Delivering More Connected Care Through AI: CVS Health's Josh Weiner
Me, Myself, and AI · 27 May 2025
Source: https://sloanreview.mit.edu/audio/delivering-more-connected-care-through-ai-cvs-healths-josh-weiner
Transcript: https://podcasts.opennash.com/shows/me-myself-and-ai/delivering-more-connected-care-through-ai-cvs-healths-josh-weiner-abb690/ (via OpenNash Podcasts, machine-generated)

Organizations can use AI to improve healthcare by making it more personalized, connected, preventative, and accessible. Find out how one leader thinks about these four dimensions on today's episode.

I'm Josh Weiner from CVS Health, and you're listening to Me, Myself, and AI.

Welcome to Me, Myself, and AI, a podcast on artificial intelligence and business. Each episode, we introduce you to someone innovating with AI. I'm Sam Ransbotham, professor of analytics at Boston College. I'm also the AI and business strategy guest editor at MIT Sloan Management Review.

And I'm Shervin Khodabandeh, senior partner with BCG and one of the leaders of our AI business. Together, MIT Sloan Management Review and BCG have been researching and publishing on AI since 2017, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and to deploy and scale AI capabilities and really transform the way organizations operate. Hi everyone, today Sam and I are speaking with Josh Weiner, senior vice president of consumer engagement and analytics at CVS Health.

Hi Josh, welcome to the program.

Thanks, Shervin. Pleasure to be here.

I'm sure most of our listeners know what CVS is, but maybe you could give us an overview of the company and your role there.

Happy to do so. Let me actually start with healthcare. Healthcare is deeply personal to me. I was born with a rare genetic condition, and from an early age up until now, I've dealt firsthand with challenges. Whether it's cost or authorizations or this premise of navigation and figuring out what's covered. And trying to repeat 35 years of medical history to every new clinician I see and wishing there was just a summary of that or a way to share that.

So for those reasons, I've decided to dedicate my career to improving healthcare, and my personal passion is affordability of healthcare. CVS is one of the largest integrated healthcare companies in the US, with subsidiaries that include Aetna, a health insurer, Caremark, pharmacy benefits manager, the largest retail pharmacy, our core CVS pharmacy, as well as increasingly some care delivery businesses. You may have seen in the news Oak Street Health, Signify. And with respect to my role, as you said, I'm the senior vice president of consumer engagement and analytics.

My priority is developing experiences for our consumers to help them improve their health, and increasingly, we're using AI to revolutionize those experiences. And I believe that's a focus for today's discussion.

Wonderful. So maybe let's just dive right in. What are some examples of the kind of experiences that you're trying to create? And what is AI's role there?

Great question. As a structure, I'm optimistic about AI's opportunity to revolutionize consumer healthcare experiences. And there are four areas I'm focused on more personalized care, more connected care, more preventative care, and more accessible care. And within each of those, we have priorities that we're confident will make a big difference for you and I and our consumers. When I think about some of the most exciting examples, several come to mind, but I'll start with our development of personal health agents.

Historically, the opportunity to truly develop personalized care plans for consumers has been impossible. I mean, the unfortunate reality is, for a given person, based on a condition, and if you're lucky, your gender, your age and some understanding of behavioral preferences, you're assigned a care plan. And that care plan doesn't really update depending on how well you're progressing. And when you get another checkup in a year or so, maybe you review, and then maybe at that point you'll make some changes.

And that's about the best we could do before AI. The reality is there's just too many people that we need to create some clusters. We need to go ahead and use humans to develop care plans and move forward. AI has changed this. With AI, we can now make the most of our longitudinal set of unstructured health data, medical claims, electronic medical records, labs, as well as wearable data. And use agents to say, here's what a personalized goal might look like, based on how you're actually progressing on a day-to-day basis.

Here's an updated goal. And from streaming those health data sets, identify emerging gaps in care or opportunities to help achieve your health goals. And hopefully, more likely gets where you want to go from a healthcare perspective. The role of CVS is that of what is typically characterized as care management or clinical support pro programs, and it's something that either a pharmacist from the pharmacy business or a nurse from one of the health plans typically provides. And they are not prescribing medication or performing surgeries.

Their role is to help you close gaps in care, understand and interpret clinician care plans, and help you achieve your goals. Often with telephonic outreach on an ongoing basis to check in on how things are going. Areas like managing side effects, adhering to your medication, or understanding and scheduling the importance of preventative care visits, whether it's diagnostics or screenings, fall within the scope of the clinical teams in CVS Health's umbrella. Now, historically, it has been difficult to share those experiences and insights with provider groups, the largest systems.

We have integrations with their EMRs, and there is interoperability and it's seamless. And they know about your interactions with the care management team, and we know about your interactions with clinicians. But as you well know, we aren't where we hoped we would be from an interoperability perspective. This is actually another area that AI is transforming because we can work around different systems and different data models and have agents talk to each other to facilitate data sharing without spending three years and a lot of money trying to integrate legacy systems,

And when you say agents, you mean AI agents?

right? AI agents. Also, I would say our role is on the prevention side. You're typically seeing doctors and entering the health system when you have a problem and for management of that problem. But up until that point and after that point, that's where we come in. Because there's a variety of things you should be doing to keep yourself healthy, taking medications, moving, sleeping a certain way, getting the right number of screenings, vaccinations and so forth. That's where we've typically come in with humans.

It's been difficult to scale, and AI presents an opportunity not only to do it for everyone, but to also develop more personalized care plans.

You've mentioned personalized a couple of times, and that really resonates with me because I think we all have our own anecdotal stories. But as I think about this, I look at charts, let's say radiology reads a chart and they describe what they see there. I think that they're speaking in English, but I'm not sure because they use so many big, complicated words that I spend, you know, 20 min going through word by word, trying to understand and google things to see what they mean.

And part of that is, I think the radiologist may be writing for my physician, but I'm reading it too. What you described is pretty appealing to me, the idea that that this agent could talk to me, like, like, what I know, versus what everyone else in the system knows.

Translation is one of the most beautiful and successful use cases we've seen to date, and translation comes in a couple different forms. There's the version you're describing, where we just ingest language that was intended for the clinical community, whether it's CPT codes or prior authorization criteria, and we translate it to something that makes sense for a consumer. You know, it's almost an x-ray of what's being shared through the EHRs for you in plain English. There's another version of that that really excites me as well, and that's how we communicate with our Medicaid consumers.

Historically, it has been extremely difficult to serve these communities well. These are other great examples where we're today using AI to translate to provide more personalized care and more accessible care. And it's particularly for the folks that tend to be more underserved.

And I can see the argument the other direction that if you're talking to me about statistics, I think you can hopefully talk at a little bit higher level. And you don't need to say what an average means, or what a what a distribution is, or I hope not. But we need that coded information for the clinical community to rapidly understand things in the system. But a lot of that depends on the idea of this data sharing and this universal access to data.

I think I've heard this for the last decade or more about how we're going to get there. Where are we in this? Is this tomorrow? Is this yesterday? Is this infinitely always two weeks away? How are we going to get there?

I think AI is the unlock we've been waiting for. Let's talk about what we're doing within CVS to start. We have all these subsidiaries with different clinical platforms, different data structures. So for us, developing even an internal customer 360 has had the issues you've just described. Oh, we gotta figure out how to integrate these systems, we have to build common data structures, and it's truthfully extremely difficult, you know, it's written in different languages and so forth.

AI agents present an opportunity to interpret unstructured data and to share the data across applications that are programmed in different languages, with entirely different architectures. In a way that was previously extremely costly, and that's a hypothesis, but we're able to do that today internally to finally create consumer 360s for our patients. And it's reasonable to me that using the same set of technologies, we can begin to actually leapfrog the interoperability challenges of the broader healthcare system.

I agree with that because we're also seeing it in other industries as well. The challenge of data management and connecting different data domains that have different names and conventions in disparate data sources has always been a manual challenge. Because of the nature of those data sources, which has always been structured data that has to fit in tables in a certain way. And what you're saying is, and by the way, what was the purpose of that, to have weaker analytic models?

Use it? What you're saying is the models are so strong now that they can do a lot with data without the structure. In fact, the structure actually confuses them to to some extent. And I hope that we see this at scale. That might be the unlock you're talking about.

And AI is able to do what was very difficult for humans to do in figuring out how to integrate different types of data structures across different architectures. We have a whole NoSQL vector database of electronic medical records. It was previously impossible to interpret, impossible to track the data, structure it and share it. Now you can just throw LLMs on top of it and it works. And we can use prompt engineering to figure out how to convert it and share it with all different types of recipients.

Josh, you talked about four priorities and we talked about personalization. Do you want to cover a couple more?

Sure, so I think we also talked about more connected care indirectly. We certainly feel it at CVS, where our mission is to create better integrated experiences, and we haven't been able to solve that problem. We just spoke about how, through AI, we're delivering more connected care. But what's also fascinating, what I think will ultimately have the most profound effect on health care, is accessibility. And this is a bit farther out, but worth discussing. Just last week, I saw, I think, now the third study in mental health where they've demonstrated that an AI therapist through text can perform at a level of clinical parity to that of humans.

Today, the behavioral health landscape is really challenging. There's far more demand than supply. If you're lucky, you can get an appointment in four months, and even then you may be told, Hey, we don't really take your insurance. And there's this large out of pocket. I expect in the next two or three years, everyone can have their own AI therapist in their pocket, available to them at all times and incredibly affordable. And that's not today, that's hopefully tomorrow, but that's also part of our research agenda and what really excites me.

What I think is beautiful about that is the idea that, you know, I would hope that the best therapists are better than than next week's AI agent. And I think that's going to be true for a while. But you know, the way these technologies tend to work is that it meets an underserved market. You know, there's a market of people who are not getting access, who, let's say, wait four months or maybe wait infinite months to get access to therapy.

And this idea that you can connect these people through this technology to give them some minimum level of service is pretty appealing. Because right now, I think the competition is not, you know, I think you mentioned on par with human. I think the argument right now is that the competition is on par with nothing. So all we have to beat is this non-inferiority trial here. We don't, we don't have to beat the human, we just have to beat nothing at this point.

Or giving humans superpowers, which are enabling them to serve more patients and community members. I had previously referenced how we employ nurses and our nurses work with you to build care plans, help you adhere to care plans. Through the use of AI, we expect them to be able to serve. And then you add on things like messaging and digital channels and technologies. Maybe you can even get more leverage from a panel perspective. So we don't need to, re, we're not going to replace you anytime soon, but with superpowers, we can make you more effective.

You can help more people. It's more accessible, it's cheaper. The other thing we we haven't talked about yet, and I would be remiss not to mention, is the importance of empathy and trust. As we build these experiences to be able to actually revolutionize health care and transform, you know, our trajectory, and that's something we're being really thoughtful about. We're putting in quite a bit of process and governance to make sure that we're extremely rigorous in how we test, how we co-design these experiences, how we measure accuracy and effectiveness.

The starting point is back office and enablement, and it's only over time as we demonstrate success. And as we've made clear that this can be trusted, and we're continuously looking to help you and do what's right for you, that we can progress to these broader, more transformative use cases,

Well, let me push back a little bit on that because you mentioned an ecosystem of people that involves pharmacy and pharmaceutical companies and hospitals and insurers and a bunch of these people. And if they're all developing an agent, and those agents are going to maybe work more for that company than for me, the individual, how do we get the individual voice represented in this? Who's going to take that action? Because I don't think a lot of individuals are going to make their own agents.

right? I guess I see the future a little differently. I expect that Sam, you'll have a couple agents, predominantly, maybe from some of the major consumer platforms and operating systems you work with today. And I want to make sure that those agents have the intelligence of the agents we've built. I don't really expect you to be using your health insurance agent on a daily basis. That's just not a daily need. But I want to make sure Siri, or whomever it is, has all that intelligence and power.

And our focus is going to be how we integrate our agents with your real agent who knows you and is really kind of looking out for you with your priorities. I see it a bit more the way you think about APIs. Over the long term, it'll take a little while to get there. There has been advancement in those agent agent protocols, but I don't think any of us want a world where there's just, you know, 20 different agents from every single company I interact with that are trying to build relationships with me.

I think that seems really important, and actually, I really like the way you frame that. Because what you're saying is if all these agents can talk to each other, then we can let the invisible hand of the economy straighten this out. If company A builds a duplicitous agent that is always recommending the expensive, you know, drug or whatever, I'm just making up stuff at this point, but always going down one path, or it clearly has some sort of vested interest.

Then what you're arguing is that if we allow agents to talk to each other, then we can let the invisible hand work on that. That's actually pretty appealing to me.

Well, those duplicitous actions don't need automation and AI to exist, right? I mean, they can very well exist right now, with human human agents doing

well, I deeply suspect they do. And you know, I like the idea of what you're saying, that this allows this transparency and governance to to show some of that that's probably going on right now.

Josh, let's, uh, switch gears a little bit. Speaking of agents, I know you're a chess aficionado, and I'm guessing you're using AI somehow in advancing your own knowledge of chess. Is that true? Or are you doing anything in that domain?

Funny, you mention it. So I have been using my love of chess as a way to stay on top of advancements in the space. I've been in the process of building out an AI chess coach using some of these tools like Cursor or GitHub Copilot and so forth. Being good at chess, and certainly being good at teaching someone how to play chess, requires planning and reasoning. And it's not just machine learning or math. And certainly, there's stock fish in your incredible chess algorithm that always tells you the scientifically perfect answer.

But that's not useful in actually helping someone become a better chess player. It's not teaching them how to be better. So from what I know for the time being, the breakthrough in chess coaching is gonna have to wait a little bit for the next breakthrough in more generalized AI and AI that has more reasoning skills. But for the time being, for me, using AI to build an app when I haven't seriously written code in about a decade.

And I've been able to now roll out an app that's 40, zero lines of code across five languages, just using AI with one person to do something that might have otherwise taken five people six months. That's beautiful in of itself.

How long did it take you?

30 hours.

There you go? Wonderful, yeah. That

Incredible, right?

Josh, it's been fascinating talking. I think there's a lot of opportunity here, and we've touched on a lot of that opportunity. But you also mentioned how our barrier or our thresholds may be a bit off. We can make a lot of progress, but I think we're gonna have to be patient in this process. Thanks for joining us today and thanks for taking the time to talk to us.

Thanks for listening today. On our next episode, Sam and I speak with Julian Adams, president and CEO of Stand Up to Cancer. We hope you'll join us.

Thanks for listening to Me, Myself, and AI. Our show is able to continue in large part due to listener support. Your streams and downloads make a big difference. If you have a moment, please consider leaving us an Apple Podcast review or a rating on Spotify and share our show with others you think might find it interesting and helpful.