Guillaume de Zwirek is CEO of Artera.
Tell me about yourself and the company.
My name is Guillaume de Zwirek, which is a funny name. I was born in Canada. My mom was a physician, my dad was a software engineer, and somehow I turned into a blend of both of them. That’s surprising, right? I turned into a health tech entrepreneur.
I founded Artera 10 years ago, so we just celebrated a decade. I founded it out of personal experience of having to navigate a complex healthcare condition, having to coordinate all of my care by myself. I was frustrated with the status quo, the present state of doing everything on the phone. I didn’t feel like I was a customer, but rather just a cog that came in for a visit, revenue was booked, and I was sent on my merry way. I felt like we could deliver tech that could bring some of the relationships back to healthcare.
What relationship do people want with a hospital or health system? How do you measure the clinical and business value of those relationships?
I never went to business school, but the train of thought that most people are familiar with is that you can do two out of three things, the iron triangle. You can win on service, you can win on quality, and you can win on price.
We think that service is one of the most important things. Relationships are really, really important. They connect us to people. They are the foundation of trust. Especially in an industry like healthcare, when you’re vulnerable and conditions may be life-threatening, it’s important to you know who is caring for you. That flows all the way through from the physician at the top of the pyramid through to the folks who answer the phone and help you get coordinated to the right resources.
From a foundational level, humans want to connect. They want to build trust. They want to build relationships.
Dentistry is a great example. My brother is a dentist, and he shared an interesting anecdote with me. He bought a dental office, which is what most dentists do when they go into business. I asked him why it’s so easy to get funding for a dental practice when the dentist will change and you would expect most patients to leave. He said that I would be surprised that while he spends five to 10 minutes with the patient, the hygienist spends 30 minutes, and people don’t ever want to leave their hygienist.
You can extrapolate that thought to all of healthcare. If you think back on your best experiences, you knew the person, you trusted them, and they were caring for you. That’s what we’ve been doing for 10 years. We’re only part of it. We are the technology that the people have to be bought in on the other side. But it creates unbelievable connection and loyalty.
As a personal example, I once complained to a solo PCP because her front office staff were clearly incompetent and unfriendly. She urged me to call the practice after hours to avoid them because they would be gone and she would pick up herself. Can technology solve that or does the underlying problem need to be fixed first?
It’s both. The tech can help with monitoring your staff. We released a new product recently, a homepage, which is pretty simple concept. But it highlights how staff is doing in our tool. Are folks productive? Are patients happy with their experience? Are they waiting a long time for responses when cases get routed to a live agent?
This has been helpful for our customers because they can see where things are going well, and where things are going poorly. They can do coaching on the operational side. So we’ve definitely seen ourselves extend more to the operations.
A lot is changing in tech, like agentic AI. Do we need as many humans as we used to, or, or can these digital employees do the job better? I think the jury is still out on that, but there’s a fundamental technological shift happening right now in the world, but also specifically in healthcare.
Describe agentic AI and how it will be less frustrating than the phone trees of old, where the patient’s time and patience were valued less than preventing them from talking to a human.
The key word in agentic AI is agent. By agent, we mean agency, the ability for an AI to complete a task on behalf of a patient. That is the key condition, the product.
When you think about a generic AI solution, there’s actually at least six underlying technologies that make that possible. Most people probably think about the phone and being able to talk to a digital human just as you described. The alpha version of this technology dates back over two decades, which is what you were describing, a phone tree. Press 1 for Spanish, press 2 for English. That technology is known as dual tone multi-frequency. That technology is dead. That was Version 1.
Version 2 was natural language understanding, which was, please say “one” or “yes” if you would like to continue in Spanish. The patient could respond and say si por favor, ad we would understand that to mean yes and effectively means 1 and then we would continue them down the tree.
With agentic AI and with LLMs under the scene, we can ask open-ended questions, the patients we can respond, and the AI has agency to interpret what is being said and route them down the right path. That right path might be a skill, like resetting a portal password or canceling an appointment, or it could be a skill like routing that patient to a live human being because the agent is not capable of fulfilling that task. The technology has completely transformed.
I have yet to call a health system truly using agentic AI, that final version that I described, but the technology is there. I think for good reason, healthcare should be cautious. You do not want LLMs hallucinating and giving patients bad guidance.
But I think we are on the cusp of a good chunk of the telephonic volume going to these agents, because they can perform tasks more accurately and more quickly than a human can. That will free up our existing staff to focus on the high-acuity cases and building those relationships that I just described, those real, human-to-human relationships that engender trust and loyalty with your provider.
I assume that much of the volume of abandoned calls involves scheduling, which can be complex due to the patient’s primary preference of date, time, location, provider, or soonest visit. How can AI improve that?
Scheduling is by and large the highest use case for call volume. We process nearly 3 billion interactions a year and scheduling is more than half of the inquiries that come into our system.
The interesting thing with scheduling is that we have standards that all the EHRs comply with, and many have FHIR scheduling endpoints. The problem is that those are unusable in practice without the rules you just described that.
There are not only are patient preferences, there are provider preferences. Let’s talk about orthopedics. You can’t just schedule an appointment with any orthopedic surgeon willy-nilly. They have specialized focus. Some may work on pediatrics or adolescents. Other may work on hip replacements, but only for a specific gender. Those preferences get really, really complicated.
We have a ways to go on the scheduling side in terms of standards. It’s nice that we have the FHIR standards, but without those preferences on top of them, it is hard to deliver fully end-to-end autonomous self-scheduling with an agent. Kyruus, DexCare, and Radix are solutions that provide that filtering logic. The EMRs have started doing this, too. I’m hopeful that those folks will start exposing their APIs, because we don’t want to have to go out and build that logic again on another system.
How important is it that AI products integrate with existing systems?
Before we get to integration, let’s talk about the underlying tech. It’s a complete commodity. We believe that the underlying infrastructure has been commoditized by the big players like Meta, OpenAI, Google, Anthropic, and a bunch of other vendors under the hood.
With AI technology, it is not hard to build an alpha prototype. That’s why you see three-person companies raising a ton of money. We don’t think there is a durable advantage in the technology alone. We believe that the market will shake out in a way where three things will be important to hospital and physician group buyers.
One is that they will look to their existing vendors first, so it’s distribution. Second is content. Do you understand our workflows? Do you have an easy button for turning our documented business practices into autonomous agents? Third, to your point, is integration.
Integration is going to favor folks who have distribution and market share, because they will have connected into a lot of different systems. They will have connected into every EHR. They will have gone deep on their APIs. They may have gone to third parties. But integration in itself is not a durable moat, because anybody can do it. We have a lot of open standards. Integration is critical because the agent needs to be able to perform actions, so full stop. But that is something anybody can do.
I actually think that what is going to shake this market out more is distribution, because it is so easy to build. The folks that build it into their products quickly in a very cost-effective manner and make it super simple are the ones that are going see the most traction.
People have a lot of options when choosing a way to communicate, and many of them seem to least prefer talking to a human on the phone. How do you address that as a business in terms of preference for texting versus calling?
It comes down to patient preferences overall, and it’s not always patient preference. It’s also socioeconomic. Do you have access to a smartphone? Is there somebody who speaks your language on the other side of that text or the other side of the phone? We need to meet patients where they are.
I think the right strategy is omni-channel. It’s multimodal and allows patients to gracefully switch from one modality to another. We talked about dual-tone multi-frequency technology, or DTMF, the phone tree where you’re pressing 1 and you’re just trying to get to an agent or you’re a yelling “agent” at the phone. Poor implementation of automation or AI is worse than just having a human in the first place.
The right balance, in my humble opinion, will always include humans and AI. I do think AI has the ability to complete tasks faster. I do still think I think there will always be a yearning to talk to a human being when it’s appropriate. So I think it will be a combination, and that’s what we’re seeing from our customers. It’s going to be text, it’s going to be phone, it will be video. In some cases, there’s a combination of all three. Strategically, folks should think about how to bring all three of those things under either a single vendor or a consolidated tech stack to be able to manage those graceful transitions, including language.
There are other dimensions beyond channel that are going to be important to serving patients effectively. I’ll give you an example. For most companies, AI over the past couple of years has been a solution in search of a problem. A lot of people have spent a lot of money and launched a lot of experiments.
One experiment that took off for us more than we expected was translation. It was as a copilot, just a button in the UI. We looked at the patient’s preferred language and we auto-translated communications coming in from the patient into English, so that any staff member could read them. Then when the staff member responded, I could say, “Tim, nice to hear from you. Yes, I have scheduled you for lab work next week.” When we sent that message, we would auto-translate it into the patient’s preferred language.
You wouldn’t believe how popular that has been. A simple tool, very easy to build. So again, the advantage isn’t the technology, the advantage is we have 50,000 call center and back office agents in our application every single day, and they were all able to use it immediately. There’s nothing new to buy, there’s no net new integration, they just automatically were upgraded into capable individuals who can speak 100 languages.
How hard is it for a health tech business to stay on top of the daily changes to LLMs and also make sure they don’t negatively impact your product?
You have to be experimenting with every cloud service provider at all times. We have simulations in Azure, Google Cloud Platform, AWS, and OpenAI. We are running experiments at all times in all four systems.
I’ll give you a tactical example. The biggest issue with an AI voice agent today, a robot you talk to on the phone, is latency. By latency, I mean, when I say something, how long does it take for the AI to respond? A natural human conversation is not going to be more than, I don’t know, 1.52 seconds. Most of those agent take much longer, which is way too long. You can tell, and it’s a frustrating experience.
The technology behind that that slows things down is converting speech to text, then making an API call to the LLM, then the LLM returning text that you turn into speech that you read back to the patient. All of that adds a lot of delays. Literally two weeks ago, a new technology was introduced by OpenAI called speech-to-speech, where you don’t actually need to convert anything to text. The LLM is doing all of that natively, which significantly reduces latency.
The day that was announced, we had an experiment running and we were benchmarking the latency against our other system. Every single week, new tech is coming out. We need to understand, does it meet our HIPAA requirements? A lot of the experiments we run, we would never take to production because they’re not ready.
Second, how is it fundamentally going to change the way the software is built? It truly is, if not weekly changes to the underlying infrastructure, daily changes. That is dangerous for people in the infrastructure layer, because investment that you might have spent tens or hundreds of millions of dollars building can be wiped out overnight when one of these large providers releases an update.
What concerns and opportunities for the company do you expect to see over the next couple few years?
We’re in the business of customer service. I try to ignore the noise of what’s going on in the markets, with competition, and other other things in general. How can we make this experience incredible? How can we make the healthcare experience the best experience that the customer, the patient, will ever go through? We are always looking at new technologies. We are patients ourselves. We talk to patients. We talk to providers. How can we make that world class?
There are advancements that are ready to be taken advantage of by healthcare providers. We talked about one a lot today, agentic AI, but there are others that are even simpler. Branding is one. How often do you get a text message or a phone call from a provider and you have no idea who it is, so you don’t answer. Then you don’t find out what your lab result is. Did you know there’s technology today that allows for branded calling and branded texting? There’s a new protocol, Rich Communication Services or RCS, that Apple just started adopting late last year.
This is all technology that’s available today that can help build that trust. We are focused on being on the forefront of that, deploying that to our customers as quickly as possible to continue building that trust that they have with patients. Our goal is to be that invisible infrastructure layer. When folks work with us, they know that we will be on top of the best technology and the best possible experience that they can deliver to their patients. That’s what drives and motivates us. We will follow the markets and the technologies that come to bear over the next decade and more to come.
I think fitness as an entrepreneur, as a CEO, comes down to doing the work yourself. I’ve always been in the details. An I building code for agentic AI? No, but could I tell you every single part of our stack? Am I the first person testing a solution? I’m calling agents in French in the morning and I want to know exactly where the technology is.
That fitness is important. Like if you’re an athlete, you need to go on the track every day and make sure that you’re fit for the next race. As an entrepreneur, as an executive, as a CEO, you have to stay fit and sharp, which means you need to talk to customers every day. You need to understand the tech intimately. You need to understand how you’re deploying to customers. It’s one of the most important parts to doing a good job in business.
Hopefully you can tell that I love what I do. I’m obsessed with it. I love working in healthcare.
RE the AI GLP1 company, Washington Post has an article today by someone who used one of those compounded products,…