1ArtificialIntelligence · 2026
AI Transforming Health Operations | Arturo LoAIza-Bonilla
THE CONVERSATION
Session overview
Moving healthcare AI from experiments into operations is the challenge in Arturo Loaiza Bonilla’s presentation. He surveys administrative applications, including documentation, prior authorization, scheduling and supply chains, alongside oncology examples involving clinical trial matching. The proposed shift from generating answers to coordinating workflows depends on connecting information with patient needs and organizational processes. He emphasizes workforce reskilling, implementation champions and collaboration between clinical, technical and security teams. In discussing ethical oversight, he calls for explainable decisions, attention to training data and local differences, and continuing human involvement, positioning technology as support for patient care rather than a replacement for clinicians.
Sources and review notes
Source checked against provider captions by two independent assistants. The recording has not been independently audio reviewed.
Moving healthcare AI from experiments into operations is the challenge in Arturo Loaiza Bonilla’s presentation.
1:24–1:58
right. So, uh hi everyone. Um I'm Dr. Arturo Loaiza Bonilla as was mentioned by Glenn and it's a pleasure for me to be here today. Uh as a medical oncologist, I always want to have a perspective where I can uh you know, put into a couple of views. One from my medical oncology view, uh the other one from my administrator side, and in a the grander scheme of things on uh entrepreneurship and see how operations are changing across. So, so that's how we're going to have this conversation today. Uh so, I call it from pilot to production, the 2026 operational revolution using AI.
4:51–5:33
But now the question is how are we going to implement this in a broader scale, right? Because we we have all these tools, all these opportunities, but there is this paradox where 83% are pilots, less than 10% there is an investing in the real infrastructure for deployment. So, people are just kind of trying solutions, but but they really haven't put a full end-to-end approach to the enterprise level of this and and and I think this is a good time now to stop the experimentation. We know these tools have some you know really major opportunity here and the challenge is causing from you know pulling from the pilot to actually production to how we can make this really meaningful.
He surveys administrative applications, including documentation, prior authorization, scheduling and supply chains, alongside oncology examples involving clinical trial matching.
8:07–9:31
uh, I'm going to give you all these, you know, legacy friction points and the AI enabled velocity that we see. So, uh, you know, RCM, or revenue cycle management, uh, we have all these issues with prior authorization and denials. 60% of denials are never appealed because of attrition game. We just simply change things and say we don't want like when we talk with insurance companies or even Medicare, we're like, "Okay, we got a denial, we're going to write it off." Uh, now with AI, we can reduce, uh, the the, you know, cost to collect using, for example, integrated EMR solutions. Uh, prior authorization, the same thing. 92% of physicians have to go on the phone like myself and try to do appeals. Uh, now we can do 90% auto approval rates if we have an integrated AI tool. Uh, and that's something that, for example, Cohere Health and Highmark have been accomplishing. Uh, and with that also came out 85% reduction in the amount of people involved, uh, in those, uh, prior authorization issues. And then supply chain, right? So, we the supply chain comes not only in healthcare, but in many spaces, but in healthcare it's unique because of the complexity, uh, but it's usually very reactive, uh, and can lead to clinical shortages. Here with AI, we can predict, uh, what if stockouts are happening, so using machine learning, so we never run out of the things that are essential for operations, either a medication or a piece of equipment or or a or a spare
9:31–9:37
part, right? So, that's all part of the whole process and giving you alerts for it.
11:55–12:24
So let's talk about some use cases. So ambient AI that was the first healthcare first massive adoption win. So for example Kaiser Permanente they use a bridge uh they integrate it directly with the EMR. Uh it was 40 hospitals, 600 offices uh and they in the first year they saved about 15,691 hours uh in time spent typing and just simply having the listening uh and transcription in the system uh with uh better communication with patients.
19:01–20:36
And that's where I came about working with several innovators and in MassBio we built this multi-agent architecture called Synergy AI. And and basically we put agents that are able to do LLM extraction on oncograph which is a knowledge base and give recommendations for clinical trials in real time so we can get a single targeted patient match in in a matter of seconds. We showed this at the American Cancer you know American Society of Clinical Oncology meeting and we validated about 3,104 patients matched with Synergy AI and we were able to do the job of about you know 5,000 patients and which would have taken about you know 50,000 number of work hours in just an hour right so so the the benefits are legit. Uh So the next frontier is to move from generation to orchestration right so instead of asking just prompting we need to use multi-agent AI where the agent is able to collect database information and connect with autonomous go driven systems that able to schedule patients to get them where they're supposed to be either for regular standard of care appointment or for getting the billing ready right so the reality check is 80% of executives expect value but only 3% of agents are deployed in live workflow so how are we going to see the results if we don't have that put together? Um and and the idea here is this is going to help also on the revenue cycle side is scheduling care coordination supply
20:36–20:39
chain as we mentioned earlier.
The proposed shift from generating answers to coordinating workflows depends on connecting information with patient needs and organizational processes.
18:02–19:26
we have all these issues with, you know, interfaces where we don't have all the data surfacing in front of us. We have to deal with authorization, authorization fatigue, time-bound priorities, and sometimes we even miss biomarkers that the patient may be eligible for and can change their lives. And those are commercially available. So, it's not that we don't have the testing or the drugs. We just don't have the connectivity together. And this is mostly because of this phenomenon of data drift. As everyone can imagine, we all as humans change. You know, the our conditions change over time, our cancer progresses or stays the same, our conditions are always in a constant evolution and the same the environment around us in terms of where we are being treated, what physician is in front of us, what labs are being drawn at that moment, and the standards of care that evolve along with that. So, we have to think about this more strategically, more for like logistics. So, Uber, Instacart, that we need real-time awareness for availability. And that's where I came about working with several innovators and in MassBio we built this multi-agent architecture called Synergy AI. And and basically we put agents that are able to do LLM extraction on oncograph which is a knowledge base and give recommendations for clinical trials in real time so we can get a single targeted patient match in in a matter of seconds.
19:55–20:39
So the next frontier is to move from generation to orchestration right so instead of asking just prompting we need to use multi-agent AI where the agent is able to collect database information and connect with autonomous go driven systems that able to schedule patients to get them where they're supposed to be either for regular standard of care appointment or for getting the billing ready right so the reality check is 80% of executives expect value but only 3% of agents are deployed in live workflow so how are we going to see the results if we don't have that put together? Um and and the idea here is this is going to help also on the revenue cycle side is scheduling care coordination supply chain as we mentioned earlier.
25:44–26:12
Uh, how can we do this? Being very patient-centric. So, we need to make Uber, Amazon, Instacart, but for clinical operations, clinical trials, clinical operations overall, because manual methods right now they really fail and we need a hybrid click and mortar approach where the patient is digitally enabled with tools and then we get them to the point of care for with this AI prescribing hub. So, we monitor over time and find a solution.
27:06–27:37
How? With this flywheel, right? So, patient centered the hub in oncology, we can use this for clinical trial matching, we can check how drugs are being tested and and used and and do real-time analytics, we can activate clinical trials in point of care, we can do digital enablement and education, we can optimize biomarkers, so we can improve the process of the future, build this foundational models further and eventually get even better drugs for our patients and understanding things in real-time using the data that we have.
He emphasizes workforce reskilling, implementation champions and collaboration between clinical, technical and security teams.
23:29–24:27
So the two battleground here is transforming the workplace, right? So in the in the next seven minutes or so that I have we'll talk about how we're going to be able to transform organizations on this. So because we had 10% algorithms already there. 20% is data and technology that is being deployed but 70% is processes which is you know is the way it is we're humans at the end, right? So we need to have a lot of reskilling. 60% organizations reports is reskilling is a top challenge. Security as well. So I have a very nice chief security officer at St. Luke's but you know is is hard to get every single solution in front of that team because they're always you know very cautious rightfully so and and and certain algorithms have a level of scrutiny compared to others but no one's putting AI in the single umbrella and and that reskilling has to happen as well.
34:18–35:20
But it takes uh you know, a couple of steps. First, you need to have a champion uh or a group of champions in the organization able to show them that, you know, not everything's perfect and this is how we can make it better and this is the differentiator, and then have that conversation with them, and also have some clarity. Uh so that there's not a lot of folks who work on, you know, that you know, product side of things. In healthcare organizations, we always have IT people, and we have uh clinical informatics people, and security people. Those are the three folks that are always there. Like in even the CTO is usually someone who used to work with the EMR, and then went up and became kind of the CTO over time because they were able to integrate themselves into the RCM and P&L and all that stuff. So that savviness brought them to the place they were supposed to be. Uh but it So but having the the knowledge on the this AI tools, it takes takes a village. Um uh because that
In discussing ethical oversight, he calls for explainable decisions, attention to training data and local differences, and continuing human involvement, positioning technology as support for patient care rather than a replacement for clinicians.
28:01–28:19
So, is AI my new doctor? No. Pretty smart, but you know, still pick up the phone and and talk to patients and take pick up the pager. Once again, we need this collaborative intelligence together, but this this really helping us to be better and be focused on the on the patient in front of us.
37:43–38:36
the the to me the ethical component seems really important. We need through like three major things. One is explainability. We need to know the data provenance and how the decision was made by any tool, no matter what, you know, either is just RCM or really decision support. We heavily regulated FDA side of things, which is completely different in terms of of of what they should be label for. Like, there there should be different levels of scrutiny based on that. The other part is making sure that we understand the provenance of the of the data set, where it was trained because the there's a lot of real bias, you know, they I know people say garbage in garbage out, it become almost like a you know, what whatever one says that, it's easy to say, but the reality is more like poor data in bad output out.
39:45–40:30
breast exam. But what if the patient has breasts that are more dense compared to the European one where the models have been deployed? So, we need local models, and that's part of the ethics component as well. Understand that provenance. And then lastly, making sure that we's always going to be human in the loop because you know, no matter how good algorithm is going to be, a patient is not going to be making decisions based on an AI tool. I can I have a hard time believing that, not because I'm not an enthusiast in AI, but because I see my own patients believing in certain drugs and things that are not really effective just because it's been told by a human. So, so I don't believe a a bot's going to help. Yeah.
The original conversation
Watch. Explore. Go deeper.
- Featuring
- Arturo LoAIza-Bonilla · Glenn Tyranski
- Running time
- 43:15
- Program
- 1ArtificialIntelligence
- Video source
- YouTube
- Published to YouTube
Reference this session
AI Transforming Health Operations | Arturo LoAIza-Bonilla — Arturo LoAIza-Bonilla · Glenn Tyranski. 1BusinessWorld Watch.
https://1businessworld.com/watch/ai-transforming-health-operations-arturo-loaiza-bonilla-NoFuuu4SdLM/Having trouble with playback? Watch on YouTube.
THE PEOPLE BEHIND THIS CONVERSATION
Explore their perspectives.
EXPLORE THE CONVERSATION
Read. Find. Go deeper.
Read the transcript Automatic captions
Automatic captions may contain transcription errors. Select a timestamp to open that moment.
0:10Hi everyone and welcome back to One
0:12Artificial Intelligence. One Artificial
0:14Intelligence brings together the leaders
0:16shaping the evolution of artificial
0:18intelligence across all industries. It
0:20is a platform for innovators, pioneers,
0:23and decision-makers
0:24who are advancing AI and translating its
0:27promise into measurable performance,
0:30long-term advantage as well as
0:32responsible growth. So, thank you for
0:34joining us here in the One Business
0:35World platform and our conference both
0:37virtual as well as in New York. Uh and
0:40it's all about building community. So,
0:42thank you for being part of our
0:43community. Uh it's now my great pleasure
0:47uh to welcome to uh to our conference
0:49and our platform Dr. Arturo Loaiza
0:52Bonilla. He is the network chief of
0:55hematology and oncology for St. Luke's
0:58University Health Networks. Uh usually
1:01in the greater Philadelphia area today
1:03tonight joining us from Colombia. Uh so,
1:05Dr. Arturo, welcome. Uh good to see you.
1:08Uh I know you've had a long long travel
1:10day already. Uh
1:12I thought I thought mine was longer. You
1:13beat me. Uh
1:15I'm in a space as you can see now. So,
1:18That's it. That's it. So, welcome. Let
1:19me turn it off over to you. We look
1:20forward to your talk and your
1:22presentation. Thank you so much. All
1:24right. So, uh hi everyone. Um I'm Dr.
1:26Arturo Loaiza Bonilla as was mentioned
1:28by Glenn and it's a pleasure for me to
1:29be here today. Uh as a medical
1:31oncologist, I always want to have a
1:33perspective where I can uh you know, put
1:36into a couple of views. One from my
1:38medical oncology view, uh the other one
1:40from my administrator side, and in a the
1:43grander scheme of things on uh
1:46entrepreneurship and see how operations
1:48are changing across. So, so that's how
1:49we're going to have this conversation
1:51today. Uh so, I call it from pilot to
1:53production, the 2026 operational
1:56revolution using AI.
1:58Uh for those who want to, you know,
2:00follow up after the meeting or they want
2:01to, you know, go online, I'm in Twitter
2:03X uh
2:05in LinkedIn. Um it's called Dr. Arturo
2:07AI, no pun intended. Uh and uh you can
2:10always talk uh about, you know,
2:12different topics that you're you find
2:13interesting. So, I'm always happy to
2:14engage with you there.
2:17Uh I'm also a faculty at Temple
2:18University and an associate professor
2:20there as well.
2:23So, we come a long way uh since the
2:25advent uh of the first uh you know, AI
2:28algorithms. So, uh so for some people,
2:31this is a brand new topic, but uh
2:33as you may know, uh we have been
2:36experiencing this revolution is slowly
2:37but but steady is getting just
2:39exponential in terms of opportunity, but
2:42we we've been talking about AI from
2:44actually, you know, 1764 we're talking
2:47about about the, you know, foundations
2:48of probabilistic inference uh and
2:51talking about the Boolean logics of the
2:52zeros and ones, yes or no's. Uh and we
2:55even have a time uh the first algorithm,
2:58which is was by, you know, managed by
3:00Ada Lovelace in 1842. So, fast forward,
3:03right? So, we have uh uh Jeff Hinton
3:06talking about deep learning. Uh we
3:08started looking doing pattern recognition with ImageNet in 20 uh in in
3:112009. Uh the competition was overcome
3:14by, you know, Fei-Fei Li and then she's
3:16doing now her world models. And then
3:18Google and others started to work with
3:20uh those uh LLMs uh and the transformer
3:24came to fruition in 2017
3:27uh with the attention on all you need
3:28paper. Then biology transformed as well
3:31by AlphaFold, Nobel Prize prizes were
3:33won in 2024.
3:35Uh and and all the same time we had
3:37ChatGPT and many other LLMs coming at
3:40the same time
3:41uh taking it uh the world by storm. And
3:44now the future is multi-agent AI. And
3:46that's exactly what we said today just
3:48to have a conversation of how we using
3:50AI tools in healthcare particularly with
3:52a lens from me as a medical oncologist
3:55and and
3:57you know integrated delivery network
3:59mindset. So, we'll give you some
4:00examples of how this is happening in the
4:03field and of course it's not going to be
4:05comprehensive. It's more about to have
4:08an overview of what is happening in the
4:11space.
4:12Just to give you a little bit of my more
4:14background,
4:15you know, 22% of health organizations
4:17are now using some form of domain
4:19specific AI
4:20which is a major increase about seven
4:22times
4:24compared to 2024. So, it's it's really
4:27picking up steam across different
4:29organizations in healthcare.
4:31The
4:321.4 billion has been spent on that about
4:36two times two times the rate of
4:38deployment compared to the broader
4:39economy and 66% of United States
4:42physicians actively using some form of
4:44AI tools in practice.
4:46So, this is really something that is
4:49taking over.
4:51But now the question is how are we going
4:53to implement this in a broader scale,
4:56right? Because we we have all these
4:57tools, all these opportunities, but
4:59there is this paradox where 83% are
5:02pilots, less than 10% there is an
5:06investing in the real infrastructure for
5:08deployment. So, people are just kind of
5:10trying solutions, but but they really
5:12haven't put a full
5:14end-to-end approach to the enterprise
5:17level of this and and and I think this
5:19is a good time now to stop the
5:21experimentation. We know these tools
5:23have some you know really major
5:25opportunity here and the challenge is
5:27causing from you know pulling from the
5:29pilot to actually production to how we
5:31can make this really meaningful.
5:33And
5:34I'm going to put you in three buckets
5:35here.
5:36So, the first one is you know the
5:37economic flywheel, the undeniable ROI
5:40that we have seen with this
5:42AI deployments. The second one is the
5:44clinical revolution. So, from me as a
5:46clinician, uh I'm going to give a use
5:48case in precision oncology and the
5:50things I've been doing as co-founder of
5:51Massive Bio,
5:53and also as leader at San Lucas. And
5:55then lastly, the human bottleneck and
5:57how we can transform workforce uh as the
5:59ultimate metric of success.
6:02So, what is Genera- GenAI? We all have
6:05an idea of generative AI throughout many
6:07conferences and and things we're
6:08learning, but as uh as as you many of
6:11you you you know, this is a an area of
6:13machine learning where we use
6:15unsupervised learning and those
6:17generative models have three components.
6:20One is the architectural transformers,
6:22second, a ton of data, which is a data
6:24set uh in uh for training, and then the
6:27prompting, which is asking the right
6:28questions.
6:29And uh all based on this structure of
6:32the transformer, which really comes from
6:34this paper and we mentioned it was
6:35actually pre-print uh and it com- it's
6:38called transformers because uh at the
6:40time they were the prior project was
6:42Autobots. So, 100% cartoon-based. So, in
6:45in science fiction. Uh but all these
6:48LLMs are coming from the ability for um
6:51what we call the generative pre-trained
6:52transformer, that's where GPT comes
6:54from, and it's able to predict the next
6:56step based on probability. So, if you
6:58have a ton of data and you have a ton of
7:00examples, it's likely that you will know
7:01what is the next step. Uh and and that's
7:03what we can do for translation, in
7:05images, in
7:07uh operations, understanding coding,
7:10understanding language, all that is
7:12coming across.
7:13And um we have learned from diff- very
7:16uh incipient efforts in healthcare of
7:18these uh that uh you know, there when we
7:21do a combination between using these
7:23LLMs along with a human being, we can
7:26actually optimize the way that patient
7:28care uh is delivered more efficiently.
7:31Uh so
7:32So, how do we start with this? Uh what
7:34is the multiplier effect? And and the
7:35input here, you know, for each dollar
7:37invested, uh it's estimated that uh over
7:40a 14-month maturation, uh the output is
7:43about 3.2, uh, dollars tangible return.
7:45So, and this is a flywheel effect,
7:48right? You keep optimizing, you keep
7:49getting better.
7:50Uh, so, 85% of healthcare leaders report
7:53they're actively, uh, increasing in
7:55revenue directly because of these
7:56integrations. Uh, so,
7:59uh, it it this is just the beginning of
8:01it and and how we're optimizing, uh,
8:03it's it's up to us to decide.
8:06So,
8:07uh, I'm going to give you all these, you
8:09know, legacy friction
8:10points and the AI enabled velocity that
8:13we see. So, uh, you know, RCM, or
8:15revenue cycle management, uh, we have
8:18all these issues with prior
8:19authorization and denials. 60% of
8:22denials are never appealed because of
8:23attrition game. We just simply change
8:25things and say we don't want like when
8:27we talk with insurance companies or even
8:29Medicare, we're like, "Okay, we got a
8:31denial, we're going to write it off."
8:33Uh, now with AI, we can reduce, uh, the
8:37the, you know, cost to collect using,
8:38for example, integrated EMR solutions.
8:41Uh, prior authorization, the same thing.
8:4392% of physicians have to go on the
8:46phone like myself and try to do appeals.
8:48Uh, now we can do 90% auto approval
8:51rates if we have an integrated AI tool.
8:53Uh, and that's something that, for
8:54example, Cohere Health and Highmark have
8:56been accomplishing.
8:58Uh, and with that also came out 85%
9:00reduction in the amount of people
9:02involved, uh, in those, uh, prior
9:04authorization issues. And then supply
9:06chain, right? So, we the supply chain
9:08comes not only in healthcare, but in
9:10many spaces, but in healthcare it's
9:11unique because of the complexity,
9:13uh, but it's usually very reactive, uh,
9:15and can lead to clinical shortages. Here
9:18with AI, we can predict, uh, what if
9:21stockouts are happening, so using
9:22machine learning, so we never run out of
9:24the things that are essential for
9:25operations, either a medication or a
9:28piece of equipment or or a or a spare
9:31part, right? So, that's all part of the
9:33whole process and giving you alerts for
9:35it.
9:37And, uh,
9:38and as you may know in the medical space
9:41because you have all these you know uh
9:44interfaces that we go online and check
9:46if we have our laboratory data there or
9:48not and all the data input that we put
9:50as clinicians, uh the largest electronic
9:53medical records um uh companies are
9:56putting themselves into the AI era. So,
9:58they're embedding AI into bunch of
10:00workflows. Sometimes kind of like
10:02creating co-creating with different
10:04companies. So, for example, the ambient
10:05AI that uh Epic was doing uh with a
10:08bridge and Dax Co-pilot and then after a
10:11while Judy Faulkner, you know, knowing
10:13how she is, she learned the ropes and
10:15then they launched their own um uh
10:18ambient AI.
10:19Uh but they already have in Epic uh with
10:22um major leading market position about
10:25uh 54% of all beds in the United States,
10:28about 200 AI features in development. Uh
10:31and they have Emmy for patient-facing,
10:33they have Penny for revenue cycle. They
10:36have Art, which is the uh you know,
10:38additional one for decision support to
10:40answer questions during inside of an LLM
10:42instance in the in the system, which I
10:45actually uh have used myself even though
10:47my cell is you know, Arturo, so I guess
10:50Art is part of that. So, yes, I use it,
10:52too. Uh and they have also made a system
10:54scale deployment with all this
10:56information they had for many years of
10:58collection. So, they have this thing
10:59called Cosmos, which has about 300
11:02million records. And and Comet, which is
11:0416 billion pre-trained events.
11:07Uh Oracle on its end is also doing the
11:09same. They acquired Cerner, which also
11:12has significant amount of market share.
11:15Uh they're cloud-native now, try to do
11:17voice first agent EHR next gen, seem a
11:20very nice on top of it, talking about it
11:22everywhere. Uh the idea is to decrease
11:2430% in data documentation time. They had
11:27generated a bunch of data on clinical
11:29notes.
11:30And they also have clinical agent and
11:32deployed across uh many specialists
11:34including oncology. Uh
11:37and lastly we have FDA uh where they
11:40have approved about uh over 1,400 uh AI
11:44ML devices cleared
11:46uh with increase of 230 50% in the next
11:495 years. So
11:50uh so this is just crazy amount of
11:52information that we're getting.
11:55So let's talk about some use cases. So
11:56ambient AI
11:58that was the first healthcare first
11:59massive adoption win. So for example
12:01Kaiser Permanente they use a bridge uh
12:03they integrate it directly with the EMR.
12:06Uh it was 40 hospitals, 600 offices uh
12:09and they in the first year they saved
12:11about 15,691
12:13hours uh in time spent typing and just
12:16simply having the listening
12:18uh and transcription in the system uh
12:21with uh better communication with
12:22patients.
12:24Uh they uh but you look across is not just
12:27uh you know Kaiser many other systems
12:29are doing this. about $600 million in
12:31revenue with two 2.5 times grow grow
12:34year over year.
12:36Uh and most of them being a bridge
12:38uh nuance tactical pilot and other
12:40ambience um
12:42uh with reduction in burnout. Uh how
12:45that translates into operations uh it's
12:47it's clear and the burnout is still in
12:49the making just personal point here but
12:51we'll see.
12:53Uh because there's so many of them.
12:55So uh just in this little table you can
12:58see uh that um the ambient scribes are
13:02just far too many for my own taste. Uh
13:05but they're just trying to show
13:06differentiation and try to be the best
13:08possible. So let's see who wins at the
13:10end. Uh but we we have a couple of good
13:12players in the space.
13:16Uh and and this is just as as I
13:17mentioned earlier uh the effort for
13:19Kaiser Permanente uh actually helped a
13:21lot.
13:24Um the next uh thing is can we use this
13:27data acquisition in the medical record
13:29beyond just simple listening?
13:31And yes, we used to be using OCR optical
13:34character recognition and NLP, which is
13:36natural language processing with
13:37generative AI for collection of data and
13:40for coding, right? So we have coding
13:43because we can translate the medical
13:45record into an ICD-10 code. And many
13:49players now are trying to get into the
13:51prioritization and coding by just simply
13:54integrating themselves into the
13:55multi-modality of the electronic medical
13:57records. So they're trying to be EMR
14:01native. And many sure to complain to
14:03combine lab data, clinical notes,
14:06genomics, monitoring data, and even
14:08medical images to create these
14:10foundational models so they can deploy
14:12them for decision support.
14:15So one example of things that we're
14:17doing in medicine
14:19you may not be aware of is the seeing
14:21the unseen, right? So we not only
14:23generative AI is a probabilistic
14:24approach which sometimes tends to
14:26hallucinate. So when you talk about
14:28operations in radiology for example to
14:30optimize care, we use the concept of
14:33locality, which is basically using
14:35still machine learning efforts, but not
14:38with gen AI, but more for for example
14:40with these convolutional neural
14:41networks, which are
14:44able to do a higher accuracy level
14:46because they're based on
14:48non-generated images. They're all you
14:51know, actual data sets and recognizing
14:54the the patterns themselves.
14:56So in this case we're trying to divide
14:58into kernels and put it into these
14:59algorithms and the deep learning
15:01approach to differentiate between
15:03malignant, benign, and granuloma or
15:06something else that is different in this
15:08case
15:09with a lung nodule.
15:11But we can use the same mathematical
15:12equations for many other things
15:14including x-rays for fractures and for
15:16other findings. And if you do the three
15:19layers of colors you do red, blue, and
15:21green, you do the same mathematical
15:23calculations. You can see the different
15:25densities have different numbers. And
15:27and that those kernels can be computated
15:31and put together to give us an example
15:33of of
15:35IXM and detect things that
15:38could be helpful for screening and for
15:41making our operations faster as well.
15:44Uh and an example here is
15:47screening for cancer. So, AI is now
15:50outperforming in radiology detecting,
15:52for example, pancreatic cancer, which is
15:53very hard to find earlier
15:56even when we have,
15:58you know, family history and with the
15:59screening, we don't have it easy to
16:02detect those ones. So, in this case, AI
16:04is able to find about 38% fewer false
16:07positives with AI with very early signs
16:09and subtle subtleties that the naked eye
16:11may be hard to determine. So, we'll be
16:14we're becoming better because of it.
16:18Breast cancer research, which is a
16:19conglomerate of multiple organizations
16:21working together and using them for
16:23optimization of clinical operations on
16:24research. So, instead of making caliper
16:27calculations and looking at the you know
16:29with a little, you know, measurement on
16:31on those nodules to estimate if that
16:33treatment is working for patients,
16:35we can use these criteria from
16:39the AI tool to help us to assess
16:41responses much more efficiently.
16:43And another one is in radiation
16:45oncology. So,
16:47we have to do changes all the time based
16:49on the patient anatomy or the patient
16:51has to breathe and you want to make the
16:53beam of radiation very focused.
16:55In the past, we used to take several
16:57days. Now, we can do in a matter of
17:00minutes a full re-planning for a
17:02patient.
17:05So,
17:07moving forward
17:08in the transformation journey now in the
17:10clinical side,
17:11I love to talk about clinical trials and
17:13how AI can transform this. So, this is
17:17just an example of about the last decade
17:20and analysis where 12 million patients
17:22in the National Cancer Institute
17:23database.
17:25Basically fewer than 0.1% ever enroll in
17:28the trial. And even places that have all
17:31the mission in the world, they have, you
17:32know, they're only research centers,
17:34they do between 5 and 15% enrollments of
17:38their patients. And that's their only
17:39mission. That's why they get all the
17:41dollars philanthropy and we're not. And
17:43it's not because they don't want to.
17:44It's just because, you know, first of
17:46all, 85% of the patients are treated in
17:48the community. So, operations are
17:50happening in, you know, systems that are
17:53more remote and not just simply in
17:56in academic enclaves. And and also
17:59because, you know, time is of the
18:00essence. So,
18:02we have all these issues with, you know,
18:04interfaces
18:05where we don't have
18:07all the data surfacing in front of us.
18:10We have to deal with authorization,
18:11authorization fatigue, time-bound
18:13priorities, and sometimes we even miss
18:15biomarkers that the patient may be
18:16eligible for and can change their lives.
18:20And those are commercially available.
18:21So, it's not that we don't have the
18:22testing or the drugs. We just don't have
18:24the connectivity together. And this is
18:26mostly because of this phenomenon of
18:28data drift. As everyone can imagine, we
18:32all as humans change. You know,
18:34the our conditions change over time, our
18:36cancer progresses or stays the same, our
18:39conditions are always in a constant
18:40evolution and the same the environment
18:42around us in terms of where we are being
18:44treated, what physician is in front of
18:46us, what labs are being drawn at that
18:48moment, and the standards of care that
18:50evolve along with that. So, we have to
18:53think about this more strategically,
18:55more for like logistics.
18:57So, Uber, Instacart, that we need
18:59real-time awareness for availability.
19:01And that's where I came about working
19:03with several innovators and in MassBio
19:05we built this multi-agent architecture
19:08called Synergy AI. And and basically we
19:11put agents that are able to do LLM
19:13extraction on oncograph which is a
19:15knowledge base and give recommendations
19:18for
19:19clinical trials in real time so we can
19:21get a single targeted patient match in
19:24in a matter of seconds.
19:26We showed this at the American Cancer
19:29you know American Society of Clinical
19:30Oncology meeting and we validated about
19:333,104 patients matched with Synergy AI
19:36and we were able to do the job of about
19:39you know 5,000 patients and which would
19:43have taken about you know 50,000
19:47number of work hours in just an hour
19:49right so so the the benefits are legit.
19:54Uh
19:55So the next frontier is to move from
19:57generation to orchestration right so
19:59instead of asking just prompting we need
20:01to use multi-agent AI where the agent is
20:04able to collect database information and
20:06connect with autonomous go driven
20:08systems that able to schedule patients
20:10to get them where they're supposed to be
20:12either for regular standard of care
20:13appointment or for getting the billing
20:16ready right so the reality check is 80%
20:19of executives expect value but only 3%
20:22of agents are deployed in live workflow
20:24so how are we going to see the results
20:26if we don't have that put together?
20:29Um and
20:30and the idea here is this is going to
20:32help also on the revenue cycle side is
20:34scheduling care coordination supply
20:36chain as we mentioned earlier.
20:39And a couple of health systems are doing
20:40so effectively. So if you want to have
20:43some examples one of them is being Mayo
20:44Clinic which is now I think this year
20:47was classified as the you know best in
20:49the world
20:50they're really doing a good a lot of
20:51good job on AI I have to give them that
20:53for sure and have very good friends
20:55working there.
20:56The other one is being Cleveland Clinic
20:58being
20:59themselves very well positioned for that
21:02but Mayo Clinic has invested about a
21:04billion dollars and they are building
21:07this platforms
21:09to really set out the foundational
21:11models of the future
21:12and do it in a further the learning
21:14approach as well. And coming to clinic
21:16is just saving in EMR integration in the
21:19amount of time efforts
21:21doing documentation and all the
21:24connections basically attached to the
21:26you know the scribes encoding system.
21:31So they there's other health systems as
21:33well. So HCA Healthcare 53% reduction in
21:36charge entry time, 90% nurse approval.
21:39The vascular medicine 5% length of stay
21:41reduction, which is a pretty remarkable
21:43one because we always want to decrease
21:45length of stay for complexity.
21:47Providence 6,000 surgical cases
21:49additionally added because of
21:51optimization on the OR. Common Spirit
21:54did five times increased gap closures in
21:56the AI screening for cancer.
21:59Tampa General was able to deploy a
22:01sepsis detection
22:03with about 100 million savings
22:05operational
22:06effort. And UC San Diego 70% reduction
22:08in sepsis as well using these variables
22:11for
22:12detection of early um
22:14presence of of
22:16infection in in sepsis.
22:19In prioritization, we're seeing
22:22that you know CMS is really invested on
22:24this. They have this wiser pilot
22:27where they're trying to find
22:28prioritization automatable
22:30with health systems for certain
22:32diseases.
22:33And we know that we need to make it
22:36better because there's so much to spend
22:38on prioritization and
22:41AI is taking forefront of it. So the
22:43battle of the bots are there.
22:45So I'm here to see it.
22:49So
22:50in this case I think we need to look at
22:52this into as a way to plan. You know
22:54this is a plan chart architectural. So
22:57how can do allocation of resources? So
22:59first we have to see how we can put the
23:02patient predict you know predicting the
23:04patient flow. So we can do mathematical
23:06forecasting
23:07across all the different service lines
23:09and things we're trying to accomplish.
23:11The rest of medicine did some of that as
23:13I mentioned with the 5% reduction in
23:15length of stay.
23:16And and that basically increases the
23:19opportunity for so many patients to be
23:21able to be treated and also
23:23bring back revenue so we can keep
23:25investing in those solutions as well.
23:29So the two battleground here is
23:31transforming the workplace, right? So in
23:33the in the next seven minutes or so that
23:35I have we'll talk about how we're going
23:39to be able to transform organizations on
23:41this. So because we had 10% algorithms
23:44already there.
23:4520% is data and technology
23:48that is being deployed but 70% is
23:50processes which is you know is the way
23:52it is we're humans at the end, right? So
23:55we need to have a lot of reskilling. 60%
23:57organizations reports is reskilling is a
24:00top challenge.
24:02Security as well. So I have a very nice
24:05chief security officer at St. Luke's but
24:07you know is is hard to get every single
24:09solution in front of that team because
24:12they're always you know very cautious
24:14rightfully so
24:16and and and certain algorithms have
24:19a level of scrutiny compared to others
24:20but no one's putting AI in the single
24:22umbrella and and that reskilling has to
24:25happen as well.
24:27So but at the end of the day AI greatest
24:29operational impact will be in restoring
24:31humanity, right? So
24:32we're so
24:34burned by so many administrative stuff.
24:37So as we mentioned for every dollar that
24:40we invest we get $3.2.
24:4374% lower odds of burnout with ambient
24:45AI 70% improvement with processes using
24:49AI as well.
24:51And I always look it into the moral of X
24:53paradox which is Um, this
24:57you know, mindset where AI
25:01can do a lot of things that we can
25:02relinquish to it. You know, doing things
25:05that are very recent, right? So,
25:06language and
25:08mathematical equations that are very
25:09recent in evolution of human beings, but
25:12we took
25:13you know,
25:14uh
25:15Earth took forever, right? Millions of
25:17years and humans to evolve to be
25:21inherently good at accepting for nuance,
25:23empathetic interactions and complex
25:25physical coordination, being able to do
25:27things. So, the vision here is to
25:29relinquish computation to the algorithm,
25:31bring back the empathy and the
25:33revolution to really be humans in front
25:37of everyone. So, that that's that's what
25:38the the the dream is here with the more
25:40of the spiral logs, the collaborative
25:42intelligence as I call it.
25:44Uh, how can we do this? Being very
25:47patient-centric. So, we need to make
25:49Uber, Amazon, Instacart, but for
25:51clinical operations, clinical trials,
25:53clinical operations overall,
25:55because manual methods right now they
25:56really fail
25:58and we need a hybrid click and mortar
26:00approach where the patient is digitally
26:02enabled with tools and then we get them
26:05to the point of care
26:06for with this AI prescribing hub. So, we
26:09monitor over time and find a solution.
26:12One example is working right now is with
26:15Massive Bio, we partnered with American
26:17Cancer Society.
26:18Any cancer patient in the United States
26:21can get access to this program for
26:23clinical trials
26:25for really free to them and their
26:26physicians. We do an AI based clinical
26:29trial report for those patients
26:31using basically extraction of the
26:34information, being able to analyze it
26:36completely and we get all the support
26:38for education, for trial eligibility,
26:40for monitoring, for transportation, for
26:43lodging and this is working at scale in
26:46the whole United States. So, this is a
26:48model that needs to be replicated across
26:50not only clinical trials, but operations
26:53as a whole, but I always tell people
26:55please make people aware of this project
26:57because something that is really
26:58meaningful for everyone and we're going
27:00to change and transform our clinical
27:02research for oncology and beyond.
27:06How? With this flywheel, right? So,
27:08patient centered the hub in oncology, we
27:10can use this for clinical trial
27:11matching, we can check how drugs are
27:13being tested and and used and and do
27:16real-time analytics, we can activate
27:19clinical trials in point of care, we can
27:21do digital enablement and education, we
27:23can optimize biomarkers, so we can
27:25improve the process of the future, build
27:27this foundational models further and
27:29eventually get even better drugs for our
27:31patients and understanding things in
27:33real-time using the data that we have.
27:37We can do it safely with federated
27:38learning where we can just deploy the
27:40models and the weights in the local
27:42instance and it can transform healthcare
27:44as well by enabling large-scale
27:47privacy-preserving collaborative AI
27:49deployments across institutions
27:51is something that I'm very passionate
27:53and working on every single day so we
27:55can transform the way we deliver care
27:58at the global scale.
28:01So, is AI my new doctor? No.
28:04Pretty smart, but you know, still pick
28:06up the phone and and talk to patients
28:08and take pick up the pager.
28:11Once again, we need this collaborative
28:12intelligence together,
28:14but this this really helping us to be
28:15better and be focused on the on the
28:18patient in front of us.
28:19So, what are the trends that are going
28:20to define 2026 and beyond? So, one,
28:24ambient AI is going to become really
28:26table stakes. Everyone's going to get
28:27it.
28:28So,
28:29second one is revenue cycle. AI is going
28:32to scale and and those who are not doing
28:34it are going to fall behind.
28:36So, it does the highest ROI category so
28:39far,
28:40optimizing workflows.
28:42The third one is the agentic AI is going
28:45to move from pilot in not many places to
28:47real production in those health systems
28:49that actually taking it seriously. And
28:51they look no further, you know, as I
28:52mentioned,
28:54uh
28:55Mayo Clinic and and uh Cleveland Clinic
28:58and St. Luke's also uh we're doing the
29:00same transformation of agent TKI and
29:02we're collecting data using different
29:04bio and bill uh approaches.
29:07Um and we're doing also from Mass
29:09General Bio particular Nexus. AI
29:11governance is becoming critical, so we
29:14need to, you know, this we need to wake
29:15up and also
29:17tell uh you know, the those who need to
29:19be reskilled to wake up to that, you
29:22know, the
29:23uh the old era of software
29:25implementation is over. We need to be
29:27focusing now on implementation and in a
29:30fast rapid scale.
29:32Uh and the real battle is just that,
29:35right? So, making sure that 70% of
29:37people uh was the people and processes
29:39understand what at stake and how we can
29:41use it to be better and and be more
29:44efficient across and bring those great
29:46investments into the organization so
29:48people see the value behind it.
29:51For those who want to learn more about
29:52how approaches in medical oncology and
29:54beyond,
29:55uh there's this chapter in the ASCO
29:56educational book that um
29:59uh that uh we wrote about the
30:01multidisciplinary oncology in
30:02transformation and also HCS has some
30:04more of the information as well. So,
30:06feel free to access.
30:09So, uh this isn't futuristic, it's
30:10actionable now. So, the idea is
30:12planetary AI and operation hubs. Um but
30:15always patient first, right? Patient
30:17first network for me is the key thing.
30:18So, we can activate
30:20uh real-time operations at scale uh
30:23from uh from command centers and help
30:25the patients uh physicians and everyone
30:27else involved.
30:29So, the knowledge is uh driving
30:31knowledge to action is what we always
30:33say. Uh so, it's no longer a future,
30:35it's something that we need to
30:36operationalize now. Let's build it
30:38together. Uh for more conversations to
30:40have, you know, um I have the HGS access
30:42here for people to go and check it out.
30:46Uh, and and also and then my LinkedIn is
30:50listed there for those who want to
30:51follow me there, too, and and have a
30:53conversation. Happy to
30:55uh, talk to you.
30:57And with that, I thank you for the
30:58opportunity and time. I always finish
31:00with this Quattro de Opus Speciale,
31:01which is this first AI art, uh,
31:04um, work that actually won, uh, an award
31:07and it's a good combination of, you
31:09know, art and science and and, you know,
31:12the the human part of it, but it's also
31:14showing this beautiful future in front
31:15of us where we can orchestrate, uh,
31:18uh, things together. We can really have
31:20a bright future in front of us. Um, so
31:22thank you for the opportunity.
31:24Thank you, Arturo. Great great job. What
31:26a beautiful photo there at the end.
31:28That's absolutely so special.
31:31Uh, tha- thank you for your expertise.
31:33Thanks thanks for the presentation. Uh,
31:36seems like endless possibilities.
31:39You know, um, I mean, the obvious ones,
31:42you know, we always, you know, I'm I'm a
31:44I'm a finance guy and accountant, so the
31:46obvious ones are always things like
31:47revenue cycle, right? Supply chain,
31:50pre-authorization, and those and some of
31:52some of the statistics and stuff that
31:54you mentioned there and the
31:56advancements, um, are are obviously very
31:58very special.
31:59The clinical part always, again, is is
32:02is, you know,
32:03challenging, but so much promise as
32:06well. I I loved my favorite slide you
32:09had other than that last one.
32:10Um,
32:11maybe we're restoring the the humanity
32:14trust lost in all of the administrative,
32:17you know, stuff and and that picture you
32:20had of the of of the doctor staring into
32:23the patient and looking at the patient,
32:25right? That's always been,
32:27you know, a lot of what we what we've
32:28talked about or seen, right? It's with
32:30the advent of electronical medical
32:32records and, you know, that they are
32:34they are a godsend,
32:36but we wind up becoming very much data
32:39driven and then the element of being
32:41able to
32:43look at the patient and
32:45you know, build that confidence.
32:47If we're able to get that back and if
32:49all of this helps us with that,
32:51what a what a what an advancement that
32:53would be, right?
32:55Incredible. >> 100%.
32:56>> Incredible. >> 100%. Any presentation that has Alan
32:58turning in it, I'm always impressed
33:00with.
33:01So, thank you for reminding people that
33:03even though this may we may think this
33:05just came out of left field, there is a
33:06rich history. Sometimes it's harder to
33:08see and it's a little bit circular as as
33:11you as you as you had on that chart, but
33:13that that's always uh always
33:15fascinating. Um I mean, the ROI numbers
33:18you had, a dollar into 320 over 4
33:20months,
33:22that's that's pretty impressive as well,
33:24for sure. When you work with whether
33:26it's St. Luke's or or just uh
33:29you know, health care systems in
33:31general,
33:33how much how much direction are are you
33:35seeing or how much openness are you
33:36seeing from say like the boards, like
33:40the board of directors,
33:41uh or again, the medical boards or any
33:45any perhaps any any task force or
33:47committees in this, how much of a you
33:49know, how much of a green light do you
33:51get from that? How much how much
33:52oversight or direction or do they just
33:55say, you know what, go make it happen?
33:58How you how how do you how do you how do
33:59you deal with those interactions? What's
34:01been your experience? >> Yeah, it
34:02it's uh it's an interesting um
34:04environment because
34:06uh
34:07you know, uh
34:08folks who are in operations and and
34:10administration, you know, they're data
34:11data driven. So, if you show them the
34:13data and you show them how this is going
34:15to be meaningful, they buy in.
34:18But it takes uh you know,
34:21a couple of steps. First, you need to
34:22have a champion uh or a group of
34:24champions in the organization able to
34:26show them that, you know, not
34:28everything's perfect and this is how we
34:30can make it better and this is the
34:32differentiator, and then have that
34:33conversation with them, and also have
34:36some clarity.
34:37Uh so
34:39that there's not a lot of folks who work
34:41on, you know, that you know, product
34:43side of things. In healthcare
34:45organizations, we always have IT people,
34:48and we have uh clinical informatics
34:51people, and security people. Those are
34:53the three folks that are always there.
34:55Like in even the CTO is usually someone
34:57who used to work with the EMR, and then
35:00went up and became kind of the CTO over
35:02time because they were able to integrate
35:03themselves into the RCM and P&L and all
35:07that stuff. So that savviness brought
35:08them to the place they were supposed to
35:10be.
35:11Uh but it So but having the the
35:14knowledge on the this AI tools, it takes
35:17takes a village. Um uh because that
35:20sometimes they're scattered. So uh but I
35:23think having forms like this like we
35:25have in today, and and having the uh you
35:28know, I participate in the Beckers uh
35:30Beckers Healthcare Network and the
35:33Beckers Beckers of Folio Forum. Every
35:35month we're having, you know, reach outs
35:38and meetings where we actually
35:39discussing these things and and showing
35:41them how places like in Mayo Clinic and
35:44Cleveland Clinic, they're really taking
35:45it super serious. Like extremely serious
35:48to the point that you see, "Okay, if
35:50these are like the best hospitals in the
35:51world now going all in and investing all
35:54their money into it, there has to be
35:56something there, right?" So uh and and
35:59and when you show these examples, that
36:01usually resonates with boards and with
36:03folks that sometimes may, you know, in
36:05certain places they may be very
36:07parochial and say, "No, we just like our
36:08community." It's like, "Yes, but you can
36:10use AI to serve that community better,
36:13reinvest those dollars into programs so
36:16you can, you know, get even bigger
36:18market share and bigger support, and and
36:21you still feel that you are helping the
36:23community that you were so focused on
36:25and very well driven for, so.
36:28How about from
36:30anything of health care, you think of
36:32code codes of conduct. How often do you
36:34get
36:35asked about the ethical
36:37ramifications of of artificial
36:39intelligence in health care?
36:42Yeah, so it depends on the stakeholders
36:44that you're talking to. So, if you talk
36:46to clinicians, it's always there, right?
36:49So, because we are I mean, I'm not that
36:52bad but yeah, there's a lot of people
36:53who super risk averse and and technology
36:56they they don't like technology because
36:57technology has not been our partner for
36:59many years. It's being always a
37:01challenge. Like, how many more clicks do
37:03we have to do to get a bill or to get a
37:06patient treated when the past we just
37:07simply grab a little script and send it
37:09out and get the patient treated in front
37:10of us. Now, we have to go to all these
37:12different interactive processes and
37:14prior authorization and the ethical
37:16component comes with that, right? So,
37:18I was just at the NCCN meeting, which is
37:21National Comprehensive Cancer Network
37:22meeting and it was the United um kind of
37:26chief medical officer and she basically
37:29because we're saying I'm going to get
37:30denials by the AI, right?
37:33And and they said, "Well, no, we promise
37:35this we're not going to
37:36issue
37:38denial based on AI because we understand
37:41the stakes here." So,
37:43the the to me the ethical component
37:45seems really important.
37:47We need through like three major things.
37:49One is
37:51explainability. We need to know the data
37:53provenance and how the decision was made
37:55by any tool, no matter what, you know,
37:57either is just RCM or really decision
38:00support. We heavily regulated FDA side
38:02of things,
38:03which is completely different in terms
38:04of of of what they should be
38:08label for. Like, there there should be
38:10different levels of scrutiny based on
38:12that. The other part is
38:16making sure that we understand the
38:17provenance of the of the data set, where
38:20it was trained
38:22because the there's a lot of real bias,
38:24you know, they I know people say garbage
38:26in garbage out, it become almost like a
38:27you know, what whatever one says that,
38:29it's easy to say, but the reality is
38:32more like poor data in bad output out.
38:36So, because the problem is you can have
38:38data, for example, mammograms are being
38:41used now as a very used case of
38:43radiology changing. And And well, this
38:45is very very relevant because there was
38:47some news today coming up that the
38:50health systems in New York, the largest
38:53health systems were talking about they
38:55were ready to give up on radiology has
38:57been hired because they they're ready to
38:59get AI to take over the first pass of
39:01it.
39:02And And I think it's interesting, but
39:05Jeff Hinton said many years ago, I think
39:08at
39:09it was 2016 what he said on No, 2006.
39:12Then he said, "In the next 5 years, stop
39:14training radiology because AI is going
39:16to replace you all." And he had to
39:18recant and say, "No, I'm sorry, but what
39:20he didn't account for, I think his
39:22prediction was was right. But what he
39:24didn't account for was from regulatory
39:26aspects and how, you know, things take
39:29for people to adopt, right? So, and also
39:32the ethical component of it. So, yes,
39:33there's going to be some models that are
39:34going to be very good for mammograms to
39:36detect
39:37earlier than, you know, a human or maybe
39:39doing the as double instead of doing two
39:41radiologists, you only need one
39:43radiology to make the readout of that
39:45breast exam. But what if the patient has
39:49breasts that are more dense compared to
39:52the European one where the models have
39:53been deployed?
39:55So, we need local models, and that's
39:57part of the ethics component as well.
39:59Understand that provenance.
40:00And then lastly,
40:02making sure that we's always going to be
40:04human in the loop because
40:07you know, no matter how good algorithm
40:09is going to be,
40:10a patient is not going to be making
40:12decisions based on an AI tool.
40:15I can
40:16I have a hard time believing that, not
40:18because I'm not an enthusiast in AI, but
40:20because I see my own patients believing
40:22in certain drugs and things that are not
40:23really effective just because it's been
40:25told by a human.
40:26So, so I don't believe a
40:28a bot's going to help. Yeah.
40:30Interesting. Interesting. You had uh
40:33and it's probably a good thing to close
40:35with.
40:36You said from generation to
40:37orchestration.
40:39And when you think of an orchestra,
40:41right, and all the beautiful music that
40:43comes from it and all the sound,
40:46it it is a real person
40:48doing the
40:49doing doing that conducting, right? And
40:51bringing together your your workforce
40:54transformation, all of those players.
40:57And in the in the healthcare in the
40:58healthcare world, I mean, that's that it
41:01it it it is an orchestra. It it's it's
41:04it's a village, as you said. It it's
41:06it's it's the back office, it's the it's
41:08the it's the doctors, it's the
41:10the nursing staff, it's the technology,
41:13it's the equipment, um orchestrating all
41:16of that and getting it
41:18to sound,
41:19you know, perfect uh or or or, you know,
41:22wonderful wonderful to the ear. That's
41:25going to That's going to take it That's
41:26going to take a lot of strategy, right?
41:27And and and embedding that in into uh
41:31you know, hospitals and health systems'
41:33strategic plans. But it it's it's it it
41:36seems like it has so much exciting
41:38promise. And I mean, and you're you're
41:40right there at the forefront of it and
41:42and congratulations
41:44for all your all your success and
41:46your vision. I mean, and and again, with
41:49you I mean,
41:50seven times it turned over seven times
41:52since 2024, right, on the
41:54domain-specific
41:55piece you had. I mean, I mean, these are
41:57these are just
41:58giant leaps and bounds. So, thank you
42:01for being there. Thank you for
42:03where when it's needed not to be afraid
42:06to put up the guardrails a little bit,
42:07right?
42:09And to be and to be that conductor and
42:11to be of that orchestra, cuz this this
42:13this this this is so promising and we
42:15wish you the best of luck
42:17with your with your own work and at
42:19Saint Saint Luke's.
42:20Uh and if we can be helpful
42:23Patrizio, you please you please let us
42:24know, okay? Cuz we uh here at One
42:26Business World, we would love to
42:28be be part of this. We we've done some
42:30other some other talks in the past,
42:31probably a bunch of years ago, that were
42:33in the early days of even this and it's
42:35amazing to see to see where it's come
42:38come to. So so uh
42:40just great great great work. So thank
42:41you thank you for that.
42:43Well, thanks for the opportunity to have
42:45me here as been a pleasure and and I
42:47talking to you and having a conversation
42:49about the future of AI and many more to
42:51come for sure. All right. So hopefully
42:52we'll see you maybe in PA and maybe in
42:54New York. All right. So have a have a uh
42:57have a safe safe travels this week,
42:58okay? Thank you. Thank you again for
43:00being part of our conference. Good
43:01night. Thank you so much. Good night.
43:03Bye.