Alexa for Doctors

I started college with a very divergent mindset. I wanted to explore, figure stuff out, find out what really resonated with me. I was definitely, subconsciously, rebelling against all the Indian doctors in my family and in the world — I wanted to taste as many things as possible.

Part of that was probably informed by the fact that, towards the end of high school, one of my closest friends died. He was murdered. So when I went to college, I came in thinking: no, I have to figure life out for myself. I don't want to go to trade school. There was that silver lining to the loss I experienced as a kid.

I was super idealistic — the really idealistic, art-schoolish kid who had a piece of paper in his wallet with a quote on it. The quote was: "Be patient toward all that is unsolved in your heart, and try to love the questions themselves, like locked rooms and like books written in a very foreign tongue. Don't seek the answers, which can't be given to you, because you wouldn't be able to live them. And the point is to live everything — live the questions now, and perhaps you will gradually, without noticing it, live along some distant day into the answer." That was my mantra in college: I'm just going to live the questions, figure this out, find out what's out there for me and what my place in the world is. It was a very J.D. Salinger, coming-of-age few years.

It was late in college that a visiting professor came to Carnegie Mellon, where I was — William McDonough, an architecture professor. He told this story about how we're all designers of our world. He talked about an ophthalmologist in India who had designed a revolving platform that he sits on. He brings patients in at 12, 3, 6 and 9 o'clock. He does a cataract at 12 and says "spin." Then another at 3 — spin — 6 — spin. He's just spinning all day long, giving eyesight to people. By the time I heard that lecture, he'd given eyesight to over a million people, and he'd taught the procedure to his daughter, who'd given eyesight to over 400,000. I left that lecture mind-blown: this is the kind of value I want to be part of. Not that it's more profound than any other kind of value people create — but this is mine.

So my parents secretly rejoiced that I was now creating a path towards medicine and healthcare. But I had to do a couple more things that have been really foundational for who I am and the work I'm doing today. I got really into music — I've always been into music. I had a scholarship year at Carnegie Mellon where I got to work with virtual synthesizers, do some lightweight programming, and create music. My roommate had a record label, selling records in Japan and all over the world — electronic stuff. We just made music, and it was an amazing experience. And all the while, I was building this path into healthcare, with a sideways approach as a nontraditional applicant.

I got into med school after that, and med school was really boring. A lot of memorization, and I didn't see much room for creativity — the kind that inspired me to get into the space in the first place. When I looked at the ways you could do research, I have the utmost admiration for people who do hardcore clinical or basic research, but I couldn't do that — thirteen years bench-to-bedside. I don't have the attention span for it. So I went to the University of Michigan for more training — internship, residency — and fell in love with clinical care. All of a sudden I could be creative, because that's really what clinicians are. People think they're scientists, but really they're alchemists, artists. They're being so creative, connecting so many dots across a patient's social history, their background, their family, their allergies — and what's going on with them right now — to come up with a plan. And the plan is never the answer to a USMLE or MCAT question. There's always so much more that goes into the art of medicine. So I fell in love with it.

And I also fell in love with this idea of, well, what about tech? Right around then the iPhone had come out — that dates me — and I thought, wait, this could totally change the game on what it means to create impact in healthcare. So I taught myself some Ruby on Rails and created a web app for the residency program that helped residents put all their information in one place, to justify the decisions we were making every night in the hospital, so everyone could learn from each other — this network effect across medical education. And I experienced, for the first time, some peasant-level of product-market fit, because the residency program director said: hey, we might want to pay you for this. And I was like, wait, what?

That was the beginning of me starting to converge a little. Even in med school I had that divergent mindset of exploring, but I started to recognize that the ophthalmologist story — getting to impact at scale — you could do that with tech. All that experience from being an undergrad at Carnegie Mellon suddenly felt like: oh, connect these dots — tech and healthcare. So fast-forward: I went to cardiology fellowship, and had a first startup where at night my co-founders and I were in an incubator at Carnegie Mellon. We were managing teams and engineers, and that was a great way to get steeped in all things startup and learn how you actually build something. We got to term sheets with someone, but it was a good limits test — that idea didn't really meet the mark for what I was trying to get at.

But I stuck around that health system, and got super lucky, because that health system — UPMC — is really large, has incredible scale, but also has an entrepreneurial mindset baked into its DNA. They have a payer arm, insurance companies; a provider arm, 40-plus hospitals and all the clinicians and nurses and doctors within them; but they also had this innovation arm, a tech arm, that they were starting to invest incredible amounts of capital into. So I was in the right place at the right time. People said: hey, you should split your time — half-time doctor, half-time tech innovation. And over the course of five or six years, I violated the Peter principle like ten times. I started as the token Indian doctor on a team who had clinical insights, then graduated to head of strategy, then ended up as the EVP there — which meant I could be a check writer, one of three who could invest UPMC's capital into tech startups, but also into other places. So we invested significant capital into Carnegie Mellon to start a machine learning and health program. And that's where a lot of the founding DNA for Abridge comes from.

Even at that architecture lecture I got to go to as an undergrad — this was when the d.school at Stanford was pretty young — he was talking about T-shaped thinking, T-shaped people. You can be really deep in one domain, but cross the T with a seemingly disparate discipline. And now, as one individual, you're connecting dots and coming up with ideas that would take a lot longer to get to otherwise. Even now, in the work we do on the AI side at Abridge, domain transfer comes up quite a bit — style transfer. Those spaces are so exciting to me, because it's like transposing an idea from one space into another.

As a tech CEO now, thinking about my background as a clinician, there are so many dots you can connect from one domain to another. The key, from a mindset perspective, is that you can't silo your experiences — you let one bleed into another. We were chatting earlier about this: oftentimes being a startup CEO feels like being in the intensive care unit. And that's my favorite place to be in the hospital, because you're in the ICU — I love the urgency and the amount of agency you have as a professional to save a person's life. I'm a cardiologist. Someone comes in with a heart attack, or with cardiogenic shock, and you have to very quickly build a prototype: this is what we think it is. Then you figure out a plan to test that prototype — you start a whole bunch of IV medications, do a bunch of procedures — but then you measure as fast as you possibly can. And depending on the numbers you get, you either pivot or persevere. You drive it to a conclusion, and you save people's lives.

In so many ways, that's what being a startup CEO is like, especially in the earlier stages. Despite having incredible venture capitalists, at the end of the day you're on this journey to get to product-market fit. And on the road to PMF, your patient — your startup — is in extremis. It's in the ICU, and you have to build a prototype and see if it fits, see if it works. You keep building and measuring and learning and iterating, and the faster you can go through those cycles, the better chance you have of your patient walking out of the room — which, in the startup world, is your startup actually having PMF. And then, all of a sudden, the world is different. It's a totally new game.

When we first started the company, we had strong opinions about what the experience needed to invoke for both sides — the professionals and the patients. In our first seed deck, which we pitched to Union Square Ventures, one of our major investors, we told this story about being able to benefit both sides. It's the story we're still working towards, and now unlocking at scale in the market.

What we realized very quickly was that the barrier to entry on building a solution — not just a demo — on the professional side, one that could integrate into workflows and into the medical record systems we were targeting, is high. There's a lot of deep tech you need to do to create the seamless experience you imagine in your head. For us, that meant we needed to do research — AI research — and that Carnegie Mellon DNA gave us the ability to do it. Our UPMC DNA gave us the ability to aggregate de-identified data at great scale, and not just aggregate it but annotate it in the shape and form we needed to build models that could lead to those experiences. So we spent years aggregating and annotating data and building models.

We started the company in March of 2018, just a few months before Google published that paper, "Attention Is All You Need", about transformers — this new type of machine learning model. That was in the back of our heads the whole time, and in all our papers, in some shape or form, we were leveraging a transformer and pointing it at a specific healthcare challenge in our pipeline. While we were doing that deep-tech research — which has now culminated, years later, in a full-stack solution on the enterprise side — we didn't want to hold our breath and just do research. We wanted to create value as soon as possible. So we put out a consumer version of our application. The barrier to entry on the consumer side is so low — what were we waiting for? We knew the future looked like professionals using it and then sharing with their patients; the viral coefficient there is a lot greater than going direct-to-consumer. But why hold our breath? Let's start helping people now.

So in July of 2019, we put a free consumer app out there. We made sure all the privacy and security specifications met the highest standard for healthcare. And before we knew it, we had hundreds of thousands of users on the consumer side — with permission — capturing conversations. Maybe just snippets, maybe just the next steps with their professionals. In that consumer app, we don't just capture the audio; we transcribe just the medical parts of the conversation — sort of an abridged transcript — and then we layer on health literacy. We decode all the jargon the professional didn't even know they were dropping on you, at a fourth-grade reading level. But then, most exciting, we can nudge you in a week or two: hey, we think you were supposed to get an echocardiogram, we think you were supposed to take metoprolol. We help you be the healthiest version of yourself, because there's great research suggesting people forget up to 80% of what they've heard from their doctor or nurse. It's just how we're programmed as humans.

Organic. We put a consumer app out there, and one of the first groups that latched onto it were chronic disease communities. When you think about it — say you're on a cancer journey. It's so hard to be in the moment, especially when you're getting a new diagnosis, but even after that. You're preoccupied with all the anxieties and stress related to your diagnosis, and the medications themselves can cause what they call chemo brain, where you can't remember. So those communities were the first highly engaged, highly activated patients who recognized the value proposition. To this day, every week we get some review or email saying this is the most important thing in their care journey — a game-changer, a lifesaver.

Absolutely. We're already creating value on a lot of the downstream opportunities of the document itself. When Abridge is part of the conversation, we can not just summarize it or create different versions of a note that help both sides — there are other customers of that note. Payers, insurance companies, come into the picture. Another thing that crushes my soul is having to field all the prior authorization requests from an insurance company asking me: why did you order this MRI for a patient with cardiogenic failure? Was the MRI the right thing? Should you order another test? We can start to really unburden people on use cases like that. We can unburden use cases related to the problem list and coding as well. Those are adjacent to the conversation itself.

But even further downstream, there's what you're getting at — a pure clinical-analytics opportunity. In a privacy-preserving, secure way, you can understand patterns across the care being delivered, so you can do a better job of any number of things: from identifying which patients might be good candidates for clinical trials, to understanding sentiment around a specific new medication. There are infinite use cases, because of the thesis — and I think we'll always return to that thesis, as long as it's correct: healthcare is really about people having conversations. All this opportunity is downstream of that. There's a lot of challenge here — we need to be in the workflow, we need to work with the medical records and the other systems central to how care is delivered. But pulling that off, getting into those workflows, unlocks infinite opportunity.

It's a tailwind. These technologies are an absolute tailwind, because what they do is move value up the stack. These foundation models — and other technologies that will prop up over time — are giving us all new tools, new primitives we can plug into our existing pipelines so we can focus on the next problem, and the next, and the next. It'll help companies differentiate on value that's higher up the stack. And it'll make clear how important it is to be in the workflow, to be integrated, to be trustworthy.

It's easy — to your point — but I don't know that it's much easier today than it was yesterday to create a really compelling demo, in this or any other space. You can duct-tape a lot of different services together and create a proof of concept. Especially in industries like healthcare, that's an important conceptual first step. But the real work is ahead of you: creating something that's trustworthy, that can integrate into workflows, and actually reliably works. That turns into the moat. We've always recognized that in healthcare especially, trust is the most important network effect.

These new foundation models have changed the game in the sense that they differentiate the AI-native companies from the companies that are thin layers on top of a foundation model — who need to optimize for a totally different go-to-market and growth strategy. If you're an AI-native company, you recognize that the moat is in building machine learning beside those models, on top of them, beneath them. There are companies that need to fine-tune those models for their specific use case, which means they need proprietary datasets to pull that off. If you're not doing that, then where is your moat? And number two, how are you going to build trust as an enterprise-grade company when, chances are, every time you hit that model you get a different answer? So the name of the game right now is: yes, you can create compelling demos with these new primitives and demonstrate a concept, but the real work is still ahead of you if you're trying to build an enterprise-grade company.

Now, if you're a consumer company that's all about growth first, and you've been financed to go after a growth-first strategy — even in 2023 — then what you need to do is build that thin layer, put it out there, aggregate users, and figure out how to build a traditional network effect or another form of defensibility in the future. But that's not our game. We're trying to build real, deep, AI-native. So for us, all the investment in machine learning — and having a chief scientist like Zack Lipton — is critical to our identity.

The reason most startups go sideways is a lack of focus — that they didn't sequence the steps in the right order. Every startup needs to know when to play chess and when to play Monopoly. Being able to shift gears between chess and Monopoly — to know when it's time to say, you've explored enough, now it's time to exploit and create value — that can be a difficult decision for leaders to make. And timing is a big part of it: knowing when it's time to shift from one gear to another. When I look at this dispassionately from the outside, in health tech especially, it's the companies that didn't time that mindset shift that had the worse, suboptimal outcomes. It takes a certain mindset, and a certain ethos in the boardroom, to always invoke first principles, look at the space dispassionately, and make those decisions.

To some extent, being able to channel your own intuition is really important. As much as people want to make this a numbers game, there's so much intuition that goes into it too. You feel it in your bones, sometimes, when it's time — and we've certainly felt that way for a while now, so we're in a different gear as a company at the moment.

The things that can really knock you off your game are operating from a perspective of scarcity, or fear. As soon as those insecurities start to seep in, your vision of the market and your company starts to get obscured, and you turn into that drunk person under a lamppost, looking for their keys, just hoping something will work. But when you're operating from a place of abundance — when you're always trying to channel clarity and your own inner conviction, and that compelling future that helped you start the company in the first place — you end up making the decisive decisions that won't be informed by all the data, but that help you decide which adventure to choose at any given time.

How you feel inside — that internal milieu — is how you project yourself externally, whether to your employees or to whoever. So it's so important to always be locked in. If you start to see or feel yourself fray, you've got to take that beat to get back into gear, because it's going to have ripple effects — it'll affect everything across the company before you know it.

The three things I think are so important to channel, for yourself and externally, are: clarity — be clear about what you want to achieve, so you can stay focused; certainty — you need to feel it in your bones, what you're doing and why, so it's important to start with the why. These are all airport-business-book sayings, but there's profundity in them. And finally, you need to always channel that compelling future, which is part of that why. Why are you doing this? Why would you be willing to work like an intern every year, and probably forever, for this mission? There's a compelling future you're working backwards from, and being able to stay as close to that as possible makes you indefatigable. It gives you the energy reserves to do anything — because the line between work and life is totally blurred. You're doing this because you want to. There's no distinction for me on a daily basis. If someone asked me how many hours I work a day, I don't know. It's all the same, and it's all amazing, and I'm grateful for all of it.

Yeah, absolutely. As CEO of the company, I have five jobs. I've got to nail down the strategy. I have to deliver the capital for us to pursue that strategy. I have to build the team to execute against it. I have to channel clarity, certainty, and a compelling future in how I communicate that strategy, internally and externally. And the last piece — the one you're invoking — is that I've got to hold people accountable, and hold myself accountable. There are absolutely skills on the accountability side that you have to build as a manager.

Part of this is just dispassionate. You've got this why, this mission you've got to achieve, and certain metrics you're holding yourself and everybody else accountable to. Everybody's on the same page on those metrics, everybody's aligned, we're pushing the same rock up the same hill. At that point it becomes a dispassionate conversation: here's where we are, here's what we've done, here's what we've not done, what's going on — and you move from there. So I don't see that as a big challenge, so much as a thing that's just part of the job, and part of the mission.

I have a mentor who reminds me of a few things I should always keep in mind. One is perspective. Another is: what's your purpose? And the other is gratitude. His advice was to try to channel those three things at night, before you go to bed. You'll wake up with all this drive — you'll feel unfulfilled, and that's the fire in your belly to go do something. Execute, get stuff done. But then, at the end of the day, try to pop your head up again and get to that meta level: what's going on, what's the bigger picture, how do I fit in, what's my purpose, am I achieving those things? And then, naturally, gratitude flows from there.

Another part of my daily regimen is to take a beat at the end of the day to unplug into something else, where I can lose myself a little — get into that no-ego state. Whether that's playing chess with my kids, working out, or listening to music — being able to get into a flow with something outside the norm allows the ideas inside the company that I have to incubate a little better and faster.