Abridge reaches to become the operating system of US healthcare
Silicon Valley has been trying to come up with solutions to the U.S.’s health care bureaucracy for generations. But the administrative burden on patients, doctors … everyone, really, has only gotten worse. Myriad studies over the years have identified doctor burnout as a critical problem.
It’s been so hard to fix partly because software historically hasn’t been good at transcribing and analyzing one of healthcare’s biggest data sources: doctor/patient conversations.
The other impediment has been that two of healthcare’s biggest players – hospital systems and insurance companies – are ultra conservative and slow moving. They have healthcare privacy laws to enforce. And lives are literally at stake. In healthcare, a Silicon Valley startup that boasts how it likes to “move fast and break things” gets the door slammed in its face.
Now, AI looks like it could actually be a solution to this nettlesome problem. It easily turns unstructured doctor/patient conversations into structured, easily usable data for patients and doctors. And, as you might expect, it’s set off a scramble among AI healthcare companies for market share.
Epic, the dominant electronic health record firm, said in February that it was adding AI scribe and charting features to its EHR systems. OpenEvidence, which has quickly become the chatbot for roughly half the doctors in the country, rolled out Visits nine months ago and seems to add features to that platform every quarter. Amazon just launched ConnectHealth. OpenAI jumped in last month with ChatGPT for Clinicians.
To me, however, the most interesting company to watch in all this is Abridge. It’s run by a doctor – Shiv Rao – who started it to solve his own paperwork burden. It’s hired some of the top AI minds in the country, including CTO Zach Lipton. It’s software is easy to use and works as advertised, something doctors and hospital administrators don’t see a lot. And it’s been fast and relentless at reinventing itself. That’s hard to do in a sector of the economy that radically changes every six months.
It started life in 2018 purely as a medical scribe. When the AI revolution took hold three years ago it leveraged that into taking doctor-recorded patient visits and drafting entire chart notes for the doctor to review and sign at the end of the day.
Now Abridge thinks it can leverage that foundation into every aspect of health care administration – charting, billing/insurance, research and records. It’s an opportunity in the hundreds of billions of dollars.
When I talked to Lipton about Abridge’s tech stack 18 months ago he hinted that this was where the company was headed – to become the operating system of American healthcare.
Back then, in the fall of 2024, that seemed far off. It had just convinced Epic to integrate with their system and picked up the entire Kaiser health system as a customer. But it was only in about 40 healthcare and hospital systems.
Now the company is going for it. Last week it said it now serves more than 300 health care and hospital systems in the country from the biggest research institutions and the Veterans Administration, to community health centers. It said it now integrates with other big electronic medical record companies like Oracle Health and athenahealth. And it told me that nearly 20 percent of all primary care and specialists serving about 250 million patients use it. That’s about 200,000 doctors on the platform.
The announcements fell into two broad categories: expanding its technical infrastructure and deepening its clinical reach. On the infrastructure side, it announced a deepened partnership with AI chip maker NVIDIA, critical for any AI company that has growing processing power needs. It announced an unspecified investment from Eli Lilly, the pharma giant behind among other things obesity and diabetes drugs Zepbound and Mounjaro. And it said it was meaningfully broadening the capabilities of its proprietary AI models to now include apps and functionality for nurses as well as doctors.
It also announced a slew of research publications including JAMA, the New England Journal of Medicine and most publications focused on medical specialties that will now deeply integrate their content into Abridge’s AI.
And it said it had established linkages with insurance companies like Cigna and Aetna in an effort to make the lengthy negotiations both parties go through over many patients move faster and be more transparent.
Abridge started out as just a tool for primary care physicians in outpatients settings. Now, Rao says that its tools are being used in almost every outpatient and inpatient setting and in almost every specialty. Maybe to prove it wasn’t exaggerating about its industry influence, top executives from insurers Cigna and Aetna, health systems at Johns Hopkins University and Emory University, and chip maker NVIDIA joined Rao onstage for conversations.
Rao demoed some of the tech for me last month. It’s easy to understand how it works and why you’d want to use it. Before a patient visit Abridge prepares the doctor with pre-charted notes and summaries. During the visit it suggests discussion topics based on the recorded transcript real time. Then Abridge generates clinical documentation, flowsheets, patient summaries, billing codes, and orders for the doctor to review at the end of the day.
When a doctor is reviewing his chart note at the end of the day he/she can highlight and immediately view where in the transcript the system pulled that information. He/she can listen to that portion of the recording at the same time. It does all this in more than two dozen languages.
“So I would think about this technology now like a team of assistants that can do all the different types of work that you wish you could get done. And what’s amazing about this team is that they can work 24/7,” Rao told me.
Rao understands that having first mover advantage in a field like AI where the technology is radically changing every six months, is often more of a problem than an advantage. “There’s now three variants of AI native companies,” he told me. “There’s the post transformer paper, free LLM vintage company. There’s the post LLM free agent vintage company. And then there’s the post agent company. So much of the game … now is how fast can you become the latest AI variant not only in the product that you build, but also in the way that you operate.”
It’s certainly made Abridge more careful about how much detail it reveals publicly about itself. A year ago it was happy to disclose revenues – about $150 million – and valuation – $5.3 billion and how much money it had raised -$800 million. But it wouldn’t say a thing about the Lilly investment or current revenues last week. All it would say is that the company’s workforce has roughly doubled in that period to between 400 and 500 people along with the number of hospital systems it serves.
Maybe that’s wise. It’s not farfetched to imagine cut throat competition ahead. Epic could one day stop integrating its health record with Abridge’s AI, for example. Epic has already partnered with Microsoft’s Nuance, once the dominant scribing platform for hospitals.
Anyone who followed the rise of the Microsoft monopoly during the 1990s knows that Windows was not the most technically sophisticated or reliable operating system. But it won.
And what happens to AI in healthcare administration generally if all these systems start introducing errors? All are set up to require doctor approval of any work they do. But humans under time pressure use shortcuts no matter how well trained.
No matter what happens, it’s hard not to be charmed by Rao’s vision.
“I don’t think anyone purposefully designed a system this way. I don’t think anyone set out to industrialize the care delivery experience,” Rao said during the presentation last week. “And if we’re not careful it could get a lot worse. (We could have) AI vs AI. Agents vs agents. That’s probably a race to some dystopic future nobody wants to live in.
“Our opportunity right now is to use AI to actually rethink the system. Can we compress workflows? Can I (the doctor) let (my AI software agent) attend the compliant documentation lunch so I don’t have to? Can we let AI figure out how to get all that clerical work done so I can spend more time with my patient? And what if we can shift as many of those workflows upstream?”
