Abridge's Shiv Rao: The Doctor Founder Taking On Microsoft In Healthcare AI

0:00My wife and I went through, she as the main character, me as the family member, three years of IVF with PGD. I'd go to every single visit with her, I'd be there, and the doctor or the counselor would leave and we'd look at each other and be like, what did they just say? Like, what are we supposed to do? And I'm a doctor. Doctors and clinicians are burned out. And their patients don't understand what they're being told, which means whatever it is, they're probably not going to do it. At the same time, the U.S. healthcare system is drowning in paperwork. to the tune of$1.5 trillion in admin spend every single year.

0:31Today, we're talking to a founder who's trying to bridge those two problems. His startup, Abridge, is building an AI operating system for healthcare. And in 2025, it was valued at$5.3 billion. More recently, Abridge announced partnerships with Eli Lilly and NVIDIA, and it works with more than 300 healthcare systems. Founder and CEO Shiv Rao is a practicing cardiologist even to this day. And he says it bridges values help define its positioning. All companies need to have one strong idea at least that they hold tightly to. And that's the North Star. Otherwise, especially in healthcare, you get blown in so many different directions.

1:06And like when you're at scale, as we are, there's also so many opportunities. There's a lot of low-hanging dollars. We want to make sure that we're going after the dollars that build trust. We're not going to put ads into that workflow that sacrifice trust. We are always going to be in service of delivering a better experience. and a better clinical outcome. We want to be deflationary. We want to save time, save money, save lives. Today on the Upstarts podcast, we're going to talk about how to position your startup against the Microsoft giant, the three phases of AI companies in health, and how to create a category, then compete as it gets crowded too.

1:41Plus, why finding scale in the healthcare system is like eating glass and kissing a bunch of frogs. I'm Alex Conrad, founder and editor of Upstarts Media, and this is the Upstarts podcast, our weekly show where we talk to startup founders who are punching above their weight to take on the status quo. Shiv, thanks for joining us on the show. Thank you. Super excited to be here. Now, should I say doctor, Shiv Rao? Because you are a practicing physician as well, right? Yeah, yeah, I guess so. Weekend a month, not much. This episode is brought to you by Rippling AI, the only AI built to give you full visibility into your startup and the ability to take action across every department.

2:16So Shiv, in a nutshell, what is Abridge doing for doctors, patients, you know, your customers? Yeah, we're using AI to bring doctors and patients closer together. That's it in a nutshell. We're trying to do all the work behind the scenes that would allow them to just focus on each other. And it allows us to save time, save money, and really save lives too. Health care is a huge industry. It's like 18 % of GDP. And we are building the AI layer for health care. And where we've started is health care delivery. Because when you think about the pain points there, clinicians are burning out. Doctors, nurses don't want to be doctors anymore.

2:52Patients can't understand what's going on, much less follow through to be the healthiest versions of themselves. and we're wedging into the moment, the sacrosanct moment between them to help both sides of that room. And where we've started is that conversation. We believe that it's computable and that you can build a platform on top of it. So we started out helping clinicians with clerical work like notes, placing their orders, doing accounting, doing billing, what they call revenue cycle. But now we're also helping with clinical intelligence, with decision support so that they can practice at the top of their license, as they say, so they can not only deliver a better experience, but also hopefully deliver a better clinical outcome.

3:32Should we be thinking about Abridge as a software business, as some sort of operating system? I know you resist the idea of being an AI scribe, but how should people think about the output of what Abridge is doing? Yeah, I would think about us as an AI native company for healthcare. All of us have probably seen the doctor or we've had a family member who've seen a doctor at some point in the recent past. And you're in that room and sometimes those doctors are stressed out. Sometimes they're not making eye contact. Their back is towards you because they're typing and doing all this other work that they're asked to do that actually distracts them from delivering the best possible care.

4:08And so with a bridge now, they're more present. They're making eye contact. They're fully there because they essentially have a team of assistants of agents from a bridge that are doing all of that other work for them behind the scenes. And some of that work is helping them write their notes. Some of that work is helping them with billing and accounting and revenue cycle. But some of that work is actually helping them now make better decisions. So when they're with you, they know that maybe your symptoms represent some rare disease and they should consider that. Or maybe you're a candidate for a clinical trial and they should tell you about that.

4:42Or maybe they should consider this therapeutic instead of another. And so, again, the idea is save time and we can demonstrate that. Some clinicians are saving two, three hours a day. We want to save money for the healthcare system. Like we have to figure this out as a country. You've seen those Bommel's cost disease curves and healthcare is out of control. So we believe that there's no other industry where AI can kind of create the kind of impact that we can create in being deflationary. But also we want to save lives. We want to really, you know, help every single doctor, nurse out there feel like a superhero.

5:16You work with over 300 healthcare systems, right? Including folks like Kaiser Permanente, Johns Hopkins. Would this be something that the doctors are bringing into the system or it's more top-down where a system is rolling you out across everybody? And then would the patients in these systems be aware that a bridge is operating or is it more behind the scenes? It's top-down. Everyone's aware. And for both reasons, I think related to both, it's really all about trust. This industry, I think, in a very special way, moves at the speed of trust. And if healthcare systems don't trust your solution because you can't check off all the security boxes, the compliance boxes, if you don't have the people and the processes, then they're not going to give you that data.

6:00This data is really private. It needs to be kept very secure. It's everyone's healthcare data. And that's the context that we have to engineer in order to help in the workflow. The other piece is that we are in workflow. So this isn't something that's working after the fact. It's actually a part of those moments in the emergency room, in the operating room, in the clinic, or, you know, in the hospital. And so if everyone doesn't know that this is taking part, that this is helping everyone involved, it wouldn't really work. What is the experience like that you observe for a patient when a bridge is live versus when a bridge is not present?

6:39When a bridge is not present, patients, family members maybe aren't having a great conversation in the first place. They don't know what's going on with their care plan. They aren't getting a summary at the end that sort of creates a digest of like the most important parts of the conversation they just had and what they're supposed to do, their care plan. So they can't understand and they're like less likely to follow through. they're also having a moment with their clinician where they're stressed out and the clinician is also stressed out. And so that opportunity to really build a relationship isn't there.

7:19We just did a keynote in New York city, um, some weeks ago, and we told the audience that the most exciting, inspiring demo that we could possibly do would just be people talking like this, you You know, like no technology actually getting in the middle because the technology is in the background doing all of the work for them behind the scenes, but in real time and in a way where both sides benefit even before the conversation, but certainly during and absolutely after. And the way patients are benefiting after is they were creating these summaries that are written at the right reading level that can really help them during the conversation, you know, better interaction.

8:00but doctors are benefiting even before they walk in the room because we can help them understand what questions to ask, who this patient is, what their concerns are, why they called the clinic a couple weeks ago. You know, they can kind of come in and really instantaneously build rapport. How do you know that info? That's the context engineering piece. And that's why I think our go-to-market is so focused on enterprise. There's so many important hoops that we've got to jump through in order to build the trust of the CIO of a large health system who's going to allow us to ingest a bunch of data out of these systems of record, like the EMR, for example, that we can then engineer through these models to help the clinician know who this patient is, you know, what they should talk about.

8:43And then we need to create the note the right way on the other side. So I'm a patient, I show up, is this a scheduled visit? Is this a more urgent situation? Could it be both? Yeah, it could be both. It could be both. I think like the beauty of it is like once we get through the enterprise trust sort of gauntlet, then we're in and clinicians can use us as they see fit. You know, they can use us in the urgent care. They can use us in an exam room. I think more than anything, all those clinicians can now feel like technology has their back. There's this team, you know, increasingly like that's the concept is there's this team of assistants that are working for them 24-7, you know, doing all the work that helps them deliver a better outcome.

9:23So the physician knows me better once I come in. Then do you say, hey, I'm going to be taking notes of this visit using a bridge? Yeah, that's what doctors say. So like, hey, do you mind if I use this note-taking solution? It's going to help me focus on you. You're going to get something out of this as well. You're going to get a better summary that really captures what we talked about. And I'm going to get a lot of help with a lot of the work that distracts me. Do patients ever push back that they're just like scared of AI in general? They're like, I'm not comfortable with this. And if so, what is your response?

9:56You know, not much. You know, like I'd say the zeitgeist has shifted so, I think, profoundly over these last years. Like in 2018, for example, the zeitgeist was not meeting notes. It wasn't like capturing every Zoom call. It wasn't, you know, going back to these summaries. And I think GPT hadn't taken off. And like this idea of summarization wasn't as mainstream as it is now. We're in a new world. I think really the sky started to open up in 2023. With humility, our team can take a bunch of credit for normalizing this idea of AI in the conversation and healthcare over these last few years as we've scaled.

10:35But now when clinicians tell their patients about it, there aren't many questions. They sort of intuit already what this is all about. Okay. Let's go back in time. Yeah. So you're practicing, you're making the rounds yourself. Then you start to think about sort of more companies and startups in this ecosystem, right? Tell me a little bit about what you were doing in that run up to a bridge and what the sort of light bulb moment was, if there was one. Yeah, yeah, totally. So prior to this, I was a corporate VC for a large health system. So we're putting a bunch of money into startups, putting a bunch of money into Carnegie Mellon.

11:06We started a machine learning and health program. And were you just already interested in startups? Yeah, yeah, yeah, totally. Had a first small little idea that I was pursuing with some MBAs a lifetime ago and got to learn a little bit about what that looks like, what that feels like. We had employees, we're building, like we got to an MVP, angels wanted to give us money. And at the time they told me I couldn't be half pregnant. If I was going to take the money, then I'd have to like give up being a doctor and just fully focus. We saw that that did not work. You stayed a doctor. Yeah, I stayed a doctor.

11:38But back then it was, it pivoted you a little bit. Yeah. Yeah. I was like, I think right place, right time. And this large health system, UPMC in Pennsylvania was starting to put a ton of money into innovation and I got to sort of split my time. So it was like halftime cardiologist, halftime on the innovation and corporate venture side, got to sit on boards, learn osmotically from founders, from investors, and got to learn from professors at Carnegie Mellon, especially as deep learning was like starting to really take off in the mid 2010s. And, you know, even in terms of our timing of starting this company was three months after attention is all you need.

12:16I think that there's like three vintages of AI native company. there are post-transformer paper pre-LLM companies post-LLM pre-agent companies and then now post-agent companies and the name of the game is to be the latest variant as fast as you possibly can and it's not just into in terms of the product that you're building but also in terms of the way that you operate as a company so we were post-transformer we started with BERT and BioBERT and Longformer and Pegasus and T5 we were fine-tuning all of these these transformer based models. And then all those skill sets like turned out to be incredibly valuable.

12:53Like we just announced a relationship with NVIDIA and we're like building a foundation model at the conversational level for healthcare. But all those skill sets, we still apply to this day, but when LLMs came out, obviously you refactor very, very quickly what the product can look like, what the technology can look like, how fast you can go, how fast you can scale. And now in this agent moment. It's just an entirely different game as well. And I think the companies that are going to win are going to be the ones that are super agile because the ground's going to continue to shift on us. We've talked to people on the show like May Habib, who was working with the Transformer basically as soon as the paper comes out.

13:30And then I've also spoken to people like, we both know the CEO of Zoom, Eric Yuan. And Eric was telling me recently that he actually beats himself up a little bit that if he's honest, he didn't jump on LLMs and sort of the AI potential until that chat GPT moment. Early on, were you betting a lot of the company on this technology? And I guess how much was a tech moat going to be key to a bridge, at least in the early days? Yeah, we placed a really big bet that this was going to work, that somehow we were living, we're at this intersection of moments in technology and also in healthcare and on the tech side like deep investments and annotations like when we started the company we were finding ways super creative ways because early stage startup it's rounds weren't what they look like today in those seed rounds but we absolutely felt like outsiders like we were trying to do something that in so many ways was going against the grain.

14:31Number one, just in terms of the product, recording conversations, not normal in healthcare with HIPAA and all the other privacy issues, really like going against the grain. This thesis about AI and healthcare taking off, not, you know, had not taken off. And so this idea that the conversation could be computable, that you could build a platform on top, definitely something that investors had a hard time sort of grappling with. But then from a technology standpoint, really believing that over time, this would unlock. And at the time, pre-LLM and pre-this moment that we're in, the name of the game was fine-tuning these models, but it involved lots of really expensive annotations.

15:17So we were getting really creative finding ways to get non-dilutive capital, to employ a bunch of doctors and nurses, to go through data sets and create annotations that could allow us to even get to the contours of an MVP. So, yeah, it was a very different game, but I think it's helped us. Can you give me an example of what a creative way to find capital would look like? Yeah, it was like going to UPMC, for example, and trying to frame the challenge as research and trying to find a way to get them to foot the bill on a data that they could help, that they could use as well, but that we would obviously find a way to leverage.

15:54As a corporate VC, I one day just went to my boss at UPMC, a president over there, and just said, like, I quit. I'm going to start this company. I didn't want it to be a spin out. And I just wanted to make sure that we could own and control that P &L so that we could, you know, control our own destiny. And I also said at the moment, like, hey, like, I'm going to quit. I'm going to start this company. I'm obsessed with this problem with my co-founders. And you're going to give us a check. and you're also going to give us all your data. And then we're going to annotate as much of that data as possible and we're going to build something.

16:31So it was a very... What was your boss's response? It was a very YOLO moment, but we got there. And I look back at those emails, I was looking back at them recently and I'm not sure exactly how it all worked out, but it did work out. And they agreed to that? Yeah, they did. They gave the data? They gave us a lot of de-identified data in a responsible way, but without that data, we couldn't be an AI company off the bat. without that data, we couldn't have attracted these professors or PhDs to work with us. So it gave us something that allowed us, even if we weren't capitalized as such, it made us operate like we were a deep tech company, even off the bat.

17:05When you guys started in 2018, what was the first version of the product? What was the wedge that you guys were providing? Yeah, totally. When we first put our seed deck together, there was one slide where there was a patient on one side of the slide and there was a doctor on the other side. and then there was our application in the middle. And the idea was we can serve both sides. And the barrier to entry on the doctor side, enterprise healthcare, not for the faint of heart. You have to go through a crazy idea maze. It's a labyrinth to sort of figure out how to get out there and actually get distribution and create value.

17:38So there was also a whole bunch of R &D sort of risk that we had to navigate as well. And we had the types of PhDs and professors and postdocs and we had data sets that allowed us to sort of tackle that, But we didn't want to hold our breath on Science Challenge. And so we put a direct-to-consumer app out first in 2019. And that helped people, all of us, capture conversations with doctors with permission. And the idea was to help them sort of be the healthiest versions of themselves, help them understand what they talked about, help them understand their condition, but also understand their care plan and hopefully follow through on it.

18:12But then in 2023, obviously the sky opened up on generative AI and we were able to rush into that moment and really sort of recognize that the center of gravity to really create scale here is the enterprise. And so now it's finally, you know, overnight success, I guess, like six or seven years in the making. But we go through the enterprise. We help clinicians. Then the clinicians sort of extend to their patients and we can create impact at scale across a big swath of the country right now. What was the capability or the aha moment that was able to then unlock some of those first customers, you know, in the health systems?

18:48We were eating glass for many years. And while we're eating glass, like you just have to kiss a lot of frogs. You have to a lot of business models. You have to like figure out a lot of go to markets. There's that Alex Rumpel quote, startups get disruption when they get distribution faster than incumbents get innovation. It's like every Clayton Christensen book in like one line. And in healthcare, finding a way to get distribution is hard. And so much of it is like building trust. And so much of it is finding a way to de-risk this because it's such a risk-averse industry. So I'd say aha moments for us.

19:24I don't know if there was one because there was like multiple years of us pre-selling and figuring out how to position ourselves. And 2021, 2022, especially 2022, we had the contours of a real product that could work. That was pre-LLM. It was with all that, you know, all those annotations and fine-tuned models. And then GBT comes out and we recognize, like, what's about to happen. We very quickly refactored. And once everybody in healthcare as well had used ChatGPT and sort of understood, hey, there's something here we should take seriously, thankfully they called us back. so all those calls that we made in 2020 2021 and 2022 like all that hard work ended up working out ended up paying off because we got to pilot once we piloted and demonstrated results it was like game on I remember in that pre chat gbt era when I was at Forbes our health editor and I looked a lot at IBM Watson we were trying to understand why it had shown so much promise and gotten a lot of exciting kind of announcements to be doing some of this work to digitize and improve the health system.

20:33I even remember seeing a demo where they showed an x-ray and like scans and pre-LLM AI was going to help with a diagnosis. It feels like a lot of that wasn't realized until this most recent era. Why was that looking back? I mean, what they got right is healthcare, huge, huge market. And the market has a labor addiction. One out of five Americans works for healthcare in this country. Health systems can't hire people fast enough to get all the jobs done. There are tons of jobs that nobody wants to do, no human wants to do. So it's the perfect opportunity for AI to assist and to automate as much as possible.

21:12Our concept is like, do it to bring the people closer together. And if you think about it, the people who matter most in healthcare are all of us as patients, and then our care teams. And if you can do that, you can refactor how healthcare is experienced, how it's paid for. I don't know that IBM Watson didn't recognize at a high level, huge market, huge opportunity. Where they went wrong is like, they just got so far ahead of their skis in terms of what they were saying they could do and the products that they could bring to market. We all ended up in the industry paying for it because it eroded trust.

21:45Like you sell these huge stories and then you don't deliver, it really comes back to bite you. So now we're at this moment though, like even at this keynote that we just had, it was like everything that we're going to demo for you is going to be ready for you in a matter of weeks. We're going to make sure that the industry understands like you don't need to, like we're not shipping slides, like we're shipping real product and you're going to be able to, on all your clinicians, your patients are going to be able to benefit from it in a way that like, you know, IBM could never promise. These days you can chat with AI about almost any business problem.

22:21Rippling AI is built to actually solve them. That's because Rippling AI is built on your live workforce data that gives you full visibility into your startup and the ability to take action across every department. Say you're planning your next few hires. Just ask Rippling AI, what would it cost to add two more engineers this quarter? You'll instantly see a breakdown of comp ranges by level and location. So you can make the smartest hiring decision based on where your team has gaps and where the talent is. But it doesn't stop there. Rippling AI can draft the job wrecks, route them for approval, and kick off the hiring workflow.

22:54All you have to do is tap confirm, and then get back to building. Don't settle for AI that's all talk. Head to rippling.ai slash upstarts and get AI that turns insights into action. That's r-i-p-p-l-i-n-g dot a-i slash upstarts. Sign up for exclusive access today. When you were going in to sign one of these big health systems, would you try to talk to both the doctors in the halls and the sort of administrators at the top at the same time? And would the message be different based on who the audience was? Yeah, totally. The personas are very different. So healthcare is a hard sell on the enterprise side.

23:32When you think about healthcare, I would think about health systems as one stakeholder. Another stakeholder is insurance companies. And then another one is pharmaceutical companies like the life sciences companies. If you can find a way to align across all three, like you're transforming things like you can get access to new business models that actually benefit people you can find a way to be cheaper better faster and that's like that's what we're aspiring for and it's all starting to come together for us now finally now that we're we're at lever like we're at scale and now that we've earned trust we can we're starting to like thread the needle across all three in a health system itself that's the hardest part i'd say to go after this bigger picture.

24:13There are companies, there are startups out there fully focused on pharma. There are companies fully focused on insurance. But I think to create this three-sided network, to go after this bigger operating system, you have to start provider first. Because that's where the value is actually exchanged. That's where care is delivered. So going into providers really hard, sales cycles can be like 18 months long, historically. So when we were first raising capital, all these VCs would look at this industry and be like, oh my God, this is never going to work. Like, do we have the patience for this? And we had to keep telling them, look, it's going to be different.

24:50Like AI is going to make it different. And it has been different. These last four years haven't been hockey stick. It's more like telephone pull growth for us. We're at scale now across, I mean, our health system logos represent 250 million Americans. That's how much scale that we've got, the ability to impact really everybody. But on the health system side, how do you do this? CMIO, that's the doctor sort of persona. And they just want the best tools that can help them deliver better care. They don't want to have to worry about what they call pajama time, doing clerical work after the kids are in bed, after they've eaten ants, like they're just sitting in their pajamas and they're writing notes and placing orders and making prior authorization calls or completing paperwork for clinical trials.