Latest / Becker’s Healthcare Podcast / Nirav Shah, Associate Chief Medical Informatics Officer, AI and Innovation, Endeavor Health

Transcript

Nirav Shah, Associate Chief Medical Informatics Officer, AI and Innovation, Endeavor Health · Becker’s Healthcare Podcast

• 0:02
  Every year, Becker's annual meeting brings health care
• 0:04
  leaders together to unpack the most pressing issues
• 0:07
  facing the industry.
• 0:08
  And every year, those conversations shift in profound
• 0:11
  and unexpected ways.
• 0:13
  This April, more than 3,500
• 0:15
  healthcare executives will return to Chicago for Becker's
• 0:18
  sixteenth Annual Meeting.
• 0:20
  Seven ninety five elite speakers will offer new
• 0:23
  lessons, new case studies, and predictions about what
• 0:26
  comes next.
• 0:27
  Join us April 13 through the sixteenth.
• 0:30
  For the agenda and event details, visit beckershospitalreview.com
• 0:34
  and click on the events tab in the
• 0:36
  upper right.
• 0:38
  This is Scott Becker with the Becker's Healthcare
• 0:41
  Podcast.
• 0:42
  We're thrilled today to be joined by a
• 0:44
  brilliant physician leader.
• 0:47
  We're joined today by doctor Nurev Shah, and
• 0:50
  doctor Shah serves as the associate
• 0:52
  chief medical informatics officer
• 0:55
  of AI and innovation.
• 0:56
  He's also the director of investigational
• 0:59
  innovation
• 1:00
  in infectious disease physician, a clinical physician assistant
• 1:04
  professor.
• 1:05
  He's works currently for the last decade plus
• 1:08
  at Endeavor Health. He's also spent time at
• 1:10
  University of Chicago and Northwestern,
• 1:13
  a brilliant, brilliant physician leader. Doctor Shah, can
• 1:16
  you take a moment to tell us about
• 1:18
  yourself and about your your about your career?
• 1:21
  Thank you, Scott, for having me on this
• 1:23
  podcast.
• 1:24
  And, I mean, that was a great introduction.
• 1:26
  So,
• 1:27
  my name is Nirav Shah, an infectious disease
• 1:30
  physician by training. And like you mentioned, I
• 1:32
  serve as Endeavor Health's associate CMIO for AI
• 1:35
  and innovation.
• 1:36
  So as you can imagine, this role is
• 1:38
  quite multifaceted,
• 1:40
  and I belong to a variety of different
• 1:41
  teams, including innovation,
• 1:43
  information technology,
• 1:45
  clinical slash quality, and research teams, which are
• 1:47
  all critical domains of this work.
• 1:50
  My passion really lies in leveraging emerging technology
• 1:54
  with a focus on AI to solve complex
• 1:56
  clinical and administrative problems, and I work on
• 1:59
  the entire AI continuum. So you can think
• 2:01
  about it as strategy and governance
• 2:04
  to sourcing new tools,
• 2:06
  implementation,
• 2:07
  validation,
• 2:08
  scaling,
• 2:09
  and producing
• 2:10
  either white papers for internal consumption or academic
• 2:14
  output for generalizability
• 2:16
  on this work. So a little bit about
• 2:18
  Endeavor Health. So we're an integrated delivery network
• 2:21
  in Illinois.
• 2:22
  We serve over 1,400,000
• 2:24
  patients or a third of the Illinois catchment
• 2:26
  area. We have nine acute care hospitals, 300
• 2:30
  sites of care, and generate around 6,000,000,000 in
• 2:32
  revenue, just to give you an idea for
• 2:34
  the size.
• 2:35
  And we're recognized as one of the top
• 2:37
  25 health systems for AI readiness in The
• 2:39
  US, and we're actively leveraging greater than a
• 2:42
  100 AI and automation tools to enhance patient
• 2:45
  care, improve administrative and operational efficiency, and reduce
• 2:48
  clinician burnout.
• 2:50
  Thank you. And and, doctor, talk a little
• 2:52
  bit about
• 2:53
  when you think about where AI is having
• 2:55
  impact,
• 2:56
  where are the I mean, it seems like
• 2:58
  it's very much a cascading thing where it's
• 3:00
  having different impacts in different areas.
• 3:02
  Where are you seeing the most useful or
• 3:05
  exciting use cases
• 3:06
  yet yet endeavor that you or that you
• 3:08
  see yourself, that you sort of look and
• 3:10
  say, that's really working and making a difference.
• 3:12
  Where are you most excited, and what are
• 3:14
  you seeing the best use cases so far?
• 3:17
  Think so AI is a tool that we're
• 3:20
  leveraging across a variety of different areas.
• 3:23
  So I think there is
• 3:25
  opportunity to see value,
• 3:28
  in the clinician space primarily in
• 3:31
  documentation. So that that is a solution that
• 3:34
  has spread like wildfire. I don't think there
• 3:36
  has been a technology solution
• 3:38
  that has had as fast as adoption
• 3:41
  as in the ambient space.
• 3:43
  And we're starting to think about other ways
• 3:46
  that you can use generative AI in the
• 3:48
  clinical space such as summarization,
• 3:50
  other types of generation,
• 3:53
  and thinking about clinical decision support. So the
• 3:56
  value in those isn't,
• 3:58
  as proven out as what we're seeing in
• 4:00
  Ambiance, but there's a lot
• 4:02
  of literature in areas where we're gonna start
• 4:04
  seeing how that is,
• 4:06
  moving forward kind of,
• 4:09
  the clinical space and reducing burnout and improving
• 4:12
  patient care.
• 4:13
  On the administrative and corporate side, I think
• 4:17
  there is a lot of value that's being
• 4:19
  generated.
• 4:20
  And I'll start with the corporate side. So
• 4:21
  I think health care is probably a little
• 4:23
  bit later to the game,
• 4:25
  compared to where other sectors are using kind
• 4:28
  of AI. And I think we have a
• 4:30
  lot that we can learn from how it's
• 4:31
  being deployed in tech and other kind of
• 4:34
  regulated
• 4:34
  industries such as financials and others.
• 4:37
  So there's opportunity
• 4:39
  in, you know, marketing and HR and a
• 4:41
  variety of, you know, like, legal and compliance
• 4:43
  and supply chain. So there's a variety of
• 4:45
  areas
• 4:46
  where you can see that AI potentially has
• 4:49
  the opportunity to make operations more efficient.
• 4:53
  So I think that's interesting. And,
• 4:56
  a variety of,
• 4:57
  different places that you can you know, the
• 4:59
  way that I think about AI from from
• 5:02
  a generative AI standpoint is, in order
• 5:04
  to transform an organization, I think about it
• 5:06
  in a continuum. You can have productivity
• 5:09
  hacks. So for example, when you are using,
• 5:12
  say, chat, GPT, or Gemini, or any of
• 5:15
  these things, You know, you can use it
• 5:17
  to basically summarize
• 5:18
  essentially an email or, you know, generate an
• 5:21
  email, summarize a document, a variety of different
• 5:23
  things like that.
• 5:24
  Then if you move to the next stage,
• 5:26
  it's really kind of process
• 5:28
  modifications.
• 5:29
  If you think about a more complex process
• 5:31
  than just writing an email, you know, a
• 5:33
  specific workflow or something like that, there's a
• 5:35
  lot of steps that are there. And so
• 5:38
  we're seeing that you can implement generative AI
• 5:41
  in a variety of different steps to kind
• 5:43
  of speed up how quickly that process moves
• 5:45
  forward.
• 5:46
  And the final is really AI transformation
• 5:49
  and AI first approach, which is taking a
• 5:52
  whole process, blowing it up, and getting to
• 5:54
  the outcome using novel technology.
• 5:57
  So something like that is where I think
• 5:59
  we're going to start to see a lot
• 6:00
  of value,
• 6:01
  in the administrative operational,
• 6:04
  areas and then also the clinical space. So
• 6:06
  that's probably gonna be, you know, the story
• 6:08
  that we start seeing in '26 and beyond.
• 6:11
  Thank you. When when you look at the
• 6:13
  next steps in this and what you're doing,
• 6:16
  doctor Shah, what are you most excited about
• 6:19
  as we move to this next year?
• 6:21
  Yeah. I I mean so as I think
• 6:24
  about
• 6:25
  where we're going, you know, I like to
• 6:26
  think about
• 6:28
  kind of from where we have approached this
• 6:31
  at Endeavor Health. Like, what have we done
• 6:33
  in the last few months, and where are
• 6:35
  we going in the future? So what
• 6:37
  I'm most proud about, for what we've done
• 6:39
  is we've started to build the cultural and
• 6:42
  structural infrastructure
• 6:43
  to kind of get started.
• 6:45
  It's easy to buy a piece of software.
• 6:47
  It's much harder to align massive newly consolidated
• 6:51
  health system
• 6:52
  around a shared vision for the future. What
• 6:54
  this is, it is it allows us from
• 6:56
  moving from doing AI to being an AI
• 6:59
  enabled organization.
• 7:00
  So where my focus and excitement is going
• 7:02
  forward is really moving us from being an
• 7:04
  AI enabled organization,
• 7:06
  just trying to keep up with AI to
• 7:07
  building the institutional architecture to shape our destiny
• 7:10
  and lead it. So essentially moving from a
• 7:13
  system that uses technology to a system that
• 7:15
  is defined by its intelligence. So I wanna
• 7:18
  kinda highlight four key areas to kind of
• 7:20
  illustrate this. So the first is really,
• 7:23
  socializing
• 7:24
  AI from the bedside to the boardroom.
• 7:27
  So innovation fails in a vacuum and bringing
• 7:30
  along the organization on this work is critical.
• 7:32
  So I'm working with,
• 7:34
  some of my colleagues and leaders, and we're
• 7:36
  making a concerted effort to socialize the why
• 7:39
  of AI at every level from the board
• 7:41
  and our CEO counsel
• 7:42
  to our frontline clinical and operational teams. So
• 7:45
  in a large health system, everyone is focused
• 7:47
  on the friction of daily operations,
• 7:49
  and we've been successful,
• 7:51
  in a in creating enthusiasm amongst our team
• 7:54
  members, showing them that AI isn't just one
• 7:56
  more task that they need to do, but
• 7:57
  a potential solution to some of their pain
• 7:59
  points.
• 8:00
  So where I'd like to go
• 8:02
  is moving from this top down socialization
• 8:05
  to a frontline led innovation model,
• 8:08
  where our goal is really to empower our
• 8:10
  clinical, operational, and corporate teams to not just
• 8:12
  use AI, but to articulate the complex problems
• 8:15
  it can solve. Shifting from quick productivity hacks,
• 8:18
  like I mentioned, to process modification, to an
• 8:21
  AI first redesign,
• 8:22
  which is kind of that continuum that, I
• 8:24
  was talking about. And we wanna empower our
• 8:26
  staff to blow up outdated processes and rebuild
• 8:29
  them around intelligence
• 8:31
  with the novel capabilities brought forward with AI.
• 8:34
  So and then at the leadership level, we
• 8:36
  are actively crafting a proposal for financial investment
• 8:39
  to accelerate our work and lead in the
• 8:41
  deployment of AI for community based health systems.
• 8:44
  The the second kind of area is we've
• 8:47
  launched our own internal Google Gemini platform,
• 8:50
  which has been a game changer for engagement,
• 8:53
  socialization.
• 8:55
  And, by giving our team members a safe,
• 8:57
  secure enterprise grade environment to experiment with, we're
• 9:00
  moving beyond theory into actual, like, use cases.
• 9:03
  So we're seeing real time enhancements in day
• 9:06
  to day productivity. We've kind of captured that
• 9:08
  in a variety of surveys and feedback. And
• 9:11
  what is happening is we're seeing a grassroots
• 9:14
  innovation culture, and we're starting to see some
• 9:16
  intractable systemic problems such as policy consolidation,
• 9:21
  that are getting solved. And that's a huge
• 9:23
  problem that, you know, I've kind of discussed
• 9:25
  in other,
• 9:26
  podcasts, which is quite interesting because it's a
• 9:29
  really challenging problem as you're getting the health
• 9:31
  system together,
• 9:33
  figuring out how to standardize,
• 9:36
  consolidate
• 9:37
  massive amounts of documentation. That's where kind of
• 9:39
  AI is really, really helpful and,
• 9:42
  perfectly tuned to do. So going
• 9:44
  forward from democratizing
• 9:47
  our Gemini platform, we're moving from single use
• 9:49
  tools to moving into the space of agentic
• 9:52
  orchestration. And so we have on our road
• 9:53
  map the build out of a central nervous
• 9:56
  system, so a unified infrastructure
• 9:59
  that will manage the growth of autonomous agents
• 10:01
  across our core platforms.
• 10:03
  And this, ensures that as we develop and
• 10:05
  deploy agents, we build the infrastructure to do
• 10:08
  so safely to handle critical problems within applications
• 10:11
  and then also across applications.
• 10:14
  The third kind of pillar is really building
• 10:16
  the structure and process around AI governance
• 10:19
  that enables AI transformation. So we've and what
• 10:22
  we've done in the last year is really
• 10:24
  centralizing our AI innovation among, admits con consolidation.
• 10:28
  So building AI structures are really difficult,
• 10:31
  if you talk to leaders across other health
• 10:34
  systems, and they frequently change.
• 10:37
  Doing so
• 10:38
  in normal circumstances is challenging, but it's really
• 10:41
  challenging when you're simultaneously consolidating
• 10:44
  EHR. So we're doing a three for one
• 10:46
  consolidation right now, infrastructure and clinical and operational
• 10:49
  structures
• 10:50
  as a result of, the mergers that we
• 10:52
  have, over the last few years. So I'm
• 10:54
  proud that we've managed to stand up by
• 10:56
  unified AI governance, ensuring that innovation is built
• 10:58
  into our new system identity.
• 11:01
  And going forward, we'll be launching a comprehensive
• 11:04
  center for trusted AI that builds on the
• 11:06
  centralized governance to include more domains and teams
• 11:09
  to centralize how we intake,
• 11:11
  validate, and monitor AI, how we upscale our
• 11:14
  teams, how we internally and externally communicate our
• 11:16
  AI work, and build bidirectional sophistication
• 11:19
  with state and federal government regulators
• 11:22
  and consumer perception of AI.
• 11:24
  And then finally, developing the change muscle of
• 11:26
  the organization,
• 11:27
  the ability to pivot and adopt new tech
• 11:29
  as fast as it evolves, and it's evolving
• 11:31
  very fast.
• 11:33
  Final pillar is really our AI validation schematic.
• 11:36
  This
• 11:38
  seems trivial, but it's actually so important.
• 11:40
  So we've developed a robust,
• 11:42
  repeatable,
• 11:43
  multimodal, and mixed methods validation schematic for AI
• 11:46
  tools.
• 11:48
  We don't just take a vendor's,
• 11:50
  word for it. We've deployed this framework on
• 11:52
  our most complex AI use cases that involve
• 11:55
  patient care. And so we're doing this to
• 11:57
  ensure that our models are clinically safe, operationally
• 12:00
  sound, and ethically transparent. And where I'd like
• 12:03
  to go in the future is to refine
• 12:05
  this, which in its current state is very
• 12:07
  robust and to make governance as efficient as
• 12:10
  it is effective.
• 12:11
  So I like to,
• 12:13
  I like this phrase that in health care,
• 12:16
  AI scales at the speed of trust.
• 12:18
  So far, clinicians and patients don't trust the
• 12:20
  output. The technology is useless and is not
• 12:22
  going to scale. So by building that transparent,
• 12:25
  efficient, repeatable, and fast moving governance engine, we're
• 12:28
  ensuring that Endeavor Health is ready to be
• 12:30
  a leader in this transformation.
• 12:32
  And and what resonates so well, doctor Shah,
• 12:35
  is this concept of the single use cases,
• 12:37
  like, so many of us use
• 12:39
  day to day or individual tasks on
• 12:43
  chat GPT or whatever platform anybody's working on
• 12:46
  to this next stage of doing this at
• 12:48
  scale and with systems.
• 12:50
  And I how much of a learning curve
• 12:53
  is there for people on that? Because it
• 12:55
  seems like so many of us have gotten
• 12:57
  comfortable
• 12:58
  with the single use cases, you know, the
• 13:00
  you know, whatever it might be. But but
• 13:03
  then moving towards systemic use cases and governance,
• 13:06
  talk a bit about the complication of that
• 13:09
  and
• 13:10
  and how hard is it to start seeing
• 13:12
  results and real results, and it does it
• 13:14
  it differ by area? And does it differ
• 13:16
  by team member or or
• 13:18
  or or or the teams that are leading
• 13:20
  different areas?
• 13:22
  I think
• 13:23
  this is probably the multibillion
• 13:25
  trillion dollar question
• 13:27
  is how do we take generative AI and
• 13:30
  move from what it is that we're doing
• 13:32
  right now to draft an email
• 13:34
  to transforming how we operate? And I'll give
• 13:36
  you the example. I kind of touched on
• 13:38
  it earlier with policy consolidation.
• 13:41
  So our health system,
• 13:43
  is a product of four individual
• 13:46
  health systems slash hospitals that have come together
• 13:48
  in the last five years,
• 13:50
  and we all bring,
• 13:51
  our own policies and procedures of how we
• 13:53
  do things.
• 13:55
  And that, equates to about 10,000
• 13:57
  distinct policies
• 13:59
  that we have across our system.
• 14:01
  And we're now building out a way to
• 14:04
  kind of redefine this into an endeavor way
• 14:08
  of how we do policies and procedures.
• 14:10
  And we had a single use case where
• 14:12
  we did it in a manual way where
• 14:15
  it took about 30 different,
• 14:17
  team members
• 14:18
  to consolidate a single policy, and it took
• 14:21
  about three months to do that. And variety
• 14:23
  of different meetings, in person,
• 14:26
  over Zoom, etcetera, a variety of different things.
• 14:28
  So you can imagine doing this for 10,000
• 14:30
  policies is really complicated.
• 14:32
  So I wanna kind of highlight these three
• 14:34
  different steps. So
• 14:36
  we initially, we were asked to kind of
• 14:38
  help solve this problem,
• 14:40
  from our chief clinical officer, and we took
• 14:43
  the approach of, let's just see if we
• 14:45
  can do the productivity hacks. Can we take
• 14:47
  three policies
• 14:48
  and just summarize them and consolidate them? A
• 14:51
  very basic kind of task that's, generative AI
• 14:54
  is very suited to do.
• 14:56
  We tried to do that. We realized that
• 14:58
  for this complex problem, there were certain
• 15:01
  policies
• 15:02
  that didn't completely equate to other policies. So
• 15:05
  if you for example, if we have a
• 15:06
  policy
• 15:07
  on, I don't know, catheter associated urinary tract
• 15:11
  infection prevention.
• 15:12
  So some policies would allow for the consolidation
• 15:15
  of it, but other ones would also mention
• 15:18
  other things that were completely distinct from that.
• 15:21
  And so that that baggage that we have
• 15:23
  from our how our health systems have defined
• 15:25
  these problems, you know, we were able to
• 15:26
  do some basic things, but we needed to
• 15:28
  take a, you know, a broader approach to
• 15:30
  that.
• 15:31
  So we spent time thinking, you know, how
• 15:32
  do we take the whole process
• 15:34
  and modify things that, you know, we could
• 15:37
  using generative AI? So we thought about you
• 15:39
  know, we have 10,000 policies.
• 15:41
  That's, like, 200
• 15:43
  domains. We have, like, 500 plus owners of
• 15:46
  this.
• 15:47
  How do we simplify,
• 15:48
  like, the larger process that we have? How
• 15:50
  do we reduce the dimensionality? How do we
• 15:52
  use generative AI to do that? So we
• 15:54
  took time to basically take the 10,000 policies,
• 15:58
  put them into 20 buckets, reduce the number
• 16:00
  of owners. Think about that. We use generative
• 16:03
  AI to create these buckets and do all
• 16:04
  these things. So we're figuring out how do
• 16:06
  we simplify the process where we can using
• 16:09
  generative AI. And then we tried to say,
• 16:12
  can we still figure out with this consolidated
• 16:14
  approach? Can we consolidate the policies using generative
• 16:17
  AI?
• 16:18
  Again, we really couldn't do that. And so
• 16:20
  we have this problem again where because of
• 16:23
  the,
• 16:24
  historical
• 16:25
  tendencies of how we built these policies across
• 16:27
  these different systems, generative AI was not suitable
• 16:30
  to kind of bring them together.
• 16:33
  So the final step and this this kind
• 16:35
  of took months for us to to figure
• 16:37
  out and to do.
• 16:38
  The final step is we realized that there
• 16:41
  is no way that we're gonna be able
• 16:42
  to do this.
• 16:43
  There are some policies that were procedures. There
• 16:46
  are some procedures that were policies. They're, you
• 16:48
  know, referencing different things, referencing old policies,
• 16:51
  a variety of different problems. So the final
• 16:54
  approach, the AI first approach is, you know
• 16:56
  what? We're gonna blow up the whole system,
• 16:58
  and we're gonna generate net new policies.
• 17:01
  And we're going to make sure that the
• 17:03
  prompt when we develop this is looking at
• 17:06
  what is best practice from the literature, so
• 17:08
  the evidence based practice,
• 17:10
  from societies that are, you know, providing thoughts
• 17:13
  around this, whether it's infectious disease society or
• 17:16
  OBI in society, you know, variety of different
• 17:18
  areas,
• 17:19
  looking at what are the policy drivers,
• 17:22
  what are the regulations
• 17:23
  at a state and, federal level that, you
• 17:26
  know, that would, shape these types of policies.
• 17:29
  And we put it all into one master
• 17:31
  prompt. And now what we're doing is when
• 17:33
  we,
• 17:34
  are redoing a policy,
• 17:36
  we are building a new policy from scratch
• 17:39
  so that our team members don't have to
• 17:41
  fight and try to figure out how you
• 17:43
  know, which policies do we wanna keep, and
• 17:44
  there's a lot of inertia to keep what
• 17:46
  you know from your existing legacy institution.
• 17:49
  And instead, we have a net new that
• 17:51
  is novel, that is new, that is up
• 17:53
  to date, that is tied to the most,
• 17:55
  relevant policy drivers.
• 17:57
  And they get that as a starting point
• 17:58
  and then that is speeding up the process.
• 18:01
  This so to get at your question, to
• 18:03
  do that for all of these net new
• 18:06
  processes
• 18:07
  takes a lot of time, and it takes
• 18:09
  a lot of you know, you have to
• 18:10
  be willing to blow up stuff and start
• 18:12
  fresh because we have a lot of,
• 18:14
  you know, processes that are really,
• 18:17
  complicated
• 18:18
  based on what we can do from a
• 18:20
  manual standpoint. So with this new technology, if
• 18:23
  you think about where are we starting from,
• 18:25
  where are we trying to go from, it
• 18:27
  enables you to really rethink how we do
• 18:29
  this, and that is where you'll see the
• 18:31
  value come forward.
• 18:33
  I mean, it is simply amazing. And and
• 18:35
  and doctor Shah, it it's such a fascinating
• 18:37
  cascade of how this moves from individual use
• 18:40
  cases to bigger use cases to really
• 18:43
  systematic and agentic.
• 18:45
  Talk a bit about
• 18:47
  what advice would you give to evolving leaders
• 18:50
  trying to work with AI? Where where do
• 18:52
  you start with on advice to to to
• 18:54
  evolving leaders in the AI world?
• 18:57
  Yeah. So I think three pieces of advice,
• 19:00
  for AI leaders or health systems in general
• 19:03
  that are navigating this transition.
• 19:06
  The first is really
• 19:07
  optimize the workflow and not the tool. The
• 19:10
  second is build change fitness into your culture,
• 19:13
  and the third is prioritize governance as a
• 19:15
  strategy and not a hurdle. So let me
• 19:17
  go a little bit more in-depth into this.
• 19:19
  So the the first is start with optimizing
• 19:22
  the workflow.
• 19:23
  The biggest mistake I see is leaders are
• 19:25
  looking for a magic software solution
• 19:29
  that they can bring in to solve a
• 19:31
  systemic problem. And,
• 19:33
  you know, as I've been saying, AI is
• 19:35
  not plug and play. It's really a process
• 19:37
  transformation catalyst that enables
• 19:40
  us to rethink the process and to transform
• 19:42
  completely with an AI first approach.
• 19:44
  So before you automate this process, you have
• 19:47
  to deeply understand it. And my advice is
• 19:49
  stop asking what can this AI do, and
• 19:51
  what's the most painful fifteen minutes of a
• 19:54
  team member's day, and how do we use
• 19:56
  intelligence to eliminate it, automate it, etcetera.
• 19:59
  And then that may sometimes require blowing up
• 20:02
  the whole process and starting it from fresh.
• 20:04
  And then iteratively keep doing that. What's the
• 20:07
  next fifteen minutes that are the most painful?
• 20:09
  And keep doing that to kind of,
• 20:11
  you know, make the make the day less
• 20:14
  painful.
• 20:15
  So the second issue is really change fitness.
• 20:18
  And so a differentiator will be an organization's
• 20:20
  ability to quickly adapt to change,
• 20:23
  and we know technology is moving really fast.
• 20:26
  And I think the big challenge that we
• 20:28
  face as health systems and just in general,
• 20:30
  you know, probably all sectors of the economy
• 20:33
  is our ability to absorb it. I'll give
• 20:35
  you an example. I think if you look
• 20:37
  at our EHRs, they have hundreds of solutions
• 20:40
  on their road map coming out, and it
• 20:42
  takes us a long time to implement a
• 20:45
  single solution.
• 20:46
  So we need to move away from top
• 20:48
  down mandates towards codesign and agile deployment.
• 20:51
  So we need to bring physicians, nurses, operations,
• 20:54
  corporate team members,
• 20:56
  you know, etcetera into the room on day
• 20:58
  one to help shape adoption, be able to
• 21:00
  test it, learn quickly, and scale thoughtfully,
• 21:03
  and that will be critical.
• 21:05
  And
• 21:06
  we need to also be able to pivot
• 21:08
  when a solution doesn't work because a lot
• 21:09
  of these solutions may not work. So that
• 21:11
  gets at the concept of, you know, change
• 21:13
  fitness, which is a combination of change management
• 21:16
  and then rapidly
• 21:17
  building that into
• 21:19
  the culture. And the third is really around
• 21:21
  governance. And I, you know, I can't stress
• 21:22
  this enough. You can't scale AI across a
• 21:25
  1,400,000
• 21:26
  patient system without rigorous multidisciplinary
• 21:29
  and multimodal framework
• 21:30
  that ensures accuracy, equity, transparency,
• 21:33
  and safety.
• 21:35
  And, again, AI will scale at the speed
• 21:37
  of trust. So putting in that investment to
• 21:39
  build that trust layer
• 21:41
  will allow for movement, like, 10 times faster
• 21:44
  later because the foundation
• 21:46
  is solid.
• 21:47
  So, in all I mean, all of these
• 21:49
  three things really the goal is to get
• 21:51
  to high touch. So we aren't just implementing
• 21:54
  algorithms. We're transforming.
• 21:56
  We're architecting a future where technology handles the
• 21:59
  data, and so we can handle
• 22:01
  care and relationships. So that's really the goal
• 22:04
  of, you know, all of these three different
• 22:05
  things and just in general, like, what we're
• 22:07
  trying to do with AI.
• 22:09
  Doctor Shah, thank you so much. I've I've
• 22:11
  had a chance to watch out upfront and
• 22:13
  close the Endeavor Health System. I live in
• 22:15
  the area. It's amazing the transformation of the
• 22:17
  system over the last few decades and amazing
• 22:20
  what you do on the AI side.
• 22:22
  Thank you so much for joining us, and
• 22:24
  what an impressive background you have. It's incredible.
• 22:26
  Doctor Shah, thank you for being with us.
• 22:29
  Thank you, Scott, for having me on this
• 22:30
  podcast.