company send an idea, which we can triage, review, and then prioritize. So we do a lot of listening. I like to say we’re top-down and bottom-up. We do a lot of focus groups with patients, and many of our most successful digital products came from feedback we heard just by listening to our patients.

Robbie Freeman [01:07:54]:
One example is what we call Check Symptoms and Get Care, based on feedback from patients that when they start having symptoms, they’re not always sure where to go. Do they call their PCP? Do they go to an urgent care? We’re using AI to match their symptoms with evidence-based guidelines, and within that experience, with one click, they can see a virtual care provider if appropriate, or book a same-day appointment. That all came from doing focus groups, and we’ve done the same with our employees. Any employee can submit an idea.

Robbie Freeman [01:08:30]:
One use case I love: about a year and a half ago, one of our nurses came to us saying, can we use AI to figure out which patients end up developing a pressure injury or bedsore? We went through our governance prioritization, we used the RICE framework and it rose to the top, and so we sat down and co-developed a product that’s now live across our health system. We did our own software development. It’s adding a lot of value. It’s improving quality and safety, and there’s a big financial impact when we can reduce those pressure injuries.

Robbie Freeman [01:09:00]:
So I think that top-down and bottoms-up, and measuring across the portfolio, has been just incredibly helpful.

Sanjog Aul [01:09:08]:
Mouli, your challenges, and what the team can do to help you.

Mouli Venkatesan [01:09:14]:
As I said in the beginning, as part of being Blue Cross Blue Shield Association and working in a federated way across 30-plus Blue plans, we have the privilege and honor to have 120 million Americans supported in our ecosystem, touching one in three Americans. The most important component in that is the payer-provider collaboration and the friction points that it creates. In that equation specifically, and I’ve got a lot of provider friends on the call as well, we’d love to see how, using agentic AI, we can go after that today. You take claims, you take prior authorizations, you take even benefit questions. All of that goes in paper, in phone calls, in people sending emails from the other side, through process, through workflows. Can we agentically solve that today? Because for me, solving this at scale is important.

Mouli Venkatesan [01:10:15]:
So we’re looking for partners that can help to scale this. If people have done that in a way, not just through an API or a third-party clearinghouse that’s sitting in the middle, but purely in an agentic way, can we go after solving that? Because otherwise there’s a lot of middle-layer players in the market, in the country, who are coming and sitting in between payers and providers, saying, “I got it, we will solve for both of you.” I wanted to look at it in a different way. So that’s my problem statement. How can we partner to go after that? How can we collaborate at scale? I’ve got friends here on the call with Novant, with Mount Sinai, which are all great North Carolina and New York-based plans, which we are so privileged to have as part of our network in terms of sending our members and their patients there. I’d love to get thoughts and feedback from the group.

Sanjay Gupta [01:11:13]:
Mouli, I’ll take you up on the offer. Come down in winter, when you can’t bike in Minnesota, and we’ll let you bike around Charlotte too. I would love to sit down and do that, and in fact take it a step further, where we’re not just automating the existing process. There’s no benefit in you having AI to deny a claim and me having AI to appeal that denial. Let’s actually solve it so that we take out the friction altogether.

Mouli Venkatesan [01:11:40]:
Fantastic. I’ll definitely take you up on it. We’ll not just bike, we’ll come and sit down together on this one. Sanjay, I appreciate it.

Sanjog Aul [01:11:48]:
So it’s not a leave-it-to-God problem. There is hope there.

Mouli Venkatesan [01:11:52]:
I genuinely believe there should be hope on this one. I wish the innovations happened at the core of healthcare, but unfortunately, innovations are happening at the seams of healthcare, where it’s broken, not at the core of healthcare. That’s why I wanted to directly appeal to the partners here, where we can actually fix the core, so that we don’t have so many middle-layer people in healthcare, which is all adding up to the cost for a consumer like you and me.

Michelle Crovo [01:12:21]:
I was just going to share that we have worked with clients that have done this at scale. That’s one of the ones I’ve been talking about today, because we’ve gone through the pains of trying to figure out how to make it happen. What I will also say is, if you all partner up together on how you do the payer-provider connection, platforms and other organizations will sign up to help do POCs to figure this out with you. So also think about that, because, again, we’ve done this at scale, meaning Cognizant, with many different platforms, and as I said at the beginning of the call, it’s not typically the platform or the model that’s not working. It’s, does it work for the use case and the parameters that you gave it? So I think it would be an awesome next step if, as Mouli and Sanjay wanted to partner on that, what platform or other solution providers can you go to to help you solve that problem? Because it’s a problem for the entire industry, whether it’s a smaller health system or one of the largest payers in the country. So I just wanted to put that out there as a tip: I guarantee you a platform like Chris and team, and others, would be happy to do a POC to help solve that problem, because it’s good for you and it’s good for them as well.

Chris Helzerman [01:13:40]:
Michelle, thanks for that tee-up. With 200 health systems and dozens and dozens of health plans on our platform, every day we’re thinking, “If we could only streamline this interaction here or there.” Health plans are receiving interactions from providers, and providers are spending time trying to track down information. There’s friction there that can definitely be smoothed over through agent-to-agent solutions. The provider-plan interactions are prevalent and self-evident, but I’m also seeing this in adjacent spaces in the ecosystem. Take a life sciences company that needs to check eligibility in order to enroll somebody in a copay discount or copay assistance program.

Chris Helzerman [01:14:42]:
They’re having to pay an incredible amount for labor to manually track down that information, and there’s a tremendous amount of opportunity in that space as well. Pharmacy, life sciences: it’s definitely there with providers, but there are other areas in the ecosystem where we can start to think about agent-to-agent and more efficient, streamlined interactions between participants in the healthcare ecosystem.

Sanjog Aul [01:15:15]:
Great. Before we close, I see that both Chris and Michelle were taking notes. Michelle, starting with you, any quick takeaways that enlightened us together, and which you could take home and do something with?

Michelle Crovo [01:15:32]:
Yes. I love the energy today. I love the focus on patient experience, the different demographic and the areas that you have to think about, how maybe some of the smaller health systems versus larger ones are thinking. I love being able to hear about all the different areas that AI has been implemented. I sit in the customer service space, so I may lean immediately to patient services or call centers, but hearing about how it can help actually in doctors’ offices, in hospitals, just all the different ways. I think the biggest thing to take away is being comfortable going back to leaders like yourself, as I continue on in my career, to push and point out the problem finders and the problem solvers, to your point, Mouli. There were a handful of things that you all said that made me think this doesn’t have to be as hard, and getting the leaders on board with sending that message and making sure we’re getting the right people in the room and holding them accountable to that success is what will ultimately drive adoption and make things successful.

Michelle Crovo [01:16:38]:
I really appreciated the opportunity to sit in and listen to your experiences and your ideas on what’s working well.

Chris Helzerman [01:16:47]:
I echo that. Really appreciate the opportunity for the conversation today. It was really enlightening, and I think what rises to the top here is we’re not just approaching this from the standpoint of conducting work or executing and putting solutions in place. There’s passion here around improving the outcomes, the lives of our patients and members, improving the experiences of our team members, and there’s a collective opportunity for us to all lean in and accelerate progress in a number of key areas within the space and in lowering costs. So I really look forward to ongoing conversations around how we can partner and enable your success and accelerate progress, be it with AI or just in general in terms of streamlining processes and improving experiences. There’s no shortage of opportunity.

Chris Helzerman [01:17:53]:
It is a very exciting time in healthcare.

Sanjog Aul [01:17:57]:
What a great conversation. Thank you so much, all of you: Sanjay, Robbie, Michelle, Joan, Ashok, Chris, and Mouli. Thank you again for sharing your insights. I believe what we are going through is a transformation with agentic AI in the center, but then if you stay focused on the business side and then bring agentic and make it behave, looking at all the different animals that we spoke about and using those analogies to get better at this, that is what the best outcome would be. Thank you so much again, and special thanks to Genesys and Cognizant for making this conversation possible. Thank you again, and have a great day.