Latest / Becker’s Healthcare Podcast / How AI Can Drive Meaningful Results in RCM

Transcript

How AI Can Drive Meaningful Results in RCM · Becker’s Healthcare Podcast

0:00
This is Scott Becker with the Beckers HealthCare
0:03
podcast.
0:04
Thank you for tuning in today. Today, we've
0:07
got a brilliant physician leader,
0:09
and we're gonna talk today about how artificial
0:12
intelligence
0:13
can drive meaningful results in revenue cycle management.
0:17
We're joined today
0:19
by doctor Josh Galeras.
0:21
Josh is the physician,
0:23
the cofounder,
0:24
and the head of product development
0:26
at Smarter Technologies.
0:28
Josh, you take a moment and,
0:31
introduce yourself.
0:32
Absolutely. Thank you so much for having me
0:34
on. Definitely appreciate it. And,
0:38
as you said, I'm a physician.
0:41
My cofounder at SmarterDx and I trained together
0:44
in internal medicine
0:46
at Cornell, and then I practice as a
0:48
hospitalist at Columbia.
0:50
I also have a software coding background and
0:53
data background from before
0:56
medical
0:57
school and was always sort of drawn to
1:00
the power that technology could
1:03
have
1:04
in
1:05
whatever I was doing,
1:07
including medicine. And
1:10
I used that experience to
1:12
become,
1:13
at Columbia,
1:14
an NIH and DOD funded,
1:17
medical researcher
1:18
doing data science and bioinformatics
1:21
with electronic medical record data.
1:23
I was also helping on a number of
1:25
hospital operational initiatives.
1:28
And through that work, I
1:30
learned a lot about
1:32
revenue cycle.
1:34
As you know, every hospital's
1:36
a little bit different, and
1:39
it's not something that physicians
1:41
know
1:42
basically anything about. I mean, to be blunt,
1:44
99%
1:46
of physicians
1:47
probably know basically nothing about how hospitals
1:51
actually get paid.
1:53
And so the more I got exposed to,
1:56
the more I was able to start imagining
1:59
how
2:01
technology
2:02
and
2:03
AI would be able to
2:06
augment the quite manual work of revenue cycle.
2:12
And so my cofounder,
2:15
Michael Gao and I decided to start
2:18
a company, SmarterGX,
2:20
in order to start
2:22
bringing that
2:23
the power of AI to to health care
2:26
revenue cycle. And we started
2:28
by,
2:30
creating a product that we now call rebill,
2:34
which
2:35
ingests
2:36
clinical
2:37
and
2:38
administrative data from health systems,
2:41
uses
2:43
AI on the clinical data
2:45
to try to understand what diagnoses
2:48
and procedures
2:50
were actively being treated
2:52
during a patient's
2:54
inpatient hospitalization
2:56
during their stay in the hospital
2:59
and
3:01
to compare
3:02
what the AI generates as a list of
3:04
possible
3:05
diagnosis codes to what the hospital
3:09
hospital's existing processes
3:11
had cataloged
3:12
in their review
3:13
of that chart for billing purposes.
3:16
And where we find diagnoses
3:19
that
3:21
weren't
3:22
captured
3:23
by that sort of manual process
3:26
and would impact
3:28
the hospital's
3:29
reimbursement,
3:30
their revenue, or their quality metrics,
3:33
then we present those back to a human
3:35
in the hospital for review to potentially capture.
3:39
And
3:40
the impact of that
3:42
is something like $2,500,000
3:45
for 10,000
3:47
discharges
3:47
in new revenue that the hospital
3:50
would otherwise
3:51
have foregone for care that they already provided
3:55
and are completely justified
3:57
in
3:58
in billing the payers for.
4:01
And,
4:02
you know, last year, we drove almost half
4:05
$1,000,000,000
4:06
in revenue to providers
4:09
that they otherwise would have missed out on,
4:11
and and 2026
4:13
is is looking looking great as well.
4:16
No. It's it's really a remarkable background. I
4:18
mean, you you've trained throughout the world.
4:21
You also spent time,
4:23
teaching and working at New York Presbyterian.
4:26
You're a technology person and a physician. It's
4:29
it's really an amazing
4:30
mix of of what you do.
4:33
Talk a little bit about Smarter Technologies was
4:35
created in 2025,
4:37
bringing together a handful of companies,
4:40
Access Healthcare,
4:41
SmarterDx,
4:43
thoughtful AI,
4:44
and PIECES technologies.
4:47
How would you describe the platform for this
4:48
combined platform,
4:50
and what larger problem are you trying to
4:52
solve together that's greater than the sum of
4:54
its parts?
4:56
Absolutely.
4:57
You know, every year in this country,
4:59
we spend
5:01
a $100,000,000,000
5:03
on the administration
5:04
of revenue cycle in health care, which is
5:07
just insane. That's just, like, such a huge
5:09
number, and I think I think that's a
5:10
conservative estimate. Obviously, I've heard much higher numbers
5:13
for that. And as I think about why
5:15
that's the case,
5:17
I, you know, I actually my dad was
5:19
a physician.
5:20
His dad was too, but I I I
5:22
have these, like, pretty vivid memories of being
5:24
maybe seven, eight years old and following my
5:26
dad around
5:27
a hospital.
5:29
And,
5:30
you know, this was early nineties
5:33
and pre EMRs, and and what he was
5:36
doing
5:37
everywhere he went in the hospital was, like,
5:39
walking and finding these, you know, thick
5:41
manila folders full of patient details that told
5:44
the story of what was happening for the
5:46
patient, and he would, you know, sign his
5:48
orders and,
5:50
write his notes in those folders. And
5:53
the
5:55
system at the time had to figure out
5:56
how to get compensated
5:58
for
5:59
the care that they were providing.
6:01
And the way to the best way to
6:02
do it, you couldn't send that whole thick,
6:04
you know, maybe a 100 page middle of
6:06
folder to the payer.
6:08
You had to start
6:10
by
6:11
going through the information and sort of shrinking
6:14
it down into something that
6:16
was manageable and could be fast. And that's
6:18
where the process
6:19
of
6:20
of medical coding sort of originates
6:23
and,
6:24
the DRG system, diagnosis related group system that
6:27
hospitals are largely reimbursed with today comes from.
6:30
Just this this process of trying to translate
6:34
what's happening in clinical care into something that
6:37
could fit on a single page and be
6:39
fast, you know, at the time to to
6:41
the payer. And while we've largely digitized
6:44
the system, we really haven't changed the underlying
6:48
process. And one of the core things there,
6:50
and this is what, you know, I think
6:52
why we spend so much money on health
6:54
care is it's largely manual. Right? We have
6:57
humans
6:59
going through
7:00
all of these charts for every inpatient hospitalization
7:03
and coming up with a catalog of the
7:06
codes
7:06
that were were treated.
7:09
And so
7:10
the promise of smarter technologies
7:13
is really about using
7:16
AI
7:18
to augment and improve
7:20
the the work that those humans are doing.
7:23
And while
7:25
AI is remarkable
7:27
in in many of its capabilities
7:29
today,
7:31
we
7:33
think there are still
7:35
an important need to have humans in the
7:38
loop reviewing and doing a lot of that
7:40
work. And so on
7:42
the SmarterDx
7:43
side, as I said, you know, we present
7:45
our findings to a human at the health
7:47
system,
7:49
but not every
7:51
health system
7:52
wants,
7:54
their own humans to be the ones doing
7:56
all that work. And so
7:59
that's where, you know, one part of smart
8:01
technologies comes in, Access Healthcare. It's sort of
8:04
one of the largest and most tech forward
8:08
BPOs,
8:09
that's doing health care work in this country.
8:13
And they have a great reputation for low
8:15
cost and and and high quality.
8:18
And
8:19
we are working on integrating
8:23
our AI products
8:24
with
8:25
their
8:26
services team.
8:28
Thoughtful AI
8:29
is a company that's using
8:32
agentic technology to to help improve
8:37
that efficiency as well to make those
8:40
humans
8:41
even more effective.
8:43
And then Technologies
8:45
is
8:47
an early stage company
8:49
that's
8:51
working
8:52
forward in revenue cycle at the time physicians
8:55
are actually writing their note. And the ultimate
8:57
goal there is to
8:59
integrate revenue intelligence into the physician experience and
9:03
do it in a way that makes a
9:05
physician
9:06
experience with the revenue cycle better and more
9:09
seamless
9:10
and
9:12
improves the physician experience in how you write
9:16
notes in the technologies and and and AI
9:19
capabilities that are available to to doctors taking
9:22
care of patients today.
9:24
And altogether,
9:26
our goal is ultimately to reduce the cost
9:30
cost of health care in this country by
9:32
using AI
9:34
to augment all of these manual processes.
9:37
Thank you very, very much.
9:40
When we talk about AI today in use
9:42
cases,
9:43
obviously, there are a ton of different solutions
9:47
being thrown at providers and health systems.
9:51
Everybody wants to pilot something,
9:53
Even doing a pilot for many, many companies
9:56
or many health systems is a lot of
9:58
work. It's not it's not, you know, people
10:00
say, oh, we're just gonna do a pilot
10:02
with you. It's gonna be easy. But the
10:03
lift in doing a pilot with somebody is
10:05
is significant.
10:07
And so people have to choose where they're
10:08
really gonna double down, where they're gonna do
10:10
pilots, where there's gonna be big wins.
10:13
From your vantage point,
10:15
which AI use cases are actually delivering our
10:18
way today, you know, and and what helps
10:20
them to scale in a health system? What
10:22
works? What doesn't work? Talk to us a
10:24
little bit about that.
10:26
Yeah. You know, I was having lunch with
10:29
a CIO at a very prominent
10:31
health system,
10:33
earlier this week. And one of the things
10:34
he said was, basically,
10:37
every
10:38
time savings
10:40
AI like, every every time they've tried to
10:44
do an AI initiative
10:46
with a goal of just saving time as
10:49
the value proposition,
10:50
it's fair
10:52
every time.
10:53
And,
10:54
again, very large, very prominent health system,
10:57
and they they're they're doing a lot,
11:02
of of these initiatives. And I thought that
11:04
was that was a very interesting comment, and
11:06
it and it very much mirrors something that
11:08
we've
11:10
learned through experience
11:12
at
11:14
Smarter.
11:15
You know, we piloted
11:18
a product with with a large Southeast
11:22
Health System earlier or, I guess, last year.
11:26
And,
11:28
it was it was a registries product using
11:31
AI to
11:33
help humans
11:35
who are going through medical records in order
11:38
to
11:39
abstract key information from those records for a
11:42
variety of different registries, whether it's sepsis or
11:46
an oncology registry or or or something like
11:48
that.
11:49
And what was fascinating
11:51
to us was the AI
11:54
got it right
11:55
almost, like, 90% of the time. I think
11:57
in our final our final results before we
12:00
shut it down were, like, 89% of the
12:02
time. But
12:04
it didn't save any time
12:06
because the users of the system
12:09
still had to go through
12:11
every single answer and sort of verify that
12:13
it was right. And
12:15
and the work of reviewing that chart ended
12:18
up just switching and replacing
12:20
the work that was going into
12:23
finding those answers in in the first place.
12:26
Right? And so,
12:28
you know, I think that
12:31
from
12:32
a ROI perspective, which I think was
12:36
a a core part of your question,
12:38
you really have to
12:40
do a lot of work and get further
12:42
than I think AI is really capable of
12:44
today to generate
12:46
really meaningful time savings in a way that
12:49
systems trust at scale. Right? It's like, yeah,
12:53
you can save time on this task or
12:55
that task here or there, like, exactly like
12:56
you said, these sort of, like, point things.
12:58
But to save enterprise time in a way
13:00
that ultimately saves cost is is just a
13:02
hard value proposition.
13:04
And I'll share one more sort of little
13:06
little tidbit here is
13:09
with the SmarterGX
13:11
pre bill solution,
13:13
we end up finding
13:15
more revenue opportunity
13:17
than
13:19
the cost of most of our health systems'
13:22
entire
13:23
CDI and and coding teams.
13:26
And so if we would have gone after
13:30
a sort of time savings approach,
13:33
even if we fully automated
13:35
all of CDI
13:37
and all of coding,
13:39
we wouldn't have been able to generate as
13:41
much value for the health system
13:43
as we are today
13:45
because our costs would be capped at what
13:47
they're spending.
13:48
And and that's a really high bar because
13:50
you have to, like,
13:52
fully automate
13:53
all of those those functions, which is just,
13:55
like, really hard to do.
13:57
And so,
13:58
you know, I think that that's an important
14:01
lesson that we've taken away
14:03
is,
14:05
you know, there are there is role for
14:08
augmenting humans and trying to improve their experience
14:11
and save time, but that's not gonna directly
14:14
lead to ROI. It's really about expanding what's
14:17
possible within the health system
14:21
and capturing more
14:23
or
14:24
or doing a better job
14:27
that is gonna result in in value. And
14:29
I think that is where AI paired with
14:32
humans actually makes the most impact on on
14:36
health care in the short term.
14:38
Who knows what the long term looks like?
14:40
But but but yeah. No. I think that's
14:42
exactly right. And and I and I love
14:44
this. This this concept that it's just technology
14:47
is is it's this concept that technology plus
14:50
people is going to be everything. You're you're
14:52
you're gonna need both to make this work
14:54
great.
14:56
Talk a little bit about guardrails and innovation.
14:59
Increasingly, systems are having AI governance structures in
15:02
place to try to make sure that people
15:03
don't go off the rails and do all
15:05
their jobs through AI and end up in
15:06
weird spots.
15:07
How should cut companies how does the systems
15:10
how do you look at guardrails in innovation
15:13
so things move fast,
15:15
but you still have humans in the lupus
15:17
checkpoints.
15:18
You know, how how do you sort of
15:20
put these two things together
15:22
so one doesn't get way ahead of the
15:23
other ended up in an area
15:26
that's very different than expected?
15:28
Yeah. That's that's a a a a great
15:30
question and and really important question.
15:33
And and core to the the smarter strategy
15:36
really is
15:37
to make sure that
15:39
you
15:40
you're carefully attuned to what AI is and
15:43
isn't capable of and
15:46
that you
15:48
have humans
15:50
in the loop to make sure that where
15:53
there is risk of AI,
15:55
as you say, going off the rails,
15:58
there's a human there who's incentivized
16:01
to make sure that's correct. And I think
16:04
that's
16:05
kind of the core
16:06
thing is you you know, it's very easy.
16:10
I
16:11
you know, as a physician, I would somewhat
16:13
ashamed to admit, I used to, like, copy
16:15
forward my notes. Right? I take my note
16:17
from yesterday
16:18
on a patient and and and use it.
16:20
It's very easy to sort of, like, miss
16:21
the details.
16:22
And so sometimes I would say two days
16:24
in a row, you know, the patient came
16:26
back from
16:28
the OR yesterday. Right? But
16:30
they went to the OR once. Right? And
16:32
and and so
16:34
that's a sort of example of where it's
16:37
easy to skip over details. And so what
16:40
you wanna do is present information in a
16:43
way that makes it hard to skip over
16:45
the important or risky details
16:47
that
16:48
the
16:50
AI is is suggesting
16:52
and not just
16:54
give an easy button to
16:57
moving on to
16:58
the next case or or or accepting
17:03
those recommendations. I think
17:05
that's a
17:07
pretty core part
17:09
of having humans in the loop is
17:12
is that incentive structure.
17:15
It's fascinating. It's fascinating, and and it's it's
17:17
pressing to talk to you. You gotta gift
17:20
of way of explaining it and thinking about
17:22
it.
17:23
Take us to the next issue. So much
17:26
of AI today and technology today
17:29
is spent on pre authorizations,
17:31
denials,
17:32
revenue integrity.
17:34
Where can AI
17:36
most credibly cut avoidable denials and speed appeals
17:40
in the next few years, and and and
17:42
what traps are there? Is it seems like
17:44
this is an arms race between
17:47
systems providers and essentially
17:49
Medicare Advantage and big payers.
17:52
How to sort of use AI to cut
17:53
through some of this to level the playing
17:55
field,
17:56
and keep payments moving?
17:59
Yes. Absolutely.
18:00
I I I think that's a great great
18:02
question, and
18:04
a arms race is a very good way
18:06
to describe
18:08
exactly what's happening, I think, between
18:11
providers and
18:13
payers right now.
18:15
You know, I I do think there's
18:18
perception
18:18
on both sides that
18:21
each that the other side is just trying
18:23
to fleece them or trying to take advantage
18:27
of of things.
18:29
I don't think either side
18:32
truly thinks of themselves
18:34
as doing that. I think
18:37
both sides are really trying
18:40
to
18:41
trying to tell the right story, but it
18:43
is
18:43
quite complicated.
18:45
And
18:46
I think the promise of AI is really
18:49
to
18:50
simplify
18:52
our ability
18:54
to understand what's happening with patients
18:58
and what's required
19:00
in order to justify
19:03
billing for care.
19:04
I I think that's kind of the core
19:08
challenge in revenue cycle today and why it
19:11
costs
19:12
a $100,000,000
19:14
to administer revenue cycle today
19:17
is it takes humans
19:19
looking at every single
19:21
chart, every single data point in order to
19:24
accurately describe
19:25
what's happening.
19:27
And I think the real promise of AI
19:31
is
19:32
to take that process and augment those humans
19:35
with
19:36
better,
19:37
faster understanding of what's going on. And I
19:39
think that can play in a number of
19:42
different places. So one is
19:44
on the appeal letter generation side that, you
19:46
know, at Smarter,
19:48
we have an appeal letter writer
19:50
that takes in a denial, looks through all
19:52
the clinical data data and prewrites an appeal
19:55
letter for a nurse. And I think the
19:57
core value proposition that that is that that
20:00
provides
20:01
is less on
20:03
writing the appeal letter part. It's more on
20:06
searching through the 30,000 data points that we
20:09
get on average for an inpatient chart
20:11
and
20:12
identifying what the important information is in those
20:17
in that data
20:18
so that
20:19
you can
20:21
craft a strong
20:23
appeal letter. It's really about pulling out and
20:26
simplifying the process
20:27
of trying to understand what was happening
20:30
in that chart.
20:31
I think that's also true
20:34
even before
20:36
discharge
20:37
or while the patient's still being taken care
20:39
of.
20:41
A lot of the challenge around
20:44
denials and appeals and certainly complex clinical denials
20:48
is just understanding what the requirements
20:51
that payers have in order to meet, say,
20:54
inpatient status criteria
20:56
are
20:57
and
20:58
to
20:59
be able to very
21:01
clearly articulate
21:03
why a patient that you're taking care of
21:06
meets those criteria.
21:07
If I don't know what those criteria are
21:09
or it's a, you know,
21:11
47
21:12
page document that outlines what those criteria are,
21:16
it's gonna be very difficult for me to
21:18
accurately
21:19
describe to you how the patient I'm
21:22
taking care of meets those criteria.
21:24
But AI can understand that and very quickly
21:27
translate
21:28
it and make suggestions to the physician that
21:32
that help describe that. And I think that
21:34
sort of, like, simplification
21:36
of complex clinical data in a way that
21:38
humans
21:39
can then validate
21:41
and
21:42
push forward is the promise that AI has
21:45
in the short term to to really take
21:48
some cost out of the system. And and
21:49
we're quite hopeful that we'll be able to
21:53
help, but I know others are working on
21:54
it too. And and, you know, for my
21:56
children's sake, I'm I'm really hopeful that
22:00
we or or someone successful
22:02
at
22:03
reducing the cost of care here.
22:06
Wet soap so. It's become daunting. The cost
22:09
for a family of four compared to the
22:10
average income of a family of four are
22:12
are numbers that just don't work.
22:15
And so if this is a way to
22:16
help get more people focused on the right
22:18
stuff
22:19
and and in the clinical care
22:22
versus
22:23
so many people
22:24
and time and hours and money spent on
22:26
administration, this would be fantastic.
22:29
Doctor Glarus, talk to us for a second.
22:31
CMS is moving to faster, more transparent
22:34
electric
22:35
prior authorizations.
22:37
How to change which AI investments may pay
22:40
off first, or does it change it?
22:42
Yeah. I mean, I think
22:43
this gets to what
22:45
I was I was just talking about a
22:47
little bit, which is, you know, in order
22:49
to submit those electronic prior authorizations, you first
22:51
sort of need to know what patients
22:54
need a prior authorization and what data is
22:57
relevant to send to the payer in order
22:59
to request that authorization.
23:02
And,
23:03
you know, I remain
23:04
pretty convinced
23:06
that the the core
23:08
task that we need AI to do
23:10
is
23:11
get better at understanding
23:14
complex clinical data.
23:16
And I'm really putting my
23:18
money with where my mouth is preferably in
23:20
in in the sense that, like, at Smarter,
23:22
we're really investing a ton of
23:25
our
23:26
engineering and data science resources in exactly this
23:29
problem because I think it shows up in
23:32
so many of the diff so many different
23:34
parts of revenue cycle.
23:37
And
23:38
the better we're able to
23:40
do that, the better we're able to solve
23:42
kind of all of these different problems. And
23:44
so I think that
23:46
is a is a core capability that we
23:50
as
23:50
a country,
23:52
in the health care system need is just
23:54
that
23:55
deep understanding of
23:58
what's happening with patients
24:00
and
24:01
how
24:02
to translate clinical data into the ontology
24:05
of revenue cycle.
24:07
Doctor Galeris, let me ask you one follow-up
24:10
follow-up question.
24:11
You get this fantastic blend of of deep
24:14
sort of technology experience
24:16
with deep physician health care experience.
24:19
Anything else that you'd like to share today
24:21
that we haven't touched on,
24:23
from your perspective?
24:25
You know, what I'll what I will say
24:26
is I while we have a lot of
24:28
challenges
24:29
in
24:30
front of us and
24:32
there's
24:33
there's a ton of risk,
24:35
I think that
24:37
I am
24:38
pretty profoundly optimistic
24:40
about what's possible
24:42
over
24:43
over the short to medium term.
24:47
AI is not a panacea. It will not
24:49
solve all of our problems.
24:51
And
24:53
being really
24:54
critical and thoughtful about how you use AI
24:56
is very important,
24:59
but
25:00
it is a meaningful advance
25:03
on how we do things.
25:05
And
25:07
the results I'm seeing come out of our
25:09
team
25:10
are
25:10
extremely encouraging
25:12
on a number of dimensions
25:14
that that really
25:16
give me hope that
25:18
these are surmountable problems.
25:20
And even though we've been
25:22
thinking about them and working on them for,
25:24
you know, as a as a system for
25:26
for decades now, I do sort of see
25:29
a a light at the end of the
25:30
tunnel here where
25:32
we can
25:33
make a really big impact and shift
25:36
these health care dollars
25:38
from administration
25:40
to
25:41
patient care to discovery
25:43
to research to
25:45
things that
25:47
meaningfully
25:48
improve people's lives.
25:50
And, you know, that's that's our focus.
25:53
That's what I challenge the team to do
25:56
literally every week.
25:57
And
25:59
I think we will
26:01
see some real gains over over the the
26:04
next few years. So I'm I'm excited, hope
26:06
hopeful, and optimistic.
26:08
Josh, thank you so much for joining us
26:10
today on the Becker's Healthcare podcast.
26:13
We wanna thank you. We wanna thank our
26:15
podcast sponsor, Smarter Technologies.
26:18
Just an absolute pleasure, Josh, to visit with
26:20
you and talk about all the things that
26:22
you're doing and the impact of AI on
26:24
revenue cycle management. Thank you so much for
26:27
joining us.