← Practical AI in Healthcare
Practical AI in Healthcare
S1, E36 - David Hidalgo-Gato, Founder & CEO, Cleo Health: Going a Mile Deep on Emergency Medicine — Specialization, Design Partnerships, and the Acute Care OS
00:49:33
Loading summary
Transcript58 lines
- [00:01]
A
Welcome to Practical AI in Healthcare, the podcast that cuts through the noise to spotlight real world solutions delivering real world value. From patient care to clinical research, from life sciences to patient engagement, we focus on what truly matters in healthcare today. No hype, no theory, just practical insights where AI is making a true impact. Dr. Steven Lapkoff and Dr. Leanne Rosenblitt are your hosts as we explore what's real and moving the needle in this exciting new domain. Welcome aboard and let's get to it. As many of our listeners know, Leon and I work very closely with the DCI Network Division of Clinical Informatics at Beth Israel Deakins Medical center in Boston. This June, the network is hosting Patient power Digital Health 2026. The conference will bring together patients, innovators, industry leaders, healthcare providers and policymakers to shape the next generation of real world patient centered solutions. The meeting will run from June 22nd to the 24th in Boston at Harvard Medical School. We've arranged for our listeners to get a discount on registration to the meeting. If you register between now and May 15th and use promo code PracticalAI June no spaces, you'll receive 30% off your registration fee. You can learn more at dcinetwork.org patients2026. In addition, we're always looking for sponsors. If you or your company are interested in becoming a sponsor, please reach out to admincinetwork.org see you in Boston. Hello and welcome to this week's edition of Practical AI in Healthcare. My name is Dr. Stephen Lapkoff and I'm here as I am every week with my partner, Dr. Leon Rosenblit. How's it going, Leon?
- [01:42]
B
I'm great, Steve. Good to be here and delighted to have our guests with us today.
- [01:47]
A
So our guest today, I had the pleasure of actually bumping into him on an airplane recently on my way to himss and David Hidalgo Gato is the CEO of Clio Health. He's a Yale trained data scientist and on that plane ride I got to hear about what he was doing with AI in healthcare and it was really intriguing, especially given that David's not a clinician. He's never worked in healthcare specifically or been trained in healthcare. And yet he's the CEO of a company doing some pretty impressive things with AI and in this case, in emergency room medicine. David, welcome to the podcast.
- [02:29]
C
Thanks, Dr. Lapkoff. Yeah, it's great to be here. I'm looking forward to chatting today.
- [02:33]
A
Please call me Steve. It's completely.
- [02:36]
B
You can call me Steve too, but I go by Leon.
- [02:39]
C
Perfect. Yeah, we'll do Steve and Steve.
- [02:41]
A
So getting back to the plane ride, you know, I was sitting there on the plane cranking away on some stuff, and you were cranking away on some stuff. And I glanced over and maybe I shouldn't have. I noticed it looked like you were doing some work having to do with healthcare. And then we just started chatting and all of a sudden I realized, wow, this is really interesting guy. And he's working in a space that I worked in long, long time ago, working in the er. And you're doing some really amazing things. Why don't you tell us about your background, how you got to where you are today, how you kind of got your. Your superhero cloak.
- [03:17]
C
Sure, yeah. I'll give the brief CLIO story. And so as you mentioned, Steve, background in data science at Yale. That's actually where I met my two co founders from CLIO as well. They have similar backgrounds as well as in the field of clinical machine learning research. In 2023, the three of us came together and it was sort of on the heels of the ChatGPT release to the public. Right. Where generative AI started to really, like, get its wings. And our idea for CLIO was how can we take this technology and make healthcare better more broadly? Intentionally broad and abstract. But we decided to start by focusing on one group of clinicians. We chose emergency medicine and trying to solve many problems for the same people. Right. Going a mile deep on the specialty. So that was the beginning of Clio back in 23. Throughout the course of 20, 24 and 25, we developed a wide variety of solutions, first for emergency medicine and then naturally expanding into the inpatient care setting. I'd say we're about 50, 50 at this point as an organization. And yeah, that repositioned CLIO as the acute care AI partner. But that's, that's our story. We work with private physician groups, health systems, individual clinicians. I like.
- [04:36]
A
So, you know, you mentioned that you had two other partners. Jackson Pullman was one of them. He had a. A paper in Nature, I guess, back a few years back. And Nicholas, I'm not sure how to pronounce this last name. Is it Christakis? Yeah. Yes.
- [04:50]
C
Yeah, Jackson did publish it was his senior thesis very unrelated to CLIO work. It was on the microbiome, but a super interesting read.
- [04:58]
A
Oh, okay. So not a complete connection point to your work, but he was working in. In the, in this kind of space, in the healthcare space, a bit, yes. So what drew you to emergency medicine of all things? I mean, there's so many different verticals in healthcare, primary care, obstetrics, psychiatry. What drew you to emergency medicine specifically?
- [05:19]
C
Yeah, it's a great question. The first, transparently, our first introduction to emergency medicine was just coincidences. And I have a very close family friend who's an ER doc who I've known since long before I started CLIA with my co founders. And he reached out to us saying, if you're using AI to solve problems in healthcare, can you focus on emergency medicine? Because we have so many problems to solve. So that was back in 23, and that's what exposed us to the space. But we still wanted to be very shrewd as to the group that we chose to focus on from the beginning. And the reason that emergency medicine was so appealing was one, again, emphasis on the amount of problems that there are to solve and the complexity and uniqueness of their workflow. But two, we were quickly convinced about the benefits of emergency medicine being the door to the hospital and in many ways the centerpiece of the whole healthcare system that ended up working out throughout clio's journey. And that once we figured by learning about emergency medicine, we also got a sense of that first step of the inpatient workflow and the patient moving through the hospital. And so it led to a very natural transition to the inpatient care settings. We chose to go that direction. But honestly, emergency medicine is in the middle of so much. We very, very easily could have just gone the other direction to understand the outpatient care setting and how that ends up with patients being in and out of the ed. So yeah, our goal was to expand beyond one specialty. And we thought that emergency medicine was a great specialty to start in, understanding that they would help us go to others later on.
- [07:04]
B
So, David, help us understand what it is that makes emergency medicine so different from the workflows that most AI documentation tools are designed for. Right. You have 120 competitors in the space that are trying to solve the problem at the broad level. To say we're just going to do clinical documentation and ambient AI. What are they getting wrong and what are they not understanding about the specific workflows that distinguish emergency medicine from hospitals?
- [07:36]
C
So from an ambient listening perspective, specifically, I'd say there are a few key issues in emergency medicine that jump out. One is the acute unscheduled workflow. Right. Very big difference between. David is coming in to see his PCP at 9:30 today, and it's going to be a 30 minute scheduled meeting versus in the ED. It's a Poisson process. Right. So you may see 0 patients for 20 minutes and then you may see 5 patients in the course of the next 10. Right.
- [08:09]
B
By the way, kudos for citing the right distribution. Most people would have said it's an F. It is indeed a Poisson.
- [08:15]
C
It is, yeah.
- [08:16]
B
Good education. Yeah.
- [08:17]
A
Leon's going to start geeking out on that. Be careful.
- [08:20]
C
Yeah, that stuck with me from a book because it linked. It was like an ed flow book that linked with my school training. So anyway, unscheduled workflow, a key workflow. You don't know who you're going to be seeing. Multi, multi patient management. You may have 10 to 15 patients that you are working with at any given moment. Then the two really important pieces, multiple sessions per patient encounter, very unique to emergency medicine. Within the course of a couple of hours you may revisit and modify a patient's story or append to it 3, 4, 5 times. Especially for those more complex patient stories, let's say when labs, imaging tests come back and that changes your disposition idea around the patient. And the fourth thing which is connected to the multiple sessions or reevaluations is the fact that in emergency medicine they bill very differently. So professional service fees in EM are defined by a separate CMS guideline that's called medical decision making. It came out as of 2023. It was a big transition for the field and that is just grossly misunderstood by not just other ambient listening companies within the space, but I'd say just really it's very specific to emergency and hospital medicine and it's also very complicated. So we found that it's, it tends to be overlooked when thinking about the healthcare system as a whole.
- [09:46]
B
Yeah, those are really important distinction. So I think you've done a good job of describing the multi session encounters as being somewhat different than the deeply unscheduled nature of the work. But the completely different reimbursement model really drives different kinds of documentation. Can you drill into that a little bit, help help our audience understand why that matters for AI?
- [10:07]
C
Yes, yes. So as you know that reimbursement is driven by appropriate documentation. Right. So that's the framework from which you justify, hey, I think I, the clinician think that I should be reimbursed for this CPT code or set of ICD codes for this encounter. Right. Medical decision making within emergency medicine refers to one bucket of CPT codes that they call evaluation and management codes. It's the that and procedures make up the whole of cpt. And the reason why it's so important is because it's this without getting too deep into the weeds. MDM is a complicated grid of three categories. Complexity of problems addressed, complexity of data, and risk of complications. The third section, risk, called risk. Each of those three sections has very strict and sometimes obscure rules corresponding to them. It's all centered around how complex was the patient case and what did the provider do in taking care of that patient. But it's way more complicated than the old days where it used to be. How many points are you covering in your physical exam? And that's actually as of 2022. Right. I'm sure that it might have been even wildly simpler, say, 20, 30 years ago, but these days, I would argue that the vast majority of clinicians would have a hard time citing every explicit rule that is within medical decision making. And all of it needs to be documented in the right way to get credit for the work that you're doing.
- [11:46]
B
Super interesting. And I think this highlights a phrase that you've used before in our conversation when you talked about going a mile deep, which is deep workflow understanding. One of your more interesting arguments is that when software development itself is commoditized through better AI tools, what differentiates your team and your company is knowing where to look and which conversations to have. Can you unpack that for us? It's a really interesting thesis and I think really central to your positioning, but we'd love to hear your thoughts.
- [12:19]
C
Yeah. So, first, a broad statement that is not just about clio, but I'm overwhelmingly convinced that in order to add value as a technologist in healthcare, you need to be a tremendous listener and work very hard to understand the workflow of those who you are serving. Right. The way that we have operationalized that within CLIO has been. And, Steve, you mentioned I am not a clinician. My co founders are not clinicians by background. And that has led to a very natural humility exercise since day one of operating and growing clio. Um, we clearly do not know and never will know as much about the best way to use our product as the users of our product. No, the clinicians themselves. And so because of that, we have had to listen extremely carefully and develop these tight feedback loops such that we release a feature, we hear from clinicians, and they say, I like this or I don't like this. And we need to assume that they're right. And it's a challenging exercise that we constantly try to get better at. But what it ultimately leads to is good listening leading to workflow understanding. And that workflow understanding allows us to fit into what they're doing today much Better than we would if we were just kind of pushing a product forward with our own vision and not listening carefully.
- [13:47]
A
So, you know, I want to just tease that apart a little bit more because in my experience working in the healthcare IT space for quite many years, it's a very uncommon statement that you said you're so humble about how your development goes and how you are listening to the doctors. A lot of times that piece of it is really missing and that reflects in some of the work product that comes out. I mean, you could look at many different electronic medical record systems which don't always mirror the workflow in a way that is smooth and easy to use for the clinicians. It's built because it's collecting data and it doesn't always take that into account. So kudos to you for actually pulling that off. It's not something that you hear all that often. Let's talk about the first physician group that you worked with in Colorado and how you actually did your work in getting the first product built and how did that evolve? Because you also took an awful lot of time, months in fact, to pull that together, which again, not very common to have that kind of a really long time of thinking before you stitch any code together.
- [14:58]
C
Yes. So to clarify, we did take a long time, as in we had one customer intentionally for a while this that design partner physician group. We were constantly coding but. But we were just working in a tight iteration loop with that one customer. And I'm very grateful that we did so. So happy to dive into detail there. The first time CLIO was used in the emergency department was January 2024. And we sort of publicly launched CLIO within emergency medicine in September of 2024 at the ASAB conference. So it's about nine months that we were heads down with our design partner. I highly recommend for anyone looking to start innovation to get a design partner who is a representative of the field that you're trying to and the market you're trying to go after. Because what we did with this organization is we essentially zoomed out and said, connecting back to the humility piece. We do not know how you work today, but we promise that we are going to listen and understand to the best of our ability how you work as efficiently as possible. So we asked them, what is the first problem you would like us to solve? And they responded saying, fix our documentation problem problem. At the time there were ambient scribes in the market, but they weren't tuned to the emergency medicine workflow. So we said, okay, let's build you a better scribe for the ED. And that led to a sequence of probably 40, 50 plus loops where we would come out with Clio V1. And Clio V1 was an HPI and capturing verbalized exam findings and an evaluation section. There was no medical decision making. Their immediate response was in great, good job so far. That exam section is of no use to me because I'm not going to verbalize every one of my exam findings. So Clio V2 was auto populating a default normal exam on clinician request where we would map in their pertinent findings to their normal template. They said great, thank you, the exam's better, but also what happens when I get the test results back an hour later turning into Clio V3, right where we had timestamped reevaluations and then it just, it carried on right for dozens and dozens of back and forths. But the without getting too into detail there, I'd say that that tight iteration loop and deep focus from a committed design partner was crucial to us building like a best in market product. Because they like we would not have been able to discover those problems ourselves. Right. We had to ask the end customer and we, we're very grateful that CarePoint, our first organization partner, was so willing to lean in and help us out in that way. And I'd say it's generally very hard to find someone who's willing to make a bet on an organization at that stage.
- [17:53]
A
Yeah, absolutely. I think that you're absolutely correct. In previous points in my career I've worked with EMR companies in trying to build product and finding a medical group that's willing to spend that kind of energy in refining things is actually very, very rare. But let me ask you a question about that in particular because it raises a different kind of a question which is this medicine is practiced for the most part locally and local practice patterns tend to be just that local. As you're working with a group that happens to be in the western part of the United States that works in a certain way, how extendable did you find what they taught you to do or how did you do when you started going to other places and how that actually fared in different locales, did it work or was it, you know, was it sufficiently specific or was it generalizable?
- [18:51]
C
That's very thought provoking. I hadn't thought about the local differences in this way before. So simple version of this response is the core building blocks that we developed were very generalizable and we've found that to continue to be the case as we develop core building blocks in inpatient medicine. Building blocks such as you will have multiple sessions per encounter from a technical standpoint. Right. The medical decision making as defined nationally reads. You have to check these respective boxes. Right. Even as we delved into other features that Clio had within its ambient listening product, like real time feedback on even patient experience quality that we can speak to individual behavior that leads to more positive patient experience outcomes, we found is generalizable. What's not generalizable and where organizations can be sort of snowflakes is when it comes to the size and culture of the group that you are working with. So our first customer, I would put them right in the middle in terms of the provider groups for emergency medicine within the United States in size. So I think they have around 45 +ers and several hundred clinicians. So definitely not one of the largest contract management groups. You say take a team health, for example, with thousands and thousands of clinicians. But there are also a lot of organizations that are 50 providers or fewer. There's a long tail of regional groups. So they were right smack dab in the middle. And that really helped us because we were able to flex both to the smaller groups as well as to the larger organizations. But those culture buckets map very differently how they view the rollout of any type of software. Not only implementation once you sign, but also the vetting process pre sign we found was very distinct. And because of that, Clio internally we have different kind of strategies for pre and post implementation based on the size and culture of the organization that we're working with.
- [21:02]
A
Thank you for that. So let's pivot right into your first customer and how it deployed. You'd mentioned that you're able to save nearly an hour of charting per shift per clinician. Unpack that. What did you do differently that actually retrieved that extra time? Because obviously time is money and if you're being more accurate in less time, that's like fantastic.
- [21:25]
C
Yeah, this is a very straightforward just improvement on the incumbent process, which is before Clio and the ambient listening product. You go, you have your five minute conversation with your patient or longer, and then you go back to the desktop and you dictate your note or which necessarily takes 10 plus minutes. Right. Depending on depending on the nature of the encounter, it might be faster, might be slower. Clio's drafting that note for you and assuming that you don't have to heavily edit that note, it is just strictly time savings on top of time savings relative to what you were doing beforehand. So yeah, we found it about an average of 54 minutes per shift. That was a study that we ran early in the implementation of clio. Some save way more, right? In upwards of two plus hours. Some save way less in that they may just. It's based on what your workflow was before Clio as well as how you are utilizing the Clio application. Right. Some people go all in and they filter everything through the AI. Others prefer to dictate their medical decision making thought process because they think that AI just can't get it quite right. And so we've seen a lot of variability in usage which does impact that time saving statistic that you mentioned, Steve.
- [22:42]
A
Well, that time saving statistic translates into something else which I would like to point out. You know, ERs, typically, especially during busy times, can get very, very congested. And if you're able to save a clinician, you know, one hour, 90 minutes, whatever, that translates into seeing perhaps another two or three patients per shift and that can decongest. And, you know, if you put that across, let's say an ER with, you know, three or four docs, that can be very meaningful to patients who are sitting in the waiting room and that, that's fantastic. Let's pivot around to your second product, which was real time clinical feedback and that seems to be something that you were doing. The clinical decision support is something that we've been talking about on this podcast for quite a while and you're doing it kind of in real time. Why don't you give us a sense as to how that works?
- [23:28]
C
Works, yep. So this was an early realization of the CLIO team in that we said if we are in the hands of the clinicians in real time at the point of care, then we have the ability to communicate with them intelligently, where if we say the right thing, then hopefully it leads to better experiences for the clinician and or the patient. We have been sort of respectful of what we believe is a blurred line delving into clinical decision support. And so we don't really see ourselves as much as a clinical decision support tool like moving care versus just facilitating ideas and rule sets that are coming down from, say, the organization. The best two examples that I can give are in the patient experience realm and in the quality realm, specifically in documenting clinical decision scores. So patient experience, I think, is the most interesting work that CLIO does to this day. The way that it works is that we will take in the conversation that the clinician has with their patient and assuming that this is something that the organization and clinician want to do. We have a passive, gamified approach to giving them real time feedback on hitting key patient experience criteria. Examples could be as simple as did you introduce yourself to the patient before you started speaking with them? Did you tell them how long you were going to be they could expect to be in the ED for. Did you manage up the care team? Right. Did you refer to the other members of your staff positively, such as, this is John, he's your nurse, I've been working with him for 10 years. You're in great hands with him. So we launched that a long time ago and we gamified it such that it's actually like a color coded experience where the clinician will receive green if they check all of their patient experience boxes. And we saw a tremendous improvement in patient experience scores both with the third party vendors, as you know of like the press Ganes, the NRCs of the world as well as just clio before, after, right. Where we flip on the feedback and all of a sudden now they're managing up the care team way more frequently. Patients seem to love it, staff seems to love it, everybody wins. Right. So that was again the, that was the first real use case of our real time feedback. We now do that in many different areas. But I'd say that patient experience feedback remains probably the most interesting and most like, broadly valuable work that we do because it's just such a strained part of the system right now.
- [26:04]
A
So, you know, another thing that I remember from my days in residency and, you know, I would work both in the ER and also on the floors, was that there was a role by a resident of the night and it would be the head resident. And, and the residents got a nickname for this role. It was called the schnook. And the chinook was the one who would allocate cases from the emergency room to the floors, whether they be surgical or medical, and the residents would be called up on the floor. Come down to the er, you've got a case now you're accelerating a lot of things in the er. And patient flow would be one that would be super important because patients don't want to sit in the ER overnight, they want to go to the floor where they can go to sleep and, and get comfortable. Does your tool actually help with that function? The schnooks function, helping to get patients moving faster and more effectively. Because sometimes it could take three, four hours just waiting for the resident to come down and seeing where to send the patients.
- [27:00]
C
Yeah, I mean this broader concept of ED and hospital Throughput, as you know, is just such a massive issue in the country and the world. And I'd say that we do several things where we are trying to chip away at that problem of just patient flow optimization through the hospital. Couple different ways we help with that I can highlight. This will also delve more into the inpatient work that we do, which I think is interesting and kind of important for this conversation. So Clio has a few different inpatient products. We have an ambient listening tool. We have a charge capture tool, which is the platform through which clinicians enter their billing codes. We have a clinical documentation integrity tool, which you can think of as real time feedback. It's just driving documentation specificity. Right. Sorry for slight tangent here, but I think setting the field before talking about throughput. And the fourth tool that we have that's most relevant to what you're mentioning, Steve, is the Automated Patient assignment tool that we have. So that's our kind of inpatient product suite. We just launched it recently at shm. It's called Acute Care os. And the idea for the automated patient assignment is exactly what you are mentioning, not at the point of interaction between the ED and the hospital medicine team, but at the beginning of the day, right? So in the hospital, especially at larger facilities, say if you're looking at like a 12 to 15 floor hospital census of greater than 230, 240, it is not trivial to figure out which clinicians should be seeing which patients in a given day. And Clio developed an AI based algorithm that takes the facility's priorities. So the medical director, physician provider leaders will go in and say, we want to prioritize that Dr. Labkoff sees patients that Dr. Labkoff has seen before. So if you're coming on to start your shift this Tuesday, you haven't been there in two weeks. We want to hand you patients that are familiar to you. Right? And in addition to that, we also want to say that Dr. Rosenbilt, he is going to see patients on floor seven because he has nine patients on floor seven. And we don't want him jumping between floor seven, floor 10, floor two. Right. Because that's time wasted, right? Just in, just going on the elevator from patient to patient, so little things like that, and that's not so little. But chipping away to that nature, it's all in service of, let's treat these patients more efficiently, right? Let's stop spending time walking from room to room or waiting in the ed, getting to the hospital, right? And the end goal is getting patients healthy and out the door at an appropriate cadence. So I'd say automated patient assignment's a good example. We could talk about the concept of throughput in the ED and the inpatient setting for a while, but that's probably our most tangible response to it.
- [29:59]
B
Yeah, let me sort of flag two things for the audience and then drill deeper into the patient assignment. So, you know, I, I see Steve already highlighted this piece, but the good listening skills and epistemic humility have always been absolutely key part of product management and requirements analysis. And when I used, you know, train junior requirements analysts, that was, those are the two things I would tell them to, you know, try to train them to do is just look like you have. It's like being a psychiatrist for the client, right. You have to really, really sit there and listen and understand and realize that you don't know as much as you think you do. So it's very interesting that that capability is becoming more valued as it's part of the chain becomes, becomes valuable because it used to be that the downstream development was the bottleneck and as that bottleneck goes of it, which is the listening and is becoming more and more important. But you've done such good listening that you were able to build a patient assignment algorithm that turned a four hour manual process into something that could take 15 to 20 minutes. What makes that a good fit for generative AI specifically? So it's an amazing piece of automation, right? But you could imagine doing it with various kinds of rules. Talk about the fit between that problem space and the AI.
- [31:24]
C
To clarify, when you say that problem space, you are referring to the space of automated scheduling assignment.
- [31:30]
B
Sorry, automated. Yeah, automated assignment to. Of staff to problems.
- [31:34]
C
Yes, the, the reason that generative AI is very well suited to solve this problem is because of the lack of structure in the instructions and customizations per facility. So I, each hospital is its own snowflake, right? And the rules that I mentioned are structured. They can be structured. So think geographic based rounding. The con. Everyone shares the concept of, oh, I want Dr. Labkoff to see patients on floor seven, that side that generally like that is a bucket. Geographic based rounding that people adhere to. Another bucket is patient continuity. I want Leon to see the same patients that he saw yesterday, right? So these are understood rules. There are lots of funky rules that come up and they cannot all be codified. Right. So where generative AI comes in and it can be really helpful is rather than having to rigidly fit to some tabular structure of rule X is this rule Y is that we actually let Clinicians just speak freely to the CLIO model. And the Clio model takes that free text output reasons and then maps it to a final output structure, which is patients 1 through 10 go to David. Patients 11 through 20 go to Leon. Patients 21 through 30 go to Steve. Right. So the flexibility of input kind of necessitates generative AI for a task like this.
- [33:00]
B
Super interesting. So what you're seeing in this task is that there is a piece of cognitive labor that would be very intense for human, which is taking something vague and converting to a structured input format, like if you were filling out a form, and that by automating that piece, you're actually removing the barrier entry and just allowing the system to give reasonably good answers. You know, that human, that would take humans a long time as they do. You also mentioned that in various situations, the cost of being wrong matters with patient assignment, especially with having a human in a loop, it's somewhat lower stakes problem than sort of a clinical decision, than making a diagnosis or making a recommendation. Is that part of how you think about which problems to tackle with generative AI?
- [33:49]
C
Absolutely, yes. And we speak about this at length within the context of CLIO as well as just examining AI and acute care more broadly. The best use cases for AI right now, we think are where AI can be wrong and it's okay. Okay, right. Which the easiest way to approach that is having a human in the loop in some capacity, even for the example of ambient listening, if you're drafting the clinical node. And this is super important in education, AI makes mistakes. It is way better today than it was a year ago. Right. But the concept of AI making mistakes because it is probabilistic is it's still, if we keep going in the same direction, it will be making mistakes a year from now, five years from now. Right. Just a smaller percentage of them. So everything we do at this point, our outputs are seen by a human, reviewed by a human, and then signed off. And I think that what that also does is because you have that human reveal safety net, it allows you to innovate faster and create new tools faster, because you know that you're operating within kind of an umbrella that that allows for like the clinicians to protect the edge cases.
- [35:08]
B
So let's talk about scale for a second. So you guys are at this point, I mean, very rapid. After a rapid rollout, you have millions of patients a year now, thousands of active clinicians. You got 40% of the ED groups in the US and you just launched the Acute Care OS at HIMSS. How did you get there that fast? That's very, very rapid scaling. And you could just say, because we're amazing, but we'd like to hear you.
- [35:34]
C
No, it's just. So it was a very key decision for CLIO early on was to focus on these private provider groups, right. Where they do not own the facility, they have a contractual relationship with the health system. Right. But they prioritize acute care, emergency hospital inpatient medicine from the top down. Right. Because in that kind of. We fell into that in a very nice way in that, as I mentioned, we chose emergency medicine because of how we thought it might generalize and the unique issues for that specialty. We then chose hospital medicine as the immediate next focus from emergency medicine because of the natural continuity and workflow, but also the fact that our main customers also had hospitalists. So we were able to execute the same sell or product to the same people just for a different group of clinicians. Right. And so I'd say that it's a small world, right? That world of private groups. And by cementing ourselves as the like, our intended go to acute care AI partner that spread within a very tight community that, that we were able to kind of like take hold of. And another thing that I say is very important is from the get go, as we introduced ourselves to these, these organizations, we didn't just say, here's a note writing widget. And our note writing widget is the best in the market, right? We said, we want to be your AI partner for years to come. Right. The first problem that we are looking to solve is through documentation and emergency medicine. But hopefully we want to solve 10 plus problems together over the course of the next coming years. And in the process of doing that, I also think that we've been successful in kind of working with them to understand how healthcare is changing and the impact that AI is having on the industry and on acute care specifically. Such that we've learned from them, they've learned from us, like how we can best blend the technology with the workflow together. So long winded answer. But I'd say that first, the explicit emergency medicine focus, right? Leading to emergency hospital medicine, acute care focus, zoning in on those private groups to begin with, understanding that that's a very tight sphere, right. And it was directly aligned with our organizational purpose. And the last step that I'm missing so far is those organizations have very large footprints, right? Because they have contractual relationships with most of the health Systems in the U.S. right. And so that also was a big growth channel for CLIO because We would be able to have effectively free pilots in that Clio was being used by one of our partners in the facility belonging to a health system. And so that was the natural warm exposure for the Clio product and company to the health system which who could use the Clio product offerings in potentially many different ways.
- [38:41]
B
So this is, I think, an aspect of scaling and growth that's not obvious to folks who are doing startups in healthcare space. You guys were, you know, you will run up in a space where a lot of your growth came from physician group partnerships like Apollo, Maryland, Core Clinical Partners. And you weren't in a position where you had to sell directly to hospitals hospital by hospital. That is an enormous advantage, like really, really good positioning. And I just want to highlight that that is a really clever go to market strategy for this kind of product. Right. Like you identify a group where there's a small number of customers, you deliver high quality, you develop design partnerships with a key design partner and you just hit it. So great execution. I think we want to start talking about future direction, so I'm going to turn it over to Steve.
- [39:31]
A
Yeah, thanks, Leon. And I got to tell you, David, this has been an illuminating conversation. Even things that I didn't know about, you brought out here today. So thank you for that. You know, you started in the ER and in terms of future directions, you guys aren't going to stay in the er. Patients don't stay in the er, they go to the floors. What's your future direction and path? Where's that going to take you for the rest of the hospital? Are you going to try to keep moving in? Because that's going to lead you into other things like EPIC integration and Cerner integration and a tighter connection to the hospital information system, which might be problematic to Leon's last point. So where are you going?
- [40:08]
C
Yeah, we think about our direction in terms of product offering and just clinician. So creating new products for the same people and then like delivering the same products to different people. So I'd say again, we're the acute care AI company. That remains our focus. That will be for an indefinite future. There are so many problems to solve in the hospital though, so.
- [40:33]
A
Yeah.
- [40:33]
B
Really?
- [40:34]
C
Yeah, I don't think we will ever get bored of in. In that space. That being said, so currently Clio is primarily emergency and hospital medicine. Like I said, we're about 50, 50 in terms of our scope. I think that soon we may actually exceed emergency medicine with hospital medicine and other inpatient specialties, but within that inpatient field. So Hospital medicine. And that admitting workflow is a large but not eclipsing subset of that emitter flow. So there are also a lot of, as you know, Steve, that inpatient cardiology. Right. Ob, hospitalists, critical care. There are the other inpatient specialties. Inpatient neurology is another really good example. We have a rollout coming up where we're just trying to understand their workflows well, and we think it's a very logical next step where that hasn't been our primary focus. But there are many learnings that we can take from the hospitalist workflow and kind of use as a kickstart for those other subspecialties. So that's in terms of like a user group. And I'd say another thing that's a really big focus of Clio's is just delving deeper into payment integrity. So we attack payment integrity from a lot of different angles right now we write the note through the ambient listening product, as we discussed, and that is connected to payment integrity. Because you want to accurately document what transpired in the ED or the hospital. We have our clinical documentation integrity product that will sort of intervene in real time to say, hey, you said altered mental status, but like, that's not a very specific code, right? Can you please like click through here to explain to us what you did mean? And we found that to be very helpful in a wide variety of circumstances. And then there's also this concept of generating the code ourselves, whether it be the facility code or the professional services code, like the medical decision making that we discussed, where CLIO has a very strong understanding of that. And we already do. And we already do generate codes in some scenarios. So I do see us moving from a product standpoint just deeper into that payment integrity piece, which I guess would be sort of traditional revenue cycle management. Right. But we're, as we do this, we are trying to look at the RCM space and think like, how do we interact with it and how is it going to change as AI kind of comes on the scene within healthcare? Because we think that that space is particularly ripe for disruption. I want to make sure that we're going about it the right way.
- [43:10]
A
So in that same vein, as models get more sophisticated, development times are going to drop. There's this concept of general purpose vers, specialty specific. You guys have been betting specifically on the, on the very specific nature of the, the specialty you've chosen. As the models improve, will that change or do you think you're going to stick in that. That vein?
- [43:37]
C
I suspect that we will do work primarily within the hospital for a while. That covers many different specialties, but I see a pretty stark divide between the acute versus the ambulatory world in terms of core workflows. So even as the models get better, I see it more as solving more problems for that acute care workflow versus expanding to the broader healthcare system. I think that there are things that we don't know about the outpatient world and vice versa. That being said, it is a strong belief of ours and mine that understanding the hospital is the door to understanding the healthcare system, because it's just this deeply important centerpiece that links to that outpatient world.
- [44:27]
A
So we're running a little close on time. I'm going to ask you one final question on my side and then I'll hand it back to Leon. But the question is this in one minute, what's keeping you guys up at night? What's giving you agita at what you're trying to look at with your organization or with your solution?
- [44:43]
C
Yeah, I'll be quick here. The world is changing very quickly, exponentially. And it's impossible to predict exponential curves. Right. As you know, this is about the companies beneath clio. The OpenAI is anthropics. Google's right pushing generative AI. The models get smarter exponentially. So what keeps me up at night is just understanding that the pace of change is greater than we as humans can really comprehend. Just making sure that we are steering the ship appropriately, such that it is inevitable that the world looks almost unrecognizable in five years relative today. What is unclear is whether that new world is going to be a better world than we live in today or a worse world. And I believe that we as humans and innovators and the clinicians, non clinicians alike, have the ability to steer the direction of that change. But a lot's in our control, a lot is not. And I'd say that consumes a tremendous amount of our team's thought right now, thinking, how can we best engineer this positive change?
- [45:54]
B
Yeah, I'm with you on this. And it's one of the reasons a lot of our intellectual energy has been going towards governance. The idea is that as change is fast and predictions are hard, especially about the future, to channel Yogi Berra and what. What we need to do is be in a position to wisely react in a relatively quick and thoughtful way to things as they unfold. But the one thing we can predict about exponential curves is that they're going to grow exponentially, far faster than you can Imagine. But let me ask you a more positive question after the doom and gloom section of this interview. If you had 60 seconds with a healthcare system CIO in an elevator, what's the one thing you'd want them to know about how AI should work in an acute care setting?
- [46:44]
C
Yeah, fundamentally AI should improve patient outcomes. Right. And that, that is our goal. We want, we see better healthcare as better patient outcomes. Right. Depending on the system's nature, that could look different. The first thing that you touched on, Steve, about those 54 minutes of time saved, I think that is a very tangible way that AI can improve patient outcomes because where the capacity in the ED exceeds the ability for the clinicians to provide care, we have a huge problem. Right. That leads to bottlenecks and hours first visit. So I'd say AI first should save clinicians a lot of time and it should streamline their operations, handle a lot of work that they do not want to be doing and that should in turn give them more time to spend with their patients, which will ultimately improve patient outcomes. And so I'd say very highly, very high level, Leon. That is what we are trying to do with the work. And we think there are many different product offerings, both that Clio has as well as that other organizations are delivering that are leading to that positive result.
- [47:55]
B
Awesome answer. Where can our listeners learn more about Clio and how can they reach you?
- [48:00]
C
Yeah, we have a website so cliohealth IO I can leave that somewhere if that'd be helpful. Yeah, and Clio is also very active on social media, so I say our LinkedIn page and we have a blog post on our website that we push to.
- [48:16]
B
Yeah, we'll make sure to include the link in the show notes. So we heard a clear thesis today. When everybody else is trying to boil the ocean, going a mile deep in one domain and understanding the workflow deep granular level produces tools that actually work even in a very messy environment. I just think that's a really interesting thing to eliminate. So thank you David, for sharing this really important perspective with us and your. Your exciting experience and congratulations on the success. You know, I want to thank Steve for being, as usual, for being a tremendous co host and thank our audience for joining us. If something in today's conversation resonated, send us a comment, send us flowers or just share with a colleague. And please join us next time on Practical AI in Healthcare.
- [49:11]
A
Thank you for joining us this week on Practical AI in Healthcare. If you're ready to go beyond buzzwords and hype and explore how AI is truly transforming healthcare. Stay tuned for more conversations that get us to what works. Until next time, stay practical.