Anthony: Welcome to Health System CIO’s partner perspective interview series. I’m Anthony Guera, founder and editor-in-chief. Today we’re talking with Gui de Zwirek, CEO at Artera. Gui, thanks for joining me today.
Gui: Thanks for having me, Anthony.
Anthony: All right, great. Gui, a little bit about your organization and your role.
Gui: So, I co-founded and have the pleasure of serving as CEO at a company called Artera. We used to be called Well Health Inc., a very memorable name that no other companies share. So we had to do a rebrand.
Um but look at the foundation. We’ve had the same mission for 10 years, which is to make healthcare the very best industry in terms of customer service. When we talk about customer
service, we’re really talking about the patient. So, how they experience care outside of the four walls of a clinic or hospital. And um we’ve been investing in a specific part of that for a decade, which is communication, how we allow for seamless, asynchronous, helpful engagement outside of the care setting.
And you know, over the past, let’s call it three or four years, we’ve expanded to touch on other parts of that patient journey – things like scheduling, filling out intake forms, paying your bill, in addition to that core underlying communications stream that obviously enables all of those engagements.
Anthony: All right, very good. So, um obviously one of the things everyone is looking at and trying to figure out both on the vendor side and on the health system side, is how to leverage these amazing new AI technologies to help with all the problems that are in healthcare. So we’ve gone from a point of, uh, do we use this? Is it ready to some of this stuff is absolutely ready and being very heavily used now? The ambient AI scribes is a lot of that being used. It’s almost an exception if you’re not using it at this point to help your clinicians with their notes and things like that.
Um so now it seems to me that the folks at the health systems are trying to decide maybe perhaps beyond ambient AI although there’s an element of maximizing usage there that people are still trying to get their arms around but it’s a the larger issue of where else? everybody wants it everybody thinks it’ll work everywhere or could there could be a general sentiment that you should be using it everywhere but where else should we use it where should we try it and how how do we roll things out? How do we pilot? There’s really, in my mind, there’s kind of two lines or two areas here. There’s um AI that you’re going to wind up using as a result of your vendor starting to use AI. And there’s a whole lot of issues we can talk about there about how that comes into a health system and proper notice and notification on the vendor’s side because the health systems want to know when it’s coming in. So, it might go down different governance tracks. And then there is a health system using a AI sort of independent of getting it through an existing vendor product and kind of doing it on their own which I could use more help understanding if you could illuminate me in how you think some of that’s going on. I think we’re
going to talk about that a little bit but that’s a general idea. I’d like to hear your thoughts on it if you think that that’s kind of where health systems are right now is figuring out where am I going to use this amazing tool. I’ve only got a limited amount of people. I’ve got a limited budget. Where do I go? How do I do it? How do I measure it? Your thoughts?
Gui: Yeah. Um, I think AI for a while has been a solution in search of a problem. And I think we have a lot to be thankful for with the ambient scribe space, right? It made kind of like telehealth, right, during co it made AI part of the vernacular at every single health system overnight. The thing with AI is there’s many, many different forms of artificial intelligence and we’ve been using AI in our product for eight years. Uh but the AI that people think of today when they think about AI tends to be large language models. That’s a huge component of what the ambient scribes do. So to your question on where next or what’s next. You know I think most people either look for the top line, right? Drive revenue or the bottom line, cut costs right? we’re running businesses at the end of the day even though the patient and public health you know may be the mission that we all share.
Um the ambient scribe space I think most people would say is a topline initiative. Um one making doctors more productive uh two more effective coding which is going to drive revenue. I think most people would probably agree with that statement where we have seen where where I think the next title wave is coming is in a space I call agentic AI. Um which you could make an argument it’s both a top line and bottom line driver. I think the instant ROI is actually on the bottom line and cost-cutting. Um, a lot of people are looking at how they can eek out operating expense to drive more, you know, earnings. Um, especially in an environment where nurses, doctors, everything’s getting more expensive. Um, so when we talk about Agentic AI, I’m
really talking about the phone. You know, we have a massive workforce in healthcare. Um, just look at Veterans Affairs, right? The biggest health system in the United States. I think they have upwards of 100,000 employees, tens of thousands in the call center, right? And these folks are picking up phones every day. They’re not super efficient. It’s not a great experience for patients. It’s why we started this company, candidly, to try to reduce the reliance on this synchronous 9:00 a.m. to 5:00 p.m. channel of communication where you’re getting bounced around to get to the right information. So, we’re seeing a huge push in our conversations with our customers around that agentic AI space. But back to my point earlier, around there are multiple forms of AI.
An Agentic AI solution actually has five or six different technologies under the hood that are stitched together, right? I’ll just explain what I mean by this solution. Anthony,
you’re a call center representative. Um I might call Guillaume the patient and instead of it routing directly to you, an AI voice answers and ask me how it can help me. I might say, “I need to talk to the doctor. I don’t understand my test results.” And I might do a live transfer. So, that might replace the transfer desk. Or I might say, “Hey, I’m just confused. I thought I had an appointment this week, but I can’t remember. I need somebody to help me look that up.” And that is a task that an AI today could help me fulfill end-to-end 24/7, frankly, quicker than a human being could. But to deliver that experience, there’s a large language model. There’s an AI-generated voice. There’s a bunch of other stuff under the hood that delivers that human-like experience that can replace frankly low-level work that a human might have otherwise done.
Anthony: So, what are the implications? You’re saying that’s the next that’s the next area people are going to be thinking about or looking at.
Gui: Um, yeah, there’s over a hundred there’s over a hundred companies that have been founded to solve this problem. I feel like lightning’s hit in this space. There are a lot of real world production deployments that have been very, very successful. Usually this stuff starts in the specialties orthopedics of themology, GI it gets proven out there and then it hits the mainstream. So that’s where we’re at I think in the evolution cycle of the technology. The technology is there today to allow an artificial voice to complete tasks on behalf of the patient, everything from scheduling to managing prescriptions, doing refills, things of that nature, and resetting, you know, portal passwords. The vast majority of calls today, and I’ve seen so much call center data both from health systems and, you know, independent clinics and practices, 45% of calls are, I want to verify my appointment, I’d like to cancel my appointment, I need to reschedule my appointment. 45%. That can be 100% automated with AI 24/7 telephonic or text-based or through an app. It is 100% um uh 100% automatable. Um the rest of the calls are slowly being picked off uh after that first kind of main volume.
Anthony: So in your mind, is this something that a health system attains through a vendor or is this something you build as a health system where you do it yourself or do you get it through your vendor? Your vendor that you’re using for this kind of stuff, they’re getting better and better. They work this stuff in, you know, their offering just gets more sophisticated. So now they tell you, look, this is what our tool does now. I mean, I can’t picture many health systems building this out themselves, so to speak.
Gui: It’s a good question. There’s three different ways to do this, and I think it’s important that every system or practice understands. One, you could build it yourself by connecting directly to the metal, as I say, which is going to the lowest level infrastructure and building out the integration between a low-level framework, a few different LLMs, text-to-speech vendor, a speechto text vendor a few others and you could stitch that together and create a solution it would cost a decent amount of money um and you need devops sec ops infosc, integration ML engineers uh who are going to be all in Python by the And then you need actual engineers probably coding in JavaScript or TypeScript. So you could build it yourself. Um I think minimum to have something respectable would be a $5 million investment. Um so that’s option A.
Option B is you could go to a vendor, but there’s two different types of solutions that are flooding the market right now and it’s a very, very important distinction that I think every house healthcare executive needs to understand. One, there are the companies who are basically OEMing a middleware provider. There’s a bunch of companies that do this. Syllable, Bland.AI, Vappy, your listeners should should look all of them up. Um there are a bunch of vendors who are just creating a wrapper around that and doing the integration and charging health system executives double the cost to have a turnkey solution. Um I don’t love that approach. Um, you know, when we think about how complicated interfacing and interop is, the the healthcare execs tend to have that expertise in house. My advice would be to actually connect directly with one of those middleware providers versus using a third party. So, that’s that’s kind of a second vendor option that I don’t recommend.
The third would be to contract with somebody that’s focused exclusively on healthcare that is plugged directly into the metal. That’s the the the camp that we fall in. We’ve made that upfront five plus million investment and that continued I mean I think it’s ramping it’s far north of $10 million now for us and we’ve only invested in the things that are relevant for healthcare and that gives us several advantages. one, time to value two, the product’s actually better latency is quicker three it’s trained on real real world healthcare calls you know we don’t have to use a massive large language model that is also has to be adept with e-commerce right e-commerce is irrev right? Buying a new Nintendo Switch is waste of that sense. So those are really the
three different types of options you have when it comes to these voice or text-based agentic AI solutions. Again, it’s build it yourself directly to the metal. Two, buy a vendor who is basically a wrapper on an OEM solution or build yourself on a middleware solution. Or three, contract with a healthcare specific vendor who has gone directly to the metal to build these interfaces to your
Yeah, go ahead. Go ahead, Anthony.
So, you have you have made this step. You have done this thing. You have gone from you said you you’ve had AI for eight years, but not aic AI. So, you’ve been on this course. You you’re trying to you’ve always been trying to make your product as robust and and viable and and wonderful as it can be. You as a CEO are looking at new technologies as they come along and saying, is this applicable to what we do? Can we can we inject this stuff into our stack into what we do to supercharge what we offer? And that’s what you’ve done, right? you have y through sort of AI you’ve agentically empowered your solution. Um and that’s that’s interesting because I think that’s what’s going on in the marketplace. Um and what what I would like to know is what are your thoughts on the the the transparency that’s needed with your customers because this seems to be an interesting thing as I talk to people on the health system side. they they sort of indicate that becoming an AI-empowered solution is something they need to know about. So if they let’s say they have a vendor this wouldn’t be you because you said you’ve been using AI for a while but maybe you’re using a different kind of AI now that in their mind changes the product a little bit in terms of the review and governance they want to have over it. Um, do you do you understand what I’m saying? Like does that make sense? This guy that this point that AI changes the product we were getting. We need to look at it differently. We need to know it. It it changes a little bit about the relationship in terms of transparency. Although maybe it just needs means you need to be transparent about this thing in particular because it changes how they want to handle it. What are your thoughts around that?
Gui: Well, and your assistant is spot on. We’ve been using AI for eight years, but we are using new forms of AI every two months in this company. I mean, the pace of change in this in the tech industry is unheard of. I’ve never seen this, you know, I’m 20 years in tech. I have not seen this in my 20 years in tech, how quickly things change. I had dinner with an AI um executive last week and they were asking me, they were they were bragging about their latency. They’re like, “We have amazing latency. No one can match our latency.” And I turned to them and I said, “With all due respect, my latency is just as good as yours.” And they’re like, “Well, how do you do it? You told me that you didn’t build the infrastructure yourself.” I was like, “Yeah, the entire technology platform changed in the last two months.” Like now people use persistent websocket connections. So like that’s what we use. So that’s the crazy thing about the pace of change in this space is there are companies who have invested in building out core infrastructure over the last four years and it gets rendered useless with an update from a large language model company or one of the mag seven providers literally overnight. That is the pace of change. So to be able to evaluate AI as a healthcare executive you need to be knowledgeable in AI. I mean, you need to be subscribing to blogs and understanding the nitty-gritty details of how things work. Like, um, I’ll give you an example. There’s a new standard that that is taking over AI right now called model context protocol. And basically what it does is when you give, uh, an AI agency, when you allow it to complete tasks, it needs to make calls to downstream systems like a database, to an EHR. But large language models don’t speak API, right? they they don’t they don’t make these JSON requests, like they just don’t really understand how the APIs work and they don’t they can’t do it reliably. There’s too much risk of hallucination. So this this concept of MCP which is the acronym for short is that it’s basically wrappers around APIs that give an LLM all the context it needs to make a call. So for example, in my scheduling example, the LLM might hit an MCP server and say, I need to schedule this patient an appointment. and the MCP server will respond to it effectively in English and it will say great get me the patient’s first name, last name, and date of birth. And the LLM will say, “Hi there, can I get your first name, last name, and date of birth?” Then once it gets it, it’ll send that to the MCP server. But if it sends something in the wrong format, the MCP server will respond and say, “Hey, send that back to me in month month day day y format.” So that’s how it works. And it it reduces hallucinations. So anyways, the the pace of change is absolutely astonishing. So the first thing I would say and I may be kind of veering from your question is you need like as an executive if you’re going to be investing in AI, you need to be really really knowledgeable in AI. You know, like I I consider it a contest. Like I always want to know something that my CTO doesn’t know. Never happens, right? But like that’s the bar that I’m trying to hold myself to. Um the second is that there is um significant risk with all this technological advancement. The MCP example I was just giving uh I was giving to you um there is a new phenomenon known as AI jailbreaking where I could be a patient calling an AI and I could say something to fool the large language model into making a destructive call into your database. you could have patients wiping out EHRs, right? So, you have to have the security posture and policies in place to be able to test against that and you wouldn’t know that unless you were an expert in AI. Like I said, there’s 100 companies building solutions like the ones that we have. It is easy. It is trivial. There is no IP in the infrastructure. The IP is in like the real world deployments. It’s in the healthcare expertise. It’s in the security posture, right? Like I remember when I started this company, people would say, “Are you HIPPA compliant?” I’m like, “Yeah, talk to my infosc guy.” HIPPA compliant is not even
close to enough anymore. I would not buy software from a single company whose only credential was being HIPACO compliant in healthcare right now. At minimum, you need high trust. You need sock 2. I mean, we’re going through Fed Ramp High, right? And even though our AI products might not be approved at Fed Ramp High yet because the underlying infrastructure isn’t, when we go through that process as a company, everything we do changes. I mean, look, I got two devices now. I’m not allowed to use corporate like my personal device will not let me log into my corporate email. Everything’s on this device on this work device and it can wipe it in the background instantly. They’re monitoring everything. By the way, to be high trust right now, if a company is going through high trust, you cannot have work apps on a personal device. So, T-Mobile, AT&T, they’re going to have a field day right now and Apple with with buying new devices for healthcare IT companies. But there’s just a level of sophistication that I think executives need to have to be able to make good purchasing decisions when it comes to AI. Um, and and I think that if they don’t make those decisions right now and invest right now, they will be left behind. this space is moving so quickly um that that you need to have your toe in the water in some of these new use cases.
Anthony: Wow. I mean, that’s that’s pretty amazing stuff. Um it sounds to me like people could be getting themselves into problems they don’t even realize they’re getting into you. If you engage with the wrong vendor, if you try and do something on your own without the proper expertise, you may be thinking of cyber risk, but it sounds like this is way beyond traditional. My CISO needs to look at it cyber risk. Are there other sounds like is there other risk stuff going on here that’s beyond cyber? It’s different risk than you talked about making calls in. I’m just I’m just wondering for is it possible to do your traditional risk examination of something and miss new risk that’s inherent in some of this stuff?
Gui: Oh, you you will miss the new risk. There there, everyone is missing risk. It is impossible to do a thorough uh I should never say impossible. I think it would be very very hard to run such a thorough assessment to not expose your organization to some sort of risk. I don’t mean this I don’t say this to be a doomsdayer. I actually think the investment should happen. Right? I what I would but what I would posit is that you need to invest in some of this technology but you need to build the individual level of expertise to be able to evaluate it and not outsource that. Do not outsource that just to your CISO or your CIO. There’s security risk. There’s interoperability risk. There’s patient experience risk. Right? How do you create how do you make sure that your partner right whether it’s an existing vendor or new vendor has the information systems in place to catch those for you because they will show up a new technology will be adopted new, risk will be exposed. I’ll give you an example in in the agentic AI space there’s a common um assessment framework and it’s called the the judge framework we call it LLM as a judge when we deploy a new use case we’re going to deploy an agent that can refill prescription I’m making this up. We uh actually we will never deploy that to production until we have run hundreds if not thousands of synthetic conversations. Like we’ll make up like an 83y old grandma who, you know, can barely hear on the phone and doesn’t know the type pad and maybe only speaks Farsy, right? We’ll make up all these simulations. We’ll run a bunch of conversations and then we have a judge that scores all of those conversations for accuracy. Was the task completed? How long did it last? What was the patient’s satisfaction? And then not until it reaches a certain quality score, which is use case dependent, right? And we can normally base that off of how a human like how successful human conversations are. Not until we pass that quality score will we release an agent in production. And then once the agent is live in production, we are actually running every single transcript through that judge in real time. So you could log on as an executive and you could see exactly what’s going on without reading a 100,000 call transcripts, right? You would know, hey, the 98% satisfaction, 2% had to be transferred, there were these issues. It actually recommends updates to the prompt to make the entire system run smoother. And that was like a personal investment that that that we made that every leading company has under the hood and that you should surface to executives. But no one would know to ask that because the tech is moving so quickly unless you were really following the space. I I mentioned and and feel free to interrupt me as you can tell. I’m very very passionate about this space. I mentioned earlier this new protocol called model context protocol. Mhm. It’s an open- source standard that was proposed by Anthropic a year ago and it competes with another standard called ADA.Google and Microsoft all went on uh Google, Microsoft and Meta all went on the record and I hope I’m saying this right earlier this year saying that they were all adopting MCP.
If you’re building solutions in this space and you’re not adopting the MCP framework, you are behind already,right? So like anyways, I think I’m belaboring the point.
Anthony: No, no. I think I I think you’re you’re making excellent points. Sort of the highlevel stuff I’m hearing here is if you’re going to be involved in your health system, you know, with if you if you’re an IT executive, you’re going to be involved with this stuff. you’re going to be perhaps part of the decision-making apparatus of what vendors you’re going to work with, how you’re bringing this stuff into your organization. You have to absolutely commit yourself to becoming an expert on this stuff. And it’s almost like an hourly daily intake. you have to have um I read something a day or two ago about uh a PhD saying don’t get a PhD in AI because by the time they put the materials together and yo get your PhD it’s going to be meaningless. Uh so there’s no point. It’s moving too fast for you to get a PhD. It sounds like this is almost you know X daily updates on X blogs as you said. Uh, don’t show me an ebook that was made uh two months ago, three months ago, not relevant anymore. A white paper. I I I you know, sometimes you ever have you looking up something, you can’t find a publication date on it. There’s no date. You need a date. I want to know when was this information put together. I don’t want to read something on AI that was put together three months ago. Isn’t that crazy? We’re talking about threemonths. So, that is that an important takeaway here? We don’t, as you said, you don’t want to scare people, but you you have to you can’t dabble. You got to commit yourself deeply to both hiring experts and becoming an expert yourself. And if you’re not willing to do that, uh you’re going to make big mistakes here. Does that make sense?
Gui: Oh, you you have to be an expert if you’re going to hire experts. Otherwise, it’s the blind leading the blind, right? You need to understand you need to have some base level of understanding and sophistication, right? If you’re going to build out a team to solve the problem, you’re absolute you are spot on. And the the paradox here is like I really don’t want to scare people into not making the investment because they will fall behind. I have never felt in my life such conviction around technology that is going to change the world. We are overestimating the value of the technology in the short term, but we are so far underestimating the way this is going to change things in the long term. And you have to dabble now. You have to get in the arena. Let’s talk about how to get into the arena. Okay, let’s talk you’ve already done
Anthony: Let’s talk about how to get into the arena and how you know I I think we’ve
made conflicting points, but it’s the same point. You can’t dabble because it’s too important and too dangerous. There’s too much risk. But you have to dabble. So, how do you dabble? What you know, how do you do that?
Gui: So you, individually executive needs to be have a base level of competence. So that’s
where you can’t dabble on competence. You need to form a base level of sophistication individually. Please do not outsource that there. It is not just infosc. There are too many implications to how this technology could be used. At the same time, you should field production deployments in your organization. But I would recommend starting in the least risky spaces. So let’s take ambient scribes off the table because that’s a kind of welloiled weathered machine at this point. Let’s move to where I think the next wave is, which is agentic AI omnichannel. Basically patients talking to AIS via text portal or phone. Um many EMRs have started doing this with like the clinical, you know, notes, you know, where a doctor writes back and it’s a suggested response and the doctor presses send. But I think this is going to take over the phone uh at most organizations much sooner uh than folks think. So before it’s too late and your competitor has replaced the vast majority of their call center, right? And now they’re able to handle much higher acuity needs, make sure you understand how you’re going to deploy this thing and you get organizational buyin. So let let me give an example. The bulk of calls into most organizations today is verifying an appointment time. When we talk about low-risk use cases, you could build an agent that cannot hallucinate but is able to answer that question in a wonderful way 24 hours a day, 7 days a week. That’s a great low-value use case. It requires integration, right? But it doesn’t have to be a fire interface. It doesn’t have to be anything crazy. It can be a single kind of birectional SIU message, right? HL7 SIU message. Start with that. Get real world transcripts. Get real world data. Build a dashboard that I talked about that looks at all those transcripts and shows you what percent effective it is, what percent non-effective it is. Have those recordings. Get the executive buy in. If you’re an expert and then you can get the executive buyin on how this is applying, you know, how this is working in real life and you get that security depth under the hood, you’ll be able to turn on use cases two through 100 much much quicker. And I don’t care if that’s the only use case you have live for two years, right? As long as you’ve got something live, you understand how it works, you’ve understand how to get executive buy in for a completely new type of technology. That to me is like the sliver that’s the wedge that folks should be focused on right now.
Anthony: I want you to help me understand something. You know, you talked about what you’re doing at your organization around Agentic AI. In my mind, I could be a CIO sitting there and if I’ve done a really great job bringing on excellent vendors like Artera. Let’s say Artera is one of my vendors, right? I don’t have to be running around figuring out how to get a Gentic AI into my call center, do I? Because you’re doing it. You, Artera, my vendor, like now I got this great everything’s going great or or I’m using Agentic AI because my awesome vendor put it in. So what I’m trying to understand, you could help me understand is how much can I sit back in a sense and I don’t mean like do nothing but I mean if I’ve made excellent choices in my vendor partners, they’re going to be juicing up their apps with AI, getting agentic AI and then I get the benefit of all that. To what degree do I need to do AI outside of my vendors?
Gui: Here’s the problem today. So you’re you’re not wrong. Most organizations will have vendors that do something like what I just described. The problem is they won’t just be one. It’ll probably be 10. Your call center company, your EMR, your communications. You’re going to have a you’re going to have a bunch of people all claiming they do the same thing. All claiming they can do it quicker, they can do it faster, they can do it cheaper. So it actually creates forced evaluation that delays time to value, right? When you’re like, “We’re going to do this thing, guys. I want to invest. I want I want one aentic AI lowrisk use case live in the next two months, your team is going to go out there and be like, well, sir, you know, or ma’am, uh, we got like 10 different vendors. They all say the best. They all say they’re the cheapest and these ones claim IP. Part of um uh again crusade is probably not the best word, but one of my personal objectives right now is to just educate the market on how easy this is to build. How there is not IP like anybody who claims IP in the tech, I think is taking you for a ride or they don’t actually understand how quickly the space is moving. um that that conversation I told you last week, there was a comment voice activity detection and they were like, “Well, gee, we’ve had to build our own stuff because when you know when the patient’s talking on the phone, if a caregiver is talking in the background, the LLM would get confused and hear both things. How have you solved for that?” And I was like, “This vendor, this like massive vendor, that it’s a mag 7 deployed an upgrade two months ago or whatever, a month ago that fixed that problem.” I would never think that I Artera, right, little old Artera, okay, a few hundred people, but still I am not Google. I’m not spending a hundred million a year on voice activity detection, right? Like it would be silly to think that I will ever be able to keep up with them. So we are anchoring our technology on the best providers, the best infrastructure where we know billions of dollars are being spent, right? Um, so that that is the challenge right now. and my my I I I it’s a tough time to be to be a health system exec because there isn’t a simple most EHRs haven’t like crown made a vendor yet, right? This happened in the ambient scribe space with Epic, right? They haven’t they haven’t kingmade other vendors in the other spaces of AI because they’re still being determined. So you have these executives who are getting smart, who want to be at the forefront. They’re like, I have 10 vendors who do this. I’ve got 10 vendors calling every single month telling me they do this. I’ve got my board who are investors in this venture capital firm who say they have companies they want us to use. What should I do? Right? And and so that’s why I’m spending a lot of time when I talk to executives today. I’m not even selling them. I’m like, let me just explain how this is built. I want I want you to know how easy it is. When somebody tells you, hey, I can clone your doctor’s voice. Aren’t we cool? We can all do that, Mr. or Mrs. Executive. We can all do that. That’s not special, right? What’s special is how much real world deployment do they have? Are they using the latest standards? Are they on top of tech? Can you really trust them with your security? I know they went through your security team, but are they thinking about all the things your security team hasn’t thought through? Are they thinking about AI jailbreaking? Right? Are they single-tenant? Are they multi-tenant? Like how are they structured in this AI world? And can you trust that vendor?
Anthony: Well, that’s this I get I think I understand what you’re saying here. uh you have to be you have to make yourself enough of an expert that you can’t get taken for a ride that you you need to know more than the person trying to sell to you so you can understand when they say something that doesn’t make sense when they say something that you know is out of date by two weeks could be that crazy right that’s that change by the way you know you say what you’re telling me is no longer the case because two weeks ago right whatever um so it’s this idea of developing this expertise because there’s a lot of risk. There’s a lot of snake oil salesman out there. There’s everyone’s going to tell you what can be done. Is it that’s a lot of what you’re saying, right? Make yourself an expert.
Gui: Let me give you another one. And this is this speaks to company size. Okay. The average Agentic AI solution today charges let’s call it a dollar per phone call to to any length phone call to make these calls. We went out to market a few months ago at 10 cents per call minute, which we calculated as our cost. I go to customers and say, “Look, this is literally my cost. I’m making no margin, right? 10 cents a call minute.” I recently found out, by the way, our cost is actually 12 cents per call minute. So, we’re losing two cents on every single call minute.
Now, look, I’m actually like a decently sized company. We can stomach that. We’re banking on the cost of AI going down over time. But, let me add in another wrinkle here. The average software as a service company out there today who raises money from investors which is mo most people selling into healthcare investors expect them to get to a 70% margin. So if you do the math here for a second on agentic AI I think executives need to understand this at the way at the rates that AI charges today. Let’s take my 12 cents a minute cost even though I’m I’m subsidizing it today. For us to get to a 70% gross margin, right, which is the terminal value everybody wants to get to, these solutions in today’s terms would cost $40 per hour. That is more expensive than a human. Right? So this premise that this is going to bring down our costs, it is true today when you’re charging 10 cents a minute, the costs of AI do have to go down. They will go down, right? They’ll probably go down 50%. That’s what happens as you know ubiquity and abundance happens with technology. But I also think executives need to understand that when they think about like 10 person vendors versus you know my 10,000 person EHR there’s a balance somewhere in the middle of like the price at which this is being brought to market today is unsustainable. It is 100% .com boom level things. Can the company sustain this when the market crashes? the AI market will crash. Not because it’s a bad market, because we’re overestimating its value right now, but underestimating its value long term. So, you also need partners, frankly, I would argue, who are like self- sustaining, right? Who who are at least not spending more than they’re making and have the wherewithal to last through that crash and go out the other end. So, that’s just a crazy like no one is really talking about that. AI is subsidized right now at all parts of the stack. And if we were truly charging what investors expect today, literally it’d be $40 per human hour. And I’ anybody who pushes back on that, I’d welcome the push back. I know how every stack of the technology charges, you know, and and we have volume discounts at our scale. So, wild times right now, Anthony,
Anthony: I’m telling you, it’s it’s uh it’s certainly giving people a wakeup call to uh to study up at the very least and and understand this stuff at a deep level. Um, you know, you can hire for that, but you said you have to know enough to make sure you’re hiring the right person. Um, you know, CIOS have to have a wide breath of knowledge and expertise. So I don’t know if I mean what are your thoughts on that chief AI officer role we see at a lot of health systems sometimes you know and there’s titles are all over the place some a lot of times is combined with data chief a data and AI officer which makes a lot of sense to me from what I understand um you know the CIOS all got uh digital in their title a about a year or two ago they all wanted digital I wonder if they wish they would have waited and left digital off to the side and throw AI on Right. Or are we gonna say,
Gui: “Hey, but that degree that degree expires in two weeks, Anthony. So what good is that going to do?” Right? It’s it’s just it’s just like you have like you know the internet happened, mobile happened. There’s a base level of understanding that everyone is expected to have in their job interop mobile web, like these are just security right there’s a base level of understanding. Um, I don’t really have an opinion on the chief AI officer or head of AI thing, but I don’t think there’s any substitute for building out that muscle memory yourself. Um, the space is moving so quickly. It’s the way we run our business today. I think I think we we each need to hold ourselves to having that level of um, competence and not outsource that. Please, right, do not hire a consulting firm to be your AI expert. Um, maybe Anthony, you and I should start an AI university. You know, the university completely refreshes every week, new course load. It’d be the most unforgiving job. Uh, you know, the thankless the most thankless job on the planet.
Anthony: You you mentioned your CTO in a in a in a nice way. Um, obviously you are very passionate and deeply connected to becoming as much of an expert as you can be in this space. Um, and you mentioned your CTO as even a more do you expect him to be even more of an expert than you. So, yeah, you do. So, he wouldn’t last he wouldn’t last a day here if he didn’t know more than me that would that you know like he he should and he is he’s he’s phenomenal. what he’s done. I know maybe you were going here, Anthony. He actually has a peer group of CTO’s in other industries, not in healthare. He’s got construction tech, he’s got ad tech. So, he’s actually hearing from these folks on what’s working, what’s not working, what model providers are you doing? Uh what do you think of MCP versus A to what do you make of this pricing long term? Is Deepgram Graham or Google better for VAD voice activity detection? Like these are the things that they talk about on a daily basis. And you know you you learn through we tend to have a bubble I think in healthcare right you think about all these these associations and groups that we have I think it’s time that you know we branch those out a little bit and bring in other industries and build these peer groups because we’re going to do things faster here in healthcare like the ambient scribe space but we’re going to do other things slower than other industries and I think that diversity of perspective is really really useful um to being an educated leader
Anthony: it’s interesting this whole idea of being an educated leader because a lot of times you have the the notion of I need to think about certain things like I need to be focused on patients and the clinician experience and whatever and that’s what I want to that’s my true north and I have people who handle some of these things right and I can hire someone who’s going to understand this AI tool right so but you’re almost saying you can’t do that with this one you can’t do this so let
Gui: Let me let me explain how this is so helpful. Um, so my CTO started getting really really fired up on MCP a couple months ago. He was like, gee, it’s going to be MCP or A to A. And I was like, help me understand. I don’t really get it. Like I like I’m technical, but I’m not as technical as you are. So, you know, we’ve spent hours, you know, weekends I go on walks and we talk about these really riveting acronym, three-letter acronym topics. And um we got to a point I had a breakthrough a couple weeks ago and I was like wow. I was like ashu every ehr should be adopting MCP. He was like yeah you’re absolutely right. I’m like are they? He was like no I doubt it. Like I doubt they are. I was like we could help with that. Like this is a revol we have you have hundreds of vendors trying to integrate. They’re wrapping all these API calls. It’s creating a lot of risk for the industry. But to my to the point that I’m trying to make is that investing the time and understanding the space deeply as an expert in healthcare in your patients in your organization in rev cycle in everything that is a health system or a healthcare practice. You will be able to figure out what problems the technology can solve. You will have these epiphies and if you’ve got partners, right? Right. And I’ve got my favorite customers do this all the time. They call me, they’re like, “Gee, crazy idea. Could you do this?” And I’m like, “Tell me more. How would that be helpful? Does anybody else have this problem?” Like, “Yeah, technically, here’s how we would do it. How much would you pay for this?” And like that is where I think the beautiful ideas will blossom and we’re going to create a lot of impact on this industry. So, either everyone’s gonna think I’m a madman, Anthony, or um we will hopefully inspire uh another wave of personal development when it comes to AI with your with your listener base.
Anthony: No, I mean it’s just interesting that this is a whole different area of risk. Um I’ve been talking to people about how they’re sensing more risk on the user side uh with some of the tools that they’re rolling out u that they need they need to do. There’s a big push around user education around using these tools and how to use them and how to watch for some of their deficiencies, the hallucinations we talked about. Yeah. And to understand them and deal with them and not throw the whole thing in the garbage because of them as a user. You don’t want a user saying, “Oh my god, it returned this crazy answer. I’m done.” Yeah. So they want to get ahead of that on the front end and educate about the risks so they understand the risks and take them into account and don’t don’t get scared away. So there’s that whole
area that’s going in which is totally different kind of than the level you’re talking you’re talking about the level of engaging with particular technologies and particular vendors and bringing those in the risks you could be bringing in understanding technologies at a deep level. There’s this whole other level I’m talking about inside the health systems of educating the users on using some of these tools that they’ve already rolled out it sounds like and are now almost like we need to do a better job of training and education around these tools that we are going to unleash or have just unleashed. So it’s a whole examination. It’s like as you it’s moving very fast and there’s a slight bit of take a you know let’s look back and see what we just did and oh we need we need need to fix some stuff here we had the mobile wave remember when every system needed to create an app at least that was like a thing that was defined people had time you know there’s five years to do that
Gui: this is like a whole this is on par with co and tellahalth, you woke up one day And you needed to deploy tellahalth across the board. And you know who the happy health systems were? The one that had yes a tellaalth thing running in home health or in some department. And it might not have wide scale but they like knew how it worked. They had learnings and like my urge to your readers and listeners is like do not be on the wrong side of this one. Have that small deployment because this space is moving. It is moving fast. You saw it in the ambient scribe space. I am confident Agentic AI is the next one. Get smart. Have something high impact, lowrisk that will help you build that organizational buy in and give you the blueprint for expansion when the time is right for your organization because we have not seen the likes of this level of technology change ever. And I know I only go back like 40 years, but even if you got folks who have been in this industry for 80, I’m telling you like and and I spend so much time here. No one I haven’t heard anybody say this is beyond the internet. This is beyond mobile. This is you know it is truly and I was a pessimist. I was a pessimist Anthony. I was like healthcare like I’m the biggest believer now. I’m like waving the flag. It is unbelievable um the change this is going to bring to our industry.
Anthony: I think your example that you gave of the teleaalth and having dabbled
before COVID hit and being lucky in a sense, but also you consider nothing luck and say, you know, that was just a good thing to have done. I think that’s brilliant. I think that’s right on. Um, so experimentation, yeah, be on the right side when something’s going to happen. You could have a COVID type thing where all of a sudden, you know, do go now go and if you’ve dabbled, you’re going to be a lot better shape. So, um, I think they’re all dabbling, but you made some excellent points to help them dabble intelligently and safely.
Gui: So, Anthony, look at look at reimbursement, right? There could be that event later this year. Medicaid is getting stripped. Depending on what state you’re in, if your state’s not going to provide a backs stop, you might lose 10, 15, 20, 25% of your revenue base. What are you going to do as a health system when that happens, come October? And if not this October, what about next October? Right? Like there there’s actually the chance for these cataclysmic events and you need to be able to turn the dials in the background to continue to exist and thrive as an organization for your patients.
Anthony: That’s brilliant. It’s a great example that the the idea that something couldhappen where you’re going to really wish you had at least been a little tuned up and ready to go with this stuff. So, you made a lot of a lot of excellent points today, Gee. Anything else you want? Any other final piece of advice you want to give? Yeah, we’ll go on. We’ll go on for another hour. I know, right? No, no. How do I frame this one up? Uh, you know, health system. Let’s let’s look at the C. You could address whoever you want. The CIOS, the there’s a billion titles. Seuite IT executives. I mean, you’ve made a many many excellent points. Any fe final piece of advice? Um, and let me let me just uh say that uh they want to be active. They want to be seen as as innovative and and moving their organizations forward. Um, but as you said, they have to be careful. They want to stay tied to the business objectives, to the organizational objectives. There’s that whole governance thing where you want to make sure you’re working on the right things. I mean, but they’re expected to be what you’re talking about. They’re supposed to be supporting and educating the business objectives and the operators who are going to move those forward. They’re supposed to be that trusted consultant, especially around AI. So, they’ve got to know this stuff. Any final piece of advice?
I don’t envy their position. At the same time, I’m kind of jealous because the the opportunity for impact in a in a space that is so high impact to the recipient, the patient is astounding, right? So the the quicker I would encourage everyone to get as obsessed as I am about this. Yeah. Because I think they will see so many more opportunities than I do. And then whether they build, buy or partner, I think to me the key is like deep understanding, somebody transparent on the other side of the table that will tell you how it is, right? And work with you to get to a great end outcome. And then look for high impact, lowrisk use cases to notch that wedge. And if that’s where you end until you know the the hammer comes down and you have to go quickly, you will be so far ahead. So thanks for your time as always, Anthony.
Anthony: And it sounds like for anyone who finds this dull, get in a different line of work because there you go. All right. If you if someone’s like, “Ah, this doesn’t interest me,” you’re in the wrong line of work.
