Latest / Becker’s Healthcare Podcast / How CVS Health Is Using Agentic AI Twins to Transform Patient Experience

Transcript

How CVS Health Is Using Agentic AI Twins to Transform Patient Experience · Becker’s Healthcare Podcast

• 0:00
  Everyone. This is Jacob Emerson with the Becker's
• 0:02
  Health IT podcast.
• 0:04
  Thrilled today to be joined by CVS Health's
• 0:06
  vice president of enterprise customer experience and insights.
• 0:10
  Sri, thank you so much for taking the
• 0:12
  time to be with me on the podcast
• 0:13
  today. Pat, thanks, Jacob. Appreciate the time. Excited
• 0:16
  to be here with you. Likewise. And before
• 0:18
  we dive into everything we wanna talk with
• 0:20
  you about, some interesting new tech that CVS
• 0:22
  has been rolling out, can you first tell
• 0:24
  our audience a little bit more about yourself,
• 0:26
  your background in health care, and a little
• 0:28
  bit about your role at CVS today?
• 0:31
  Yeah. So,
• 0:32
  as you mentioned, I lead, customer experience and
• 0:35
  insights for CVS Health. I I started my
• 0:37
  career actually as an economist.
• 0:39
  So,
• 0:40
  looking more at, you know, markets and evaluating
• 0:43
  them for competitive dynamics and
• 0:46
  defending mergers and and acquisitions to department of
• 0:49
  justice.
• 0:50
  And, you know, that's where I kinda first
• 0:51
  got exposed to health care. I used to
• 0:53
  look at things like hospital mergers and and,
• 0:55
  you know, medical device mergers, and I got
• 0:57
  really fascinated by health care. And then my
• 1:00
  career kinda took a a bit of a
• 1:01
  winding path. I worked at, you know, GE,
• 1:04
  including GE Healthcare,
• 1:06
  Did spend some time at Wells Fargo,
• 1:08
  you know, in the banking industry, so got
• 1:10
  a little bit more exposure to the to
• 1:11
  the regulated banking industry
• 1:13
  before joining here, at CVS Health. And my
• 1:16
  role at CVS Health
• 1:18
  is I look at all of the consumer
• 1:20
  feedback we get, whether that's through surveys
• 1:24
  or,
• 1:25
  social media data, digital session data,
• 1:28
  call transcripts, whatever we can get our hands
• 1:30
  on
• 1:31
  to understand,
• 1:32
  what are some of the patient client, you
• 1:35
  know, member pain points that we have across
• 1:38
  our entire enterprise, so retail at
• 1:41
  health care delivery, Caremark.
• 1:43
  And how do we improve the experiences for
• 1:45
  those stakeholders and make sure that we're delivering
• 1:48
  best in class experiences
• 1:50
  so that those people can get the health
• 1:51
  care they need.
• 1:53
  When I when I kinda give my team
• 1:54
  a tagline, I say our job is to
• 1:56
  learn from the past, improve the present, and
• 1:58
  predict the future,
• 1:59
  particularly with respect for our members who we
• 2:02
  serve as the internal advocate for
• 2:04
  across all of our our internal teams, whether
• 2:07
  that's in technology or the business or elsewhere.
• 2:10
  Understood. So your background is more in the
• 2:12
  business and finance sector. Now you're here at
• 2:15
  CVS,
• 2:17
  modeling what patients and and customers
• 2:19
  are doing. And and you're here with us
• 2:21
  today, Shree, to talk a little bit about,
• 2:24
  something really interesting that we I I haven't
• 2:26
  necessarily heard a lot from other health care
• 2:28
  organizations, at least publicly.
• 2:31
  CVS, since late last year, has been rolling
• 2:33
  out Agentic AI twin simulations.
• 2:37
  So talk to our audience about what that
• 2:39
  is. When you talk about
• 2:41
  agentic twins,
• 2:43
  what does that actually mean, and and what
• 2:44
  does that actually look like in practice
• 2:46
  for CVS specifically?
• 2:48
  Yeah. And I bet on this podcast, you
• 2:50
  talked to a lot of leaders who talk
• 2:51
  about being consumer centric. Right? And I think
• 2:53
  that's something that
• 2:55
  everyone says. You probably see it in every
• 2:57
  single 10 k
• 2:58
  out there, every CEO,
• 3:00
  dialogue. I I think what's hard is being
• 3:03
  consumer centric is really difficult. Getting the voice
• 3:05
  of that customer into all of the aspects
• 3:08
  of what you do
• 3:09
  is challenging because it's hard to get to
• 3:11
  customers. It's hard to talk to them. It's
• 3:12
  hard for you to test things with them
• 3:15
  or and learn from them.
• 3:17
  You're limited in terms of how you can
• 3:18
  outreach these folks. You know, you can leverage
• 3:20
  surveys or maybe the scale.
• 3:22
  You know, you have to do pilots.
• 3:24
  And
• 3:25
  it's being consumer centric is challenging. You have
• 3:28
  that challenge simultaneously
• 3:30
  with
• 3:31
  the advent of AI and all the advancements
• 3:33
  you're having in AI.
• 3:35
  And what we realized in in working with
• 3:38
  these, you know, company from Stanford founded by
• 3:41
  a bunch of PhDs and computer science from
• 3:43
  Stanford
• 3:44
  is that we can actually build
• 3:47
  agentic twins of our customers or or patients
• 3:50
  or clients
• 3:51
  where we can actually
• 3:52
  build the way they make decisions.
• 3:55
  So not just what they did in the
• 3:56
  past, but how do they actually make decisions?
• 3:58
  How do they actually,
• 4:00
  handle trade offs? What are the behaviors that
• 4:03
  they they would exhibit in a given situation?
• 4:06
  And then we can generalize that to new
• 4:09
  scenarios and new situations. So if you do
• 4:12
  that at scale, and in our case, something
• 4:14
  like a 100,000
• 4:16
  plus
• 4:17
  Agenctic Twins,
• 4:19
  you will have a set of customers with
• 4:21
  you all the time
• 4:22
  that are, you know, effectively amplifying the consumer
• 4:25
  voice.
• 4:26
  So they are with you from when you
• 4:28
  develop the product, when you wanna test the
• 4:31
  outcome, if you wanna test some operational change.
• 4:34
  And they it creates that real consumer centricity
• 4:38
  because
• 4:39
  that representation of the consumer, how they think,
• 4:41
  how they behave is with you throughout the
• 4:43
  entire process.
• 4:45
  So, you know, it's the way we've gone
• 4:47
  about doing that is we ingest a lot
• 4:49
  of different information. Of course, all consumer consented,
• 4:53
  but we take things like a a short
• 4:55
  behavioral interview, behavioral signals,
• 4:58
  spend data, things like that where we can
• 5:00
  build a twin
• 5:02
  of each individual consumer. And, again, you do
• 5:04
  that at scale
• 5:06
  where you actually have their thought process with
• 5:08
  you, and this agentic twin of the customer
• 5:12
  lives on and is always on,
• 5:15
  for you to investigate and challenge new concepts,
• 5:18
  new ideas, and to actually even perform dress
• 5:20
  rehearsals of new things you're trying to launch
• 5:22
  out. And it's really helped our decision making
• 5:26
  our understanding of the consumer throughout the whole
• 5:28
  product life cycle,
• 5:30
  and and really just it's brought them to
• 5:31
  life in a way that I don't think
• 5:33
  was possible in the past.
• 5:35
  And now, we really we really do. It's
• 5:37
  kinda putting your money where your mouth is
• 5:39
  in terms of centricity.
• 5:40
  You're having these customers with you across the
• 5:42
  entire journey.
• 5:44
  Wow. I mean, this is fascinating. You you
• 5:46
  literally copy and pasted
• 5:48
  your customer base, and and now you can
• 5:50
  see how they interact with all your different
• 5:52
  businesses.
• 5:54
  So
• 5:56
  back backing up here, Sri, why did CVS
• 5:58
  wanna go down this route? How why do
• 6:00
  you wanna leverage this kind of technology?
• 6:02
  And what does it actually do problem wise
• 6:05
  for your patients and for your customers?
• 6:08
  How does it solve issues for them on
• 6:10
  the ground?
• 6:11
  Yeah. So I think the reason you would
• 6:13
  wanna we wanted to undertake this is we
• 6:15
  really do take consumer centricity seriously. I mean,
• 6:18
  we wanna build products and services
• 6:20
  that help people access health care in an
• 6:23
  effective way
• 6:25
  and really solve the big challenges with The
• 6:27
  US health care system around, you know, simplicity,
• 6:30
  you know, ease of access, all those things
• 6:32
  that you hear about constantly from from patients.
• 6:35
  And I think where we we're we're landing
• 6:37
  is we have the ability to get to
• 6:39
  these folks via traditional methods, whether that's panel
• 6:42
  research or surveys
• 6:43
  or even to a degree looking at operational
• 6:45
  data. But each one of those things had
• 6:48
  challenges. Right? Either it was
• 6:50
  speed, you know, or fidelity or any of
• 6:52
  those things that
• 6:54
  that challenge you in terms of getting to
• 6:56
  that group. So when we heard about this
• 6:57
  methodology and this approach, it was really exciting
• 7:00
  because
• 7:01
  it would it would really unlock what we've
• 7:03
  been trying to do, which is get the
• 7:05
  patient
• 7:06
  front and center on the way we build
• 7:08
  things. And I think if you if you
• 7:10
  look at what we've seen so far, the
• 7:12
  four main things we've gotten out of these
• 7:14
  twins,
• 7:15
  just in terms of the value of how
• 7:17
  we build, is one speed.
• 7:19
  You go from weeks to months to generate
• 7:22
  insights. So if you have to go get
• 7:23
  a panel and test something, whether that's messaging
• 7:26
  or some product or service,
• 7:28
  it takes time. You gotta go find those
• 7:30
  patients. You have to go interview them. You
• 7:32
  know, you have to you have to distill
• 7:34
  those insights.
• 7:35
  Now we're talking about fifteen to thirty minutes.
• 7:38
  And, you know, I'll tell you honestly, Jacob,
• 7:39
  the first handful of times we did this,
• 7:41
  I was stunned. Like, things that would take
• 7:42
  us eight weeks, we were doing in fifteen
• 7:45
  to twenty minutes.
• 7:47
  The second is fidelity.
• 7:49
  You know, you you can only test so
• 7:51
  many things with a human being. Right? There's
• 7:53
  only so many
• 7:54
  scenarios you can put in front of them
• 7:55
  before their brain sort of breaks. They'll get
• 7:58
  fatigue. You know, the signal quality reduces.
• 8:01
  They're not able to answer. With these AI
• 8:03
  representations,
• 8:04
  you can test thousands of real world scenarios,
• 8:07
  and you're not extrapolating out what they're gonna
• 8:09
  what they're gonna do in a given situation.
• 8:12
  You can present them with tons and thousands
• 8:14
  of different scenarios or situations
• 8:16
  and see how they'll actually behave. And that
• 8:19
  actually also solves a big problem in research
• 8:21
  where you have what's called stated intent.
• 8:23
  So what someone tells you and what they'll
• 8:25
  actually do. And here with this scenario testing,
• 8:28
  just because you've generalized how someone makes a
• 8:30
  decision or how they think,
• 8:32
  you can actually get to the actual behavior
• 8:35
  and even understand that gap between stated intent
• 8:38
  and actual behavior.
• 8:39
  The scale here, you have all these folks
• 8:41
  on
• 8:42
  all the time, so you don't have to
• 8:43
  generalize the small pilots.
• 8:45
  You know, you have thousands,
• 8:47
  tens of thousands, and so some cases for
• 8:49
  the populations we looked at, agentic twins available
• 8:52
  to us at all times. So that scale
• 8:54
  of the consumer voice is is tremendous
• 8:57
  and now available to us to access,
• 8:59
  you know, as we go forward. And then
• 9:01
  the final component is the reach. Like, these
• 9:04
  hard to reach populations
• 9:06
  that we often have,
• 9:07
  you know, whether that's particular disease states or
• 9:10
  conditions where it's just difficult to find those
• 9:13
  people or that can be really costly.
• 9:15
  Now we only have to really find them
• 9:17
  once because then we can create this twin
• 9:19
  of that person, and then they'll be with
• 9:21
  us the whole time. And it goes back
• 9:22
  to, I think,
• 9:24
  amplifying that consumer voice with that hard to
• 9:26
  reach population particularly because
• 9:28
  that's a group that's, you know, often
• 9:31
  marginalized because we can't get to that group,
• 9:34
  to understand what they're experiencing.
• 9:35
  So if you if you wrap all that
• 9:37
  up, we're getting away from reacting
• 9:40
  to small groups
• 9:42
  and with long timelines to get their feedback,
• 9:45
  to predicting behavior, acting almost in near real
• 9:48
  time, and having these consumers with us along
• 9:52
  the entire journey,
• 9:53
  it's transformative. It's it's changing the way we
• 9:56
  deliver care because we're able to understand
• 9:59
  what's gonna actually impact our consumers in a
• 10:01
  real way while we're building these products
• 10:05
  and not having to do things like, you
• 10:07
  know, really narrow pilots and constant testing.
• 10:10
  We can go out with a lot more
• 10:11
  confidence in what we're doing
• 10:13
  because we've been able to dress, rehearse it
• 10:15
  with, you know, tens of thousands of customers
• 10:17
  before we actually hit the market, and and
• 10:18
  anyone has to actually experience what we're building.
• 10:22
  Absolutely. I mean, listening to you talk about
• 10:24
  this is incredible. It it it sounds transformative.
• 10:27
  I I I've done marketing initiatives myself in
• 10:30
  terms of doing those panels like you said
• 10:32
  that take eight weeks and boiling that down
• 10:35
  to fifteen minutes is that's incredible.
• 10:38
  Not to mention what to hook on the
• 10:40
  last part of your answer, the
• 10:42
  the equity of it all of being able
• 10:43
  to reach your entire customer base and know
• 10:46
  how they all respond,
• 10:48
  and not having to put everybody into pockets.
• 10:50
  It's it's amazing. I I can't imagine what's
• 10:52
  gonna be coming from this. And so I
• 10:54
  think that's that's the natural follow-up question here,
• 10:56
  Sri, is how does this change the way
• 10:58
  that CVS
• 11:00
  makes decisions compared to those traditional approaches you
• 11:03
  just laid out for us even just from
• 11:05
  a marketing perspective. But maybe you could expand
• 11:07
  a bit on in terms of clinical decisions,
• 11:10
  other areas of the pharmacist of the of
• 11:13
  the industry that the company reaches.
• 11:16
  How does this change the day to day
• 11:18
  for CVS and and the business overall?
• 11:20
  Yeah. And I I can give you a
• 11:21
  couple of just concrete examples of that we've
• 11:24
  that we've tried so far, that we've seen
• 11:26
  success with. I think one, you know, one
• 11:28
  classic one for us in the pharmacy side
• 11:30
  is adherence.
• 11:32
  That's an area we we've got a tremendous
• 11:34
  amount of focus. Obviously, we want people to
• 11:36
  be adherent to their medication. We want them
• 11:38
  to stay healthy.
• 11:39
  Accessing medications
• 11:41
  is really important to
• 11:42
  us. That can be a a difficult group
• 11:44
  to talk to, particularly nonadherent patients. There's a
• 11:47
  lot of reasons people don't, you know, adhere
• 11:49
  to medication. Some of them related to stigma,
• 11:52
  fear.
• 11:53
  And you're in a sense, if you're going
• 11:55
  to talk to that group,
• 11:56
  you're making them relive whatever that was the
• 11:59
  challenge that they had. And that can be
• 12:01
  really difficult that you it can lead to
• 12:04
  false signals. It can lead to, you know,
• 12:06
  different types of challenges in terms of getting
• 12:08
  to the population,
• 12:10
  fatigue, lack of sample, those sorts of things
• 12:12
  because they're just a difficult group to reach.
• 12:15
  What we were able to do with the
• 12:16
  twins is we were able to actually go
• 12:18
  through and and investigate in detail adherence issues.
• 12:22
  So get because you don't have to face
• 12:24
  that and it's you know, you don't have
• 12:26
  that same sort of challenge
• 12:28
  with the Gentic twins because they're AI representations,
• 12:31
  we were able to get way deeper than
• 12:33
  I think we could have in a standard
• 12:35
  qualitative or even quantitative interview.
• 12:37
  What we're we were able to figure out
• 12:39
  is a lot of the things about adherence
• 12:41
  were the same that you would expect that
• 12:42
  we've always seen.
• 12:44
  You know, it's,
• 12:45
  you know, fear of side effects and fear
• 12:48
  of, you know, you know, not being able
• 12:50
  to talk to a clinician and all those
• 12:51
  things that we've been solving,
• 12:53
  over the, you know, course of the past
• 12:55
  few years.
• 12:56
  But a new one's emerged, and it was
• 12:58
  really around
• 12:59
  the fear of getting a refill, which sounds
• 13:02
  really you know, it sounds a little bit
• 13:04
  strange. It wasn't what we expected, but we
• 13:06
  were able to see, like, you know, a
• 13:08
  lot of what we were observing was people
• 13:10
  were saying, okay. Now once I start a
• 13:12
  medication, how do I make sure I'm always
• 13:13
  gonna get it? What happens if I can't
• 13:15
  get that medication?
• 13:17
  You know, does that is that gonna have
• 13:18
  outcomes for me? And all those things that
• 13:20
  we were able to see that that were
• 13:22
  clear in the data.
• 13:23
  And I think what what it pointed to
• 13:25
  is that we have a lot of best
• 13:26
  in class digital tools we're building and can
• 13:28
  have built
• 13:30
  that are making accessing medications
• 13:32
  much easier. So it gave us a lot
• 13:34
  more confidence
• 13:35
  in those tools we're building. And in addition,
• 13:37
  we're able to actually look at how do
• 13:39
  we optimize some of those tools to ensure
• 13:41
  that,
• 13:42
  you know, we're we are making it as
• 13:44
  easy as possible on the refill.
• 13:46
  And I I think what it really pointed
• 13:48
  out is
• 13:49
  this anxiety, and it's patient anxiety that you
• 13:52
  were able
• 13:53
  that you were able to see in in
• 13:54
  the agentic twins around all the process related
• 13:57
  to the prescription. So it wasn't just the
• 13:59
  medication.
• 14:01
  It was the process
• 14:02
  and the experience
• 14:03
  that led to adherence. And, you know, we
• 14:05
  were able to back that up with a
• 14:06
  lot of the actual data we were able
• 14:08
  to we were collecting from the customers.
• 14:11
  When I say actual, I mean, operational data
• 14:13
  that we were collecting from the customers.
• 14:16
  It it it was a it's a it
• 14:17
  was a way for us to unlock something
• 14:19
  that we didn't we didn't see in the
• 14:21
  past.
• 14:22
  And I think what's particularly unique about it
• 14:24
  is we're talking about effectively about robots here,
• 14:26
  Jacob. We build robots that represent customers,
• 14:30
  but they were unlocking
• 14:31
  something that's as that's superhuman
• 14:34
  and anxiety.
• 14:35
  And so we were able to get to
• 14:37
  the emotion through these agentic twins, which I
• 14:39
  think is a really fascinating
• 14:42
  use case for us.
• 14:44
  And then just a a quick related one
• 14:45
  just because you mentioned marketing. Another one was
• 14:48
  related to our PetRx.
• 14:50
  I'm gonna just tell a story really quickly.
• 14:52
  You know, PetRx is our our you know,
• 14:54
  we're looking to to move into pet care,
• 14:56
  and we're looking at how do we message
• 14:58
  to those to those, pet parents, if you
• 15:00
  will. I'm a pet parent. I've got a
• 15:01
  dog.
• 15:02
  And, you know, there was a lot of
• 15:04
  the data we were expecting, like emotional data
• 15:06
  that people care a lot about their pets.
• 15:08
  They become, like, family members.
• 15:10
  One of the more interesting things we saw
• 15:12
  from the Agenctic twins was
• 15:14
  people were saying, you know, that even giving
• 15:16
  medication to a pet was an act of
• 15:18
  devotion, not necessarily a chore.
• 15:21
  So any messaging appealing to the emotions was
• 15:23
  gonna be successful.
• 15:24
  But what the the agentic twins allowed us
• 15:26
  to do is present scenarios
• 15:28
  beyond that. And what we found was it
• 15:30
  wasn't enough
• 15:32
  to just be
• 15:34
  connecting emotionally. You had to show that we
• 15:36
  had the capability to actually get the the
• 15:39
  scripts from the veterinarians,
• 15:41
  have an understanding what those scripts are, how
• 15:44
  the side effects work. That was really important
• 15:46
  to the pet parents. So
• 15:48
  by being able to drill in and get
• 15:50
  to the kind of stated versus actual actual
• 15:52
  behavior using the twins, we were able to
• 15:54
  uncover
• 15:56
  that the messaging needed to have both. It
• 15:58
  wasn't sufficient to just have messaging and marketing
• 16:00
  that was appealing to emotion. We had to
• 16:02
  appeal to the tactical side too, because pet
• 16:05
  owners are incredibly pragmatic despite this emotional connection
• 16:09
  that I think we all observe and see.
• 16:11
  So, anyway, those are two kinda quick examples
• 16:13
  of where we've we've seen early wins.
• 16:16
  And, you know, we're operating this at scale,
• 16:18
  so there's more to come, I think, here
• 16:20
  as we keep moving forward.
• 16:22
  Oh, absolutely. I I can imagine so. And
• 16:24
  for pretty much any company listening in right
• 16:26
  now, I would imagine they are taking notes
• 16:28
  on how to replicate something like this, Sri,
• 16:31
  because the the possibilities seem endless just listening
• 16:34
  to even those those two examples you shared,
• 16:36
  and I really like how you framed it
• 16:38
  of unlocking something that is human. You're you're
• 16:42
  unlocking something that the humans on the other
• 16:44
  side maybe hadn't been able to before, which
• 16:46
  is really fascinating.
• 16:48
  It's a really unique AI use case. You
• 16:50
  know? It's not it's in a sense, it's,
• 16:53
  it's not one of the ones that I
• 16:54
  think are traditional,
• 16:56
  and it's it's using AI to understand human
• 16:58
  behavior,
• 16:59
  which, again, sounds a little counterintuitive.
• 17:01
  But I think No. It makes sense. Yeah.
• 17:03
  Yeah. It's really creating this always on approach
• 17:06
  with our customers. Definitely. Definitely. Is there anything
• 17:09
  you'd add for for the health care audience
• 17:11
  that's listening in terms of the value prop
• 17:14
  here for the wider health care system?
• 17:17
  Yeah. I think, you know, if you if
• 17:20
  I were thinking about more broadly,
• 17:22
  a EO, even things like patient experience
• 17:24
  and and I'll use the experience tag because
• 17:26
  that's that's where I, you know, tend to
• 17:28
  focus.
• 17:29
  It it's
• 17:31
  ability to test messaging. Like, how do we
• 17:34
  change in particular in health care is really
• 17:36
  painful. That's observed everywhere.
• 17:38
  You can see that across multiple populations when
• 17:41
  you change people's
• 17:42
  medications, when you change formularies, when you change
• 17:45
  the way they access care, it's painful.
• 17:48
  But what this allows you to do is
• 17:49
  test different messaging and different approaches to see
• 17:52
  how you how do you smooth that, how
• 17:54
  do you make that easier on the patient.
• 17:56
  And you're able to do that here in
• 17:58
  a controlled, safe environment
• 18:00
  that you you are able to see the
• 18:02
  results before you actually have to go into
• 18:04
  the market and test that. I think that's
• 18:05
  a really big unlock for health care.
• 18:08
  The the other one, I think, is understanding
• 18:10
  journeys. Health care journeys are incredibly complex,
• 18:13
  and, you know, it's it can be difficult
• 18:15
  to actually get patient feedback along every aspect
• 18:18
  of that journey.
• 18:19
  But here, you can you can do that,
• 18:21
  and you can see how people are experiencing
• 18:24
  every aspect of where they engage with you.
• 18:27
  And I think understanding quickly where you have
• 18:29
  pain points and where you may need to
• 18:31
  do some work and may where you're excelling
• 18:33
  and maybe where you need to double down.
• 18:35
  Health care in particular to me,
• 18:38
  this is a way to
• 18:40
  test and learn
• 18:42
  quickly,
• 18:43
  with those audiences.
• 18:45
  And I think it's it's a major unlock
• 18:48
  because I I think that's where and I
• 18:49
  know I keep using the word unlock because
• 18:51
  I think in in these populations,
• 18:54
  traditionally, it's really difficult to get the feedback,
• 18:56
  and they're also really difficult to pilot with.
• 18:59
  Because if you're running tests, you're actually running
• 19:01
  tests on someone that's experiencing something,
• 19:04
  or trying to access care. And that's not
• 19:06
  always easy, and that's not always right, and
• 19:08
  it's it can be difficult.
• 19:09
  Here, you're able to do those tests in
• 19:11
  this controlled environment without having to worry about
• 19:13
  the patient risk or outcome. It makes it
• 19:16
  safer, and it and it makes it more
• 19:17
  controlled.
• 19:19
  And I think, you know, the follow on
• 19:21
  effect is you're gonna deliver better care,
• 19:24
  for those patients and for those stakeholders.
• 19:26
  Sure. Let me ask you a a bit
• 19:28
  of a prodding question.
• 19:30
  Just as this technology
• 19:32
  continues to develop, we'll continue to see it
• 19:34
  across many companies,
• 19:36
  different industries.
• 19:38
  How are you thinking about how CVS wants
• 19:41
  to
• 19:41
  ensure responsible use
• 19:44
  of of agentic twins at enterprise scale? You
• 19:46
  know, you're able to now test an an
• 19:49
  external pressure
• 19:50
  on on your customer
• 19:52
  base. How do you balance,
• 19:55
  some of the potential negatives of that or
• 19:57
  or ensure that there's gonna be human judgment,
• 19:59
  trust,
• 20:00
  good governance of this technology as it begins
• 20:03
  to as already has started to influence,
• 20:06
  real health care on the ground?
• 20:08
  Yeah. It's a really important question.
• 20:10
  I I think one, we do. We have
• 20:12
  very strong controls around this, and there's a
• 20:14
  couple different ways we do it.
• 20:16
  We continuously test
• 20:18
  these twins,
• 20:19
  against not only our own data,
• 20:22
  but third party studies
• 20:24
  to ensure that we have the highest quality
• 20:26
  information.
• 20:27
  And, you know, one of the things that
• 20:28
  we're doing that these twins do as well,
• 20:31
  is they actually live in the world, so
• 20:32
  they ingest current events.
• 20:34
  You know, if you are a, if you
• 20:36
  have a a twin that's a democrat, they'll
• 20:38
  read democratic news organizations. If you have one
• 20:41
  that's a republican, they'll read republican news organizations.
• 20:44
  So you're able to actually keep them up
• 20:45
  to speed,
• 20:47
  you know, even on current events. But
• 20:49
  regardless of that, we're always back testing. We're
• 20:51
  always making sure that they're meeting our standards
• 20:54
  of of
• 20:55
  matching known,
• 20:57
  human results,
• 20:59
  that we have. The second is, you know,
• 21:01
  right now, we have my team of researchers.
• 21:04
  I have a team of,
• 21:05
  you know, analysts
• 21:07
  and market researchers that we have organizationally that
• 21:10
  are are the ones who are administering this
• 21:12
  test because they represent human behavior.
• 21:14
  So you can lead the witness. So if
• 21:16
  you don't have someone, you know, at the
• 21:18
  controls who's engaging with effectively the way it
• 21:21
  works is an agentic chat
• 21:23
  and you're not asking the questions appropriately, you
• 21:26
  can get results that are misleading. So right
• 21:28
  now, it's my team of researchers
• 21:30
  that are are looking at that information and
• 21:32
  validating it, and making sure it's meeting the
• 21:35
  standards that we would have for human studies.
• 21:38
  But one of the things I point out
• 21:39
  there, though, is not to live in an
• 21:40
  echo chamber. On occasion,
• 21:42
  the Agenent twins will teach us something new
• 21:44
  we didn't know, and we shouldn't just be
• 21:45
  sitting there saying,
• 21:47
  oh, wait. It it didn't match exactly what
• 21:49
  we thought in the past, so, therefore, it's
• 21:50
  wrong. What we need to do is investigate
• 21:53
  it further. So we make sure we have
• 21:54
  that process,
• 21:56
  built out.
• 21:58
  And then I think the third thing and
• 21:59
  probably most importantly, I feel like everyone always
• 22:01
  frames AI as replacement.
• 22:04
  And as like, what is this gonna replace?
• 22:05
  It's always the first question you get when
• 22:07
  you're deploying an AI solution. I don't think
• 22:09
  this is replacing talking to our customers.
• 22:12
  It's augmenting our understanding.
• 22:14
  It's amplifying their voice, as I've said.
• 22:17
  But, ultimately, we're gonna have to always talk
• 22:19
  to our customers because that's important, and that's
• 22:21
  what good companies do. You need to listen
• 22:23
  to people, and you need to understand what
• 22:24
  their friction and their pains are. And then,
• 22:26
  also, it's important to do that to potentially
• 22:28
  catch emerging issues and things like that. So
• 22:31
  I view this as additive to what we're
• 22:33
  doing with our human populations.
• 22:36
  And I think if you you combine those
• 22:37
  three things,
• 22:39
  you know, the the the back testing,
• 22:42
  the researcher at the wheel, and then always
• 22:44
  talking to the human populations as well,
• 22:46
  it's a major it's a it's a set
• 22:48
  of governance that gives you a lot more
• 22:50
  confidence in how the twins are operating.
• 22:53
  And it also helps you amplify the impact,
• 22:56
  because when you're going out to the various
• 22:57
  stakeholders
• 22:58
  and you're doing the change management of saying,
• 23:00
  hey. Listen to listen to these agentic twins.
• 23:02
  They're telling us, you know, what to do.
• 23:05
  The people you're working with have more confidence
• 23:07
  in you because they know that you're taking
• 23:08
  the the care
• 23:10
  and you're doing the work to make sure
• 23:12
  these things are accurate.
• 23:13
  Absolutely. And like you said, Shari, I mean,
• 23:15
  this is it sounds like this is unlocking
• 23:17
  things that your customers were already feeling and
• 23:19
  thinking.
• 23:20
  Anyway, this is now just another way for
• 23:23
  you to for you to reach to reach
• 23:24
  that. So really, really fascinating.
• 23:27
  Anything else we're missing here? Any final thoughts
• 23:30
  you wanna share with our audience?
• 23:33
  Yeah. I think, you know, at at CVS
• 23:35
  Health, it's not just that we're experimenting with
• 23:38
  this because I think a lot of people
• 23:39
  will frame this as it it's it is
• 23:41
  new. And I I get it's an emerging
• 23:43
  technology, but we were at the forefront here.
• 23:45
  And I think it's gone beyond experimentation.
• 23:48
  You know, we're we're at this point. We're
• 23:50
  at operating at scale. As I mentioned,
• 23:52
  a 100,000
• 23:54
  twins. You know, my hope is to get
• 23:56
  us to 300,000
• 23:57
  or more in the in the future, but
• 23:59
  we're using this responsibly at that scale to
• 24:02
  improve people's lives.
• 24:03
  And this is giving our teams unprecedented
• 24:06
  clarity
• 24:07
  and foresight
• 24:08
  into, you know, what's happening in the industry,
• 24:10
  what's happening with our consumers, how can we
• 24:13
  better serve them, how can we build
• 24:15
  better experiences? How can we learn faster,
• 24:18
  and design better?
• 24:20
  It's it's helping us it's a step change
• 24:23
  for us in terms of how we actually
• 24:25
  deliver,
• 24:26
  health care.
• 24:27
  And I think in the end, it's gonna
• 24:29
  help us deliver better health and support for
• 24:31
  every customer member,
• 24:33
  that we serve, and we're gonna do it
• 24:35
  with more confidence, empathy, and understanding.
• 24:37
  And it's gonna help us achieve our goal
• 24:39
  of being the most consumer centric company in
• 24:41
  health care. So,
• 24:43
  yeah, I I hope everyone listening out there,
• 24:45
  particularly those in health care,
• 24:48
  understand that
• 24:49
  this is a,
• 24:50
  this is a a a tool
• 24:53
  that I think can really give your customers
• 24:55
  an unprecedented voice in the way you operate.
• 24:59
  100%. I mean, this does seem to be
• 25:01
  the one of the largest
• 25:03
  commercial rollouts we've seen of this technology or
• 25:05
  that we've been, we've heard about in health
• 25:07
  care on our end. So, Sri, really appreciate
• 25:10
  you taking the time to chat with us
• 25:12
  about this, and then please keep us updated
• 25:15
  as as this all evolves. We'd love to
• 25:17
  hear more about it. Yeah. Always a pleasure,
• 25:19
  talking about this, and it was a pleasure
• 25:21
  talking with you. So excited to come back
• 25:23
  and tell you all the great things we're
• 25:24
  doing in the future.
• 25:25
  And to our audience, if you'd like to
• 25:27
  listen to more podcasts from Becker's Healthcare, you
• 25:29
  can visit beckershospitalreview.com.