Health systems embrace AI scribes as business booms

Racing to combat burnout, health systems and doctors are flocking to technology at an unprecedented pace as they test and trial AI medical scribes.

Why it matters: Notoriously technophobic hospitals are jumping to adopt tools from Big Tech companies and Silicon Valley startups alike.

Zoom in: Investors have flooded the zone. AI scribes raised $800 million in 2024 alone, compared to $390 million in 2023, per PitchBook. As the space crowds, it's becoming ripe for a massive shakeout.

What we did: For this story, Axios interviewed health system leaders at UCSF, UC Davis, Kaiser Permanente and Providence Health, as well as CEOs of AI scribe companies including Nabla and Abridge.

The intrigue: UCSF requested that the names of the companies they were using remain anonymous, but the other health system leaders say they are using one or more tools from just two companies: Abridge and Microsoft.

• Several health systems say they have trialed or are currently testing multiple companies' AI scribe tools, while others have opted for a single vendor.

What they're saying: Leaders from all four hospitals say they view AI scribes as vital for provider recruitment and retention.

• AI scribes are "absolutely now essential for us to manage the [patient] volume," says Rick Campbell, VP of The Permanente Medical Group (TPMG), adding that "using AI scribe technology is critical to us being able to have an advantage in recruiting" as well.
• "These are set to become a commodity in the coming years," says UCSF VP and chief health AI officer Sara Murray. "I think health systems will be forced to deploy them; if they don't have them, physician retention is going to be an issue."

By the numbers: While all four health systems began piloting AI medical scribes in 2023 or 2024, estimated adoption rates vary widely.

• At Kaiser, "at least 65–70% of all physicians are using some elements of Abridge AI scribe technology," says Campbell. (The group launched Nabla in 2023 but switched to Abridge in 2024.)
• At UCSF, 800 out of 2,000 eligible ambulatory providers are using AI scribe tools, or about 40%, per Murray.
• At UC Davis Health, 350 out of 800 eligible ambulatory care physicians are using Abridge's scribe tool, or about 44%, and the system is rolling out the tools to roughly 100 new clinicians per month, according to executive director of the UC Davis Medical Group Debbie Aizenberg and director of clinical informatics Melissa Jost.
• At Providence, 1,700 providers — most in outpatient care — are using Microsoft's AI scribe tools out of about 6,500 eligible providers, or about 26%. The system is expanding its use of AI scribes to ER physicians and advanced practitioners and plans to roll them out to inpatient physicians next.

Faster implementation

Unlike electronic health records (EHRs), whose implementation process can take health systems years, AI scribes can move from pilot to full implementation in months, the leaders say.

• The required Abridge training for UC Davis providers, for example, is 11 minutes, says Jost.
• As they trial various tools, health leaders are assessing their accuracy, ability to integrate with existing EHRs, data security protocols and customization capabilities.
• "We really wanted to look at a technology that wasn't going to bring providers out of the system," says Jost, noting that integrating Abridge with its Epic EHR took about 60 days.

Between the lines: Each system also takes into account provider satisfaction and insights. Kaiser, for example, has physician technology leads in each department charged with sharing product updates and collecting feedback.

• UC Davis' Debbie Aizenberg highlights the system's use of weekly feedback focus groups, while UCSF's Murray says it works closely with vendors to share providers' thoughts as the tools evolve.
• "Even when we don't see dramatic improvement in time savings, [providers] describe this tremendous release of cognitive burden," says Murray. "They feel like they can just sit there and have a conversation with their patient and look at them."

Ripe for rollups

Health system leaders foresee consolidation and attrition in the scribe space similar to the trajectory of EHRs in the 2010s.

What we're watching: "Ultimately, I suspect it's going to be very much like Epic, where the big successful ones buy out the small[er] ones ... and then there'll be two or three [left] 3–5 years from now," says Kaiser's Campbell.

State of play: In the largest tech M&A deal of 2021, Microsoft acquired Nuance in an all-cash transaction worth $19.7 billion — and money poured into the medical scribe space thereafter.

• Philadelphia-based Abridge has raised $212.5 million from investors including Bessemer Venture Partners and Kaiser Permanente Ventures and was valued at $850 million in February 2024.
• Silicon Valley-based Suki has raised roughly $120 million from investors including Venrock and Flare Capital, per CEO Punit Soni.
• San Francisco-based Augmedix was acquired by Commure in an all-cash $139 million deal in October 2024.
• San Francisco-based Ambience has collected roughly $100 million, per a spokesperson, from backers including Kleiner Perkins and the OpenAI Startup Fund.
• U.S. and France-based Nabla has raised $43 million, per a spokesperson, from investors such as Cathay Innovation.

Moving beyond transcription

Health systems' wide embrace of AI medical scribe companies gives the businesses a critical foothold alongside established tech vendors.

Why it matters: That foothold could turn competitive as scribes creep into increasingly advanced clinical terrain.

Reality check: Most health systems are only using or testing AI scribes in one or two practice areas, the most popular being primary care.

• That's largely because the tools aren't capable of handling the nuances and intricacies of other fields, such as surgery or psychiatry, the way human providers can, the leaders said.
• "The AI scribe doesn't really get the essence of some of the mental health exam," UC Davis' Aizenberg says.

Case in point: One area the tools don't currently help is pre-chart planning — or supporting providers as they hand off patients from one caregiver to another by summarizing the notes ahead of time.

• "Being able to go through the [previous provider's] note and leverage — like with the lens of an otolaryngologist or a rheumatologist — all of the information and pull it into the note ahead of [the next provider] actually walking into the room" would be a boon for the system, says Aizenberg.
• Another area of potential improvement is customization to align with a provider's writing style — concise versus verbose — or way of thinking.
• "The note-taking process oftentimes is used for clinicians to really think through as they're developing a plan, so sometimes, when a clinician is kind of forced or directed to think or synthesize information in one way, that can be challenging," says UC Davis' Jost.

The fine print: UC Davis, Providence and Kaiser require their providers to tell patients they are using AI scribes. Abridge and Nabla suggest the disclosure, but do not require it.

• "We don't establish policy for organizations who are deeply responsible and have their own well-considered frameworks," says Abridge CEO Shiv Rao.
• Campbell says Kaiser takes the onus of training its providers to get consent while initiating patients virtually or in person. "We feel this approach is essential is preserving our patients' trust," he adds.
• At UC Davis, providers "verbally consent patients at each and every outpatient clinic visit and document that consent within their note," says Aizenberg. The system gives patients information about AI scribe tech before and during visits via physical handouts.

What's next: AI medical scribes only scratch the surface of generative AI's potential in health care. Industry experts see opportunities in other lower-stakes areas like billing, as well as higher-risk applications like suggestions on diagnoses or interventions.

• That latter category is known as clinical decision support (CDS) and is a category of AI tool currently regulated separately from scribes because it can directly influence patient care.
• CDS is "one of the things" on Abridge's road map, says Rao. "We don't have that ... feature live today, but we're certainly doing the R&D to be able to deploy that in the future."
• "That sort of agentic decision support model obviously has way higher stakes associated with it," Rao adds. "We'd have to be very, very careful in terms of where we point it."