Latest / Becker’s Healthcare Podcast / Rob Purinton, Chief AI Officer, AdventHealth

Transcript

Rob Purinton, Chief AI Officer, AdventHealth · Becker’s Healthcare Podcast

0:02
R one is the leader in health care
0:03
revenue management, helping providers achieve new levels of
0:06
performance through smart orchestration.
0:09
With more than twenty years of experience, r
0:11
one partners with 1,000 providers,
0:13
including 95 of the top 100 US health
0:16
systems and handles over 270,000,000
0:19
payer transactions annually. If you want to learn
0:21
more about how you can transform your revenue
0:23
cycle operations, visit us at wwwr1rcm.com.
0:30
Hello, and welcome to the Becker's Healthcare podcast.
0:34
My name is Will Riley from r one.
0:37
Joining me today is Rob Purinton.
0:40
Rob is chief AI officer at AdventHealth. Welcome
0:44
to the podcast, Rob. Thank you. Thank you
0:46
for having me. You bet. Rob, to start
0:49
off, tell us a little bit about your
0:51
role. Tell us a bit about AdventHealth.
0:54
Sure. So, I'll start with AdventHealth. So AdventHealth
0:57
is a
0:58
a health system primarily operating hospitals, but also
1:01
a total of 2,000 care sites that include
1:04
urgent care clinics,
1:06
physician offices,
1:07
specialty,
1:08
clinics,
1:09
outpatient surgery.
1:11
And, we operate in nine states and see,
1:14
over 8,000,000 patients a year.
1:16
My role as the chief AI officer is
1:18
a new role in 2025,
1:20
one that I'm I'm honored to serve the
1:22
organization where I've worked for twenty years.
1:25
So, originally, I was a software developer out
1:27
of college,
1:28
but I I caught the bug early on,
1:30
the lean bug,
1:32
and have worked in health care to
1:34
streamline process,
1:36
eliminate waste,
1:38
overall improve efficiency,
1:39
and and help improve clinical outcomes,
1:42
for all of those twenty years. So I
1:45
do view this role as a chief AI
1:47
officer as a continuation of that mission
1:50
to to streamline,
1:51
to standardize,
1:53
and to help our organization become a
1:57
better value proposition
1:58
for our communities, for our patients, and for
2:00
our our purchasers.
2:02
Fantastic. Okay. Great. I'm really looking forward to
2:04
talking to you. Sure.
2:06
Tell me,
2:07
a little bit perhaps to set the scene
2:09
about some of the strategic priorities that AdventHealth
2:12
has going into 2026.
2:14
Yeah. So,
2:15
first,
2:16
AdventHealth has what's called Vision 2030,
2:19
which is,
2:20
a strategic framework, a set of goals,
2:23
where we expect to arrive by 2030. It
2:25
includes
2:26
consumer focused experience. It includes
2:30
robust set of clinical outcomes. It includes financial
2:32
stewardship,
2:34
management of of population and risk, and there
2:37
are specific measurable goals. We're, in 2025,
2:40
so we're halfway into that ten year,
2:43
football game as it were. We're at halftime.
2:45
And, and so that's a good point to
2:47
evaluate our approach,
2:49
accelerate when we need to, and and really,
2:53
stabilize and lock in,
2:55
performance where we have it on some of
2:56
those strategic initiatives.
2:58
The the strategic goals of AdventHealth
3:01
haven't changed just because AI is now part
3:03
of the equation.
3:05
AI is an accelerant
3:06
to a number of our strategic goals. So
3:09
for example, in in clinical outcomes,
3:12
we know that AI can help some of
3:14
our clinicians,
3:15
make better diagnostic decisions,
3:17
augment their clinical intelligence,
3:20
with,
3:21
predictive algorithms that can help them make, the
3:24
right choice for the next step of patient's
3:26
care.
3:27
We know that it can
3:29
generate time savings back for our outpatient providers,
3:31
so they spend less time writing notes,
3:34
less time sitting down with fingers on keyboards,
3:37
documenting in the chart, and more time,
3:40
assessing patients,
3:42
dialoguing with them, planning care, and then and
3:44
then helping them feel comfortable with their diagnosis.
3:46
Yeah. So
3:48
that's what I would say about AI is
3:49
it while generative AI is a new technology
3:52
and the new implementations of AI are exciting,
3:56
the goals for health care value
3:58
have not really changed just because AI is
4:00
here. It's just given us a new, new
4:03
lens to look through and a new set
4:04
of tools to apply.
4:06
It it does seem though that it is
4:08
changing,
4:09
the way health care is thinking about innovation
4:12
and perhaps the pace of innovation and using
4:14
technology.
4:15
Health care providers haven't necessarily been known for,
4:19
like, being on the cutting edge of technology.
4:21
They've always had a relatively conservative
4:24
approach to technology.
4:25
But it does feel like that's different with
4:27
AI. Do do you agree with that? Oh,
4:29
yeah. Yeah. Absolutely. So everything has Why is
4:32
that there? Everything has sped up with AI.
4:34
And Yeah.
4:35
In part, you can look at at, where
4:38
where is some of the AI, not just
4:40
in health care, where is it making the
4:42
biggest difference? So one of the biggest differences
4:44
is
4:45
AI is being used to write computer code.
4:48
So if you think about how long does
4:49
it take to develop a new piece of
4:51
software for health care,
4:53
ten years ago, you were not using AI
4:55
to to write code at all.
4:57
Fast forward to 2025, you wouldn't dream of
5:00
writing,
5:01
any sizable,
5:03
piece of software without using AI for a
5:05
big part of it. And and that means
5:07
the cycles between
5:09
this generation of software and the next and
5:12
the next, they're getting shorter. Everything's speeding up.
5:15
So, we have to be our learning cycles
5:18
have to be faster.
5:20
Our,
5:21
our pilot experiments
5:22
have to be more robust around what we're
5:25
measuring and and how we're determining if they
5:27
were successful or we learned something different from
5:29
it. And, and then we have to operate
5:32
those cycles, you know, much more quickly than
5:34
we have in the past.
5:36
The typical,
5:38
life cycle for an application
5:40
today
5:41
is much shorter than it was for an
5:44
application released ten years ago. Mhmm. So by
5:47
the time you get it implemented and adopted
5:49
by, say,
5:51
all of your your clinical workforce, you already
5:54
have to be looking for the next new
5:55
thing. Right. So those innovation cycles have really
5:58
sped up. The other thing that's changed is
6:01
that
6:02
is is that tools have become more accessible.
6:04
AI is is not something that is only
6:07
there when companies introduce it to team members.
6:09
We're all getting to use it all the
6:11
time. So
6:13
every person at this conference, every person in
6:15
our health systems,
6:16
you know, know how to go access chat
6:18
GPT or their favorite, you know, AI tool.
6:21
And, and so they're seeing firsthand
6:24
how quickly they're getting value from it just
6:26
in their everyday lives.
6:27
Their expectations have been raised
6:30
about getting that value from AI back in
6:32
their work in the health system. Yeah. Okay.
6:34
Okay. Shrinking development cycles, and
6:37
this stuff works Yeah. Essentially. And the expectations
6:40
are going up there. Expectations. See with your
6:41
own two eyes, that works. Yeah. Yeah. When
6:44
you think about,
6:45
innovation in health care,
6:47
I think it's possible to look at it
6:49
through the lens of a couple of kind
6:50
of archetypes.
6:52
Right? You've got incumbents in the health care
6:54
system, large providers, large payers, large technology
6:58
companies, right, who have the data, the infrastructure,
7:01
the systems, the knowledge, and the institutional knowledge,
7:03
and so on. And then another group could
7:06
be you could call them insurgents,
7:08
new entrants, people coming into health care for
7:10
the first time, maybe from a technology perspective,
7:13
AI native,
7:14
companies and applications and platforms and so on.
7:17
When you look at, like, how you're advancing
7:20
your AI agenda at AdventHealth,
7:22
like, are you taking like, how are you
7:25
managing between those two things? Are you are
7:27
you an incumbent that's welcoming insurgence, or are
7:30
you, like how how are you doing it?
7:32
Yeah. It's a it's a tricky,
7:35
line to walk. So
7:37
we we wanna learn as much as we
7:38
can from AI native or AI first companies.
7:41
So we, you know, we we love that,
7:44
for example,
7:45
startups,
7:46
in the health care space
7:48
are are going and getting, you know, to
7:50
their their first 100,000,000
7:53
of of of ARR very quickly. And we
7:55
wanna learn what works
7:57
in that accelerated
7:58
cycle
7:59
so that we can adopt some of that.
8:01
Right? But then at the same time, we
8:03
have a stability,
8:05
that's expected around a health system our size
8:08
where we don't want big swings in performance.
8:11
We don't want big risks in in our
8:13
business cycles and in our finances. Right. So
8:15
we have to be very careful about, you
8:17
know, what we bring in and what we
8:19
make as part of our system.
8:21
I will say though, and I I like
8:22
your,
8:23
your your dynamic around the incumbents and the
8:25
insurgents and then the disruptors.
8:28
It it's really hard for an incumbent
8:31
to completely transform into an AI native company.
8:33
That's that's just not really realistic.
8:36
I think for many incumbents,
8:38
the the goal,
8:40
ought to be not to be the slowest
8:42
gazelle in the herd. K. Yeah. So you
8:45
don't need to outrun,
8:48
the fastest one. You just need to to
8:50
not,
8:51
get beat by the slowest one. That's right.
8:53
Yeah. And, and so that's that's part of
8:56
what we're learning. It's it's not about, like,
8:58
hey. We just you know, we we wanna
8:59
stay with the pack. It's that,
9:01
we don't wanna take any excessive risks.
9:05
Our communities trust us. Our patients trust us.
9:08
Our caregivers trust us not to just be
9:10
experimenting with their work. And and so we
9:13
we take that sacred trust very seriously.
9:16
And they also expect us to innovate
9:18
at a speed that they see the rest
9:20
of the world happening. So it's a very
9:22
tricky balance. Yeah. What I will say is
9:25
in in how we're evaluating AI, we're applying
9:28
all of the responsible AI,
9:31
best practices. So from NIST, the the National
9:34
Institute of Standards Technology,
9:37
new standards from the Joint Commission in CHI,
9:39
we're applying those to how we evaluate AI.
9:42
We're very focused on,
9:45
a fast cycle time for those evaluations
9:48
and,
9:49
and then creating feedback loops. So our job
9:52
with assessing risk in our governance process is
9:55
to determine,
9:56
is a particular implementation of AI low, medium,
9:59
or high risk?
10:00
And we may not always get that risk
10:03
assessment perfectly right,
10:05
but the way we,
10:06
hedge is
10:08
we may say, well, that's medium risk, but
10:10
we need it to come back and be
10:11
reevaluated every month.
10:13
And so even if we were wrong,
10:16
it comes back every month. We can determine,
10:18
actually, it's low risk. We can we can
10:20
dial down some of the follow-up. Or, actually,
10:23
we determined it's high risk, and we really
10:25
need to ramp up the the observability and
10:27
the monitoring here.
10:28
So I think it it's really important in
10:31
this age of innovation where everything has sped
10:33
up. That's not your imagination. It's real. It's
10:35
real.
10:36
That we very much focus on speed so
10:38
that we're keeping up with expectations and that
10:41
we're keeping up with, the pace of innovation
10:43
around us. And when you sort of make
10:45
a determination around risk Sure. Like, what what
10:48
what how do you how do you factor
10:51
risk? Yeah. Yeah. Great. Great question. So so
10:53
we'll look at, like, what is the AI
10:54
being used for? Is this a purely administrative
10:56
task that's just gonna save some minutes out
10:59
of desk work?
11:00
Is this something that's actually leading to to
11:03
clinical diagnosis
11:04
or treatment? And then we also look, is
11:06
there a human in the loop? So does
11:09
a does a licensed clinician get to check
11:11
the output before it actually reaches a patient?
11:14
We we look at if there are any
11:15
existing reports on the accuracy, bias, or performance
11:19
of that particular implementation of AI.
11:21
We look at have they implemented it at
11:23
other places first, and is there anything we
11:25
can learn from those?
11:27
So based on these kind of criteria, human
11:30
in the loop,
11:31
clinical
11:32
application,
11:33
use of PHI,
11:35
standardized security,
11:37
measures like, are they SOC two compliant?
11:40
We'll make a determination about their level of
11:42
risk. We have a a rubric that helps
11:44
sort in that low, medium, high. Yeah. And
11:46
then if we even though we may make
11:49
that determination about the level of risk, we
11:51
can still always say there needs to be
11:53
some mitigation on the back end. Like, we
11:56
need a report on this every week Right.
11:58
Every month or what have you. Right. Yeah.
11:59
And how does you you're you're saying, we.
12:02
So there's clearly a group involved. There's a
12:04
group. Can you tell us a bit about
12:06
that governance broader governance
12:08
framework that you've established? Yeah. There's two levels.
12:10
There's an AI governance committee Okay. Which is,
12:14
populated by executives.
12:15
And so they help us determine
12:18
where are the bright red lines, what are
12:20
the things that we're gonna, you know, do
12:21
or not do sort of as policy for
12:23
the health system. That's helpful. It meets four
12:26
times a year Okay. Evaluates kind of big
12:28
picture thinking and questions.
12:30
There's a technical group
12:32
that meets every week,
12:33
for, anywhere between one and two hours depending
12:36
on our our agenda.
12:38
And, and so it goes through,
12:41
all of the the reviews completed by
12:44
technical researchers over the previous week
12:47
so that we can ultimately land on a
12:49
determination,
12:50
low, medium, high.
12:51
And so in that forum, we'll have lots
12:54
of opportunity to ask questions,
12:56
to say, well, we need to go dig
12:58
a little deeper on this one. Have we
13:00
connected with the the clinical department that's looking
13:03
at this, and
13:04
have they considered that we already have a
13:06
solution that does this inside the system? Right.
13:09
So we get to ask all those questions.
13:10
We get to get answers and then make
13:11
our determination.
13:13
And where does your role fit into this
13:15
as chief AI officer? Because it's Yeah. An
13:17
unusual role Sure. I think at the moment
13:19
for health system to have. How did your
13:21
role come about, and who sort of set
13:23
it up? Where did the idea come from,
13:25
and how do you sit in this leadership
13:27
framework? Yeah. So our our previous CEO, Terry
13:30
Schott, his executive committee, created
13:33
the chief AI officer role, Brad Mahal. I'd
13:37
been doing some
13:39
work in, in my previous
13:41
role in analytics and performance improvement
13:44
related to just getting,
13:46
our doctors up to speed
13:48
on what's coming with AI, how we should
13:50
be thinking about it,
13:52
what are some use cases that would be
13:54
really impactful.
13:55
And,
13:57
you know, maybe I just drew the short
13:58
straw, but I I,
14:01
any no. I'm I'm kidding. I'm I'm very
14:02
blessed. I was I was asked to serve
14:04
as our first chief AI officer.
14:06
And,
14:07
and so I'd I'd I'd like to think
14:09
it's this combination of
14:11
a focus on performance,
14:14
speed, value,
14:15
as well as a robust technical underpinning and
14:18
the knowledge about how AI works
14:21
and and what are the what are the
14:22
risks that it can pose in our clinical
14:24
environment.
14:25
So, yeah, I that that that was I
14:27
was invited to to take the reins on
14:29
this one, and,
14:31
I yeah. It's I I I wouldn't trade
14:33
it for anything. Do you sit in the
14:35
IT team? Are you I do. I sit
14:37
in the IT department. Yeah. Group. Okay. Okay.
14:39
But I I will tell you, I work
14:41
with the entire organization. So Yeah. I work
14:43
with our with our finance leaders. I work
14:45
with our clinical leaders.
14:47
You know, there's there's very low friction to
14:51
them bringing
14:52
a
14:53
concern about existing AI or a new idea
14:56
about about AI up. I will say just
14:59
some of the,
15:01
critical few,
15:03
AI strategic priorities that probably all AI offices
15:06
are thinking about right now is
15:08
what's gonna be our platform
15:10
to let agents
15:12
start to transform our work. So AI agents
15:14
are more autonomous
15:16
bits of AI that you can allow to
15:18
do everything from, like, writing a a finance
15:21
report
15:21
to,
15:23
to to researching and transacting, you know, some,
15:26
some some things in your clinical system. You
15:28
can let them do those things. Mhmm.
15:31
Question is, should you? And so how do
15:33
you actually monitor and,
15:36
and and orchestrate
15:37
those agents to do their work? That's that's
15:39
a big one for us right now.
15:42
Another is really looking at our digital front
15:44
door. And so how is it that patients
15:46
engage with our system? Have we made it
15:48
frictionless?
15:50
Have have we made it frictionless for us
15:52
to call when there's a referral
15:54
and, and get them engaged with the right
15:57
kind of care team specialist or or what
15:59
have you? And then a third are,
16:03
a third area of, priority are really are
16:06
more around,
16:08
hospital based innovation. So
16:10
things in the surgical area, surgical robotics, computer
16:13
vision,
16:15
how are we helping to enable the next
16:16
generation
16:18
of,
16:18
of of surgeons
16:20
and, a more minimally invasive,
16:23
faster recovery,
16:25
in some of our surgical specialties. Excellent.
16:28
Which which areas of,
16:30
hospital operations do you think are most right
16:33
for full gentrification?
16:35
Well, I think revenue cycle is an easy
16:38
answer to throw out there. I think, everything
16:40
from, you know,
16:42
authorization
16:44
to to helping with documentation
16:46
in the, you know, in the midst of
16:47
your revenue cycle to to denials management on
16:50
the back end. I think all of that
16:52
is really ripe
16:53
for AI agents.
16:55
You know, I'll I'll also say there's a
16:56
lot of clinical workflow that so far we've
16:59
been hesitant, not just AdventHealth, but health care
17:02
has been hesitant to really introduce a lot
17:04
of lot of AI. But those AI assistants
17:07
and co pilots
17:09
can really help,
17:10
you know, whether it's
17:13
helping documentation to get more complete. I mean,
17:15
we've seen that with ambient scribes. Everybody's doing
17:17
ambient scribes. These are big help.
17:19
But even more along the clinical decision support
17:22
lines,
17:23
how do you make sure you're not
17:25
prematurely ruling out a differential diagnosis
17:29
that even though it's a low probability,
17:31
could pose a very great risk to a
17:33
patient? So some of those things, AI is
17:35
very good at keeping in front of you
17:37
as a low probability risk. Yeah. And, and
17:40
and so we could benefit and our patients
17:43
could benefit
17:44
from having those sorts of decision support copilots,
17:47
more in our, in our environments. Yeah.
17:50
Last question before we,
17:52
move to our conclusion. Sure. You mentioned revenue
17:54
cycle.
17:56
That is an area that historically has been
17:58
very labor intensive. Right? Very labor first.
18:02
And and technology has been used as a
18:04
sort of an aid, a support, right, to
18:06
people.
18:07
And it does feel like it's an area
18:09
that where that paradigm can flip flip completely,
18:13
where it can be largely autonomous with people
18:16
assisting.
18:17
Sounds like you agree with that premise. What
18:19
are some of the implications of that maybe
18:21
from a workforce perspective, or how are you
18:24
sort of how do you talk to
18:26
the teams about that as they as they
18:28
contemplate that? Yeah. You know,
18:30
so one one thing from a workforce perspective,
18:34
we face a challenge in health care in
18:36
that the number of patients who will need
18:38
care
18:39
will vastly outnumber the resources that we have
18:41
organized to provide it today in The US.
18:44
That that pinch point is coming.
18:47
And outside
18:48
of
18:49
some really aggressive reworking of our our workflows
18:53
and application of AI, we don't have another
18:56
great solution to to address that pinch point.
18:59
So the the clinicians are not gonna come
19:02
out of nowhere to, you know, ride in
19:04
and and come rescue us from bad ratios.
19:06
This is we just we have to go
19:08
after every bit of of non value added
19:10
work we can.
19:11
And then that may mean that we need
19:13
to,
19:14
repurpose roles into more patient facing,
19:17
you know, more,
19:19
more medical aid, more patient care tech type
19:22
roles,
19:23
and less back office, roles. I think we've
19:26
watched the trend line in health care. The
19:28
number of,
19:30
administrative staff compared to the number of clinical
19:32
staff in the last twenty years, clinical staff
19:35
have not really budged, and administrative staff have
19:38
skyrocketed exponentially.
19:40
So that's a trend we've got to reverse
19:43
and get more resources pointed back at patient
19:45
care.
19:46
A way to do that is agentification
19:48
and Yeah. Application of AI. Yeah.
19:51
Yeah. Last question. You mentioned,
19:53
you last question. You mentioned,
19:55
your digital front door. Sure. Can you end
19:57
with a couple of thoughts on how,
19:59
all of this new technology can really improve
20:01
the patient experience?
20:03
Yeah. So I think overall, the the word
20:06
is, is frictionless
20:08
that that we're looking for.
20:10
And and what we would hope to happen
20:12
is
20:13
that it's,
20:15
seamless and obvious,
20:17
you know, the way you experience care. Say
20:20
you go to your primary care physician,
20:22
suggest that you go see a cardiologist,
20:25
that within hours,
20:27
a a call or a message in the
20:29
right, format, in the right,
20:32
modality
20:33
arrives to you and makes it easy to
20:35
suggest some appointments. Maybe it's already integrated with,
20:38
like, your calendar on your on your smartphone
20:41
and can suggest sometimes that
20:43
look like they already work for you,
20:46
to to go connect with a cardiologist.
20:48
The other is when consumers want to, to
20:51
engage with our system,
20:53
how do how do those inbound calls,
20:56
get routed in the smartest,
20:58
most frictionless way possible
21:00
so that it's it's really easy to get
21:02
to,
21:03
the provider you need at the time you
21:05
need,
21:06
and that as few barriers like,
21:09
like office schedules or
21:12
or or, like, you know, authorization
21:15
questions,
21:16
as few of those as possible make their
21:17
way in. We have an initiative,
21:20
one of our CEOs
21:22
coined, burn all the clipboards. So,
21:25
you know, when you go to the medic
21:26
when you go to the physician office and
21:28
inevitably, you are handed a clipboard Yeah. And
21:31
asked to fill this out, which, of course,
21:33
you've done
21:34
triple digit times before. Yep. I refuse to
21:36
do it now. Yeah. Oh, you refuse. That's
21:38
great. I I remember that one. That's good.
21:40
But, No one says anything.
21:43
Like, they just let you in? Yeah. That's
21:44
wonderful. Brilliant. I like that. So,
21:47
so yeah. I mean, part of this is
21:49
we we we know we have your information.
21:51
Like, we Right. We this should be frictionless
21:53
and easy, seamless,
21:55
and you should not be writing down the
21:57
same things again and again and again and
21:59
again and again.
22:00
So taking those experiences
22:02
and and letting AI help to to transfer
22:05
your information
22:06
from one encounter to the next or make
22:08
that whole thing just more seamless,
22:11
that's that's what we expect, to really happen
22:13
with a digital front door. Awesome. Yeah. Rob,
22:16
thank you so much. It's been a real
22:17
pleasure talking to you. Yeah. Yeah. Thanks so
22:19
much for sharing your insights. Having me. Really
22:20
appreciate it. Yeah. Thank you all. Thanks.