Where Healthcare AI Ambient Scribes Are Heading

2024 has been the year of the artificial intelligence ambient scribe. AI scribe companies have captured the headlines, entered countless healthcare organizations, and raised hundreds of millions of dollars. Simply put, scribes are today’s leading generative healthcare AI use case.

Still, this nascent product class is quickly evolving. Let me explain how it will shake out.

How AI Scribes Ease Documentation Burdens

For every 8 hours of scheduled clinical time, clinicians like me spend an average of 2.3 hours documenting. This work steals our attention, zaps our energy, and contributes to burnout.

AI scribes help automate this process. First, they use automated speech recognition to transcribe clinician-patient conversations. Next, they pass these transcripts through large language models to generate structured clinical notes for clinicians to review, edit, and finalize.

Plenty of anecdotes, and a few peer-reviewed studies suggest that AI scribes reduce documentation time and mitigate burnout.

Are AI Scribes Commodities or Differentiated Products?

With 50+ currently available products (see figure), AI scribes seem to be becoming commoditized. Still, products differ in more ways than price alone.

Accuracy: Many AI scribes are essentially wrappers around products like ChatGPT, making them relatively inaccurate and prone to confabulate. Conversely, AI scribes trained on medical conversations generate more accurate notes. Some products also allow clinicians to cross reference note text with transcribed conversations, allaying concerns about hallucinations.

Specialty and Setting: While most companies design their products for primary care clinics, two-thirds of clinicians work elsewhere. Several companies tailor scribes for other specialties (e.g., orthopedics and cardiology) and settings (e.g., emergency medicine and inpatient care).

Personalization: Many AI scribes generate standardized notes that tend to read like a computer wrote them. Instead, some allow clinicians to specify note length (brief versus detailed) and format (e.g., paragraphs versus bullet points) to better match their preferred style.

Usability: Clinicians using standalone AI scribes must visit the vendor’s app or website to review, edit, and copy note text and then paste it into their EHR. Those using integrated products can use their EHR’s native mobile app to record conversations and review and edit the AI-generated text directly in their EHR.

Notes Are The Wedge Into Clinicians’ Workflows

As Nabla CEO Alex LeBrun explained, “Clinical documentation is not a goal per se; it’s the lingua franca bridging various downstream processes.” Scribes are adding numerous capabilities beyond note writing, including:

Clinical summarization: A little-known secret is that specialists and inpatient doctors often write most of their notes before seeing their patients. AI scribes are poised to help automate this “pre-charting” activity by automatically summarizing relevant details of patients’ histories.

Orders and Referrals: human scribes (typically pre-med students) who sit in the exam room with the patient and doctor not only write notes, but also enter orders (e.g., tests, medications, referrals) for clinicians to review and finalize. Many AI scribes will soon do the same.

Coding: AI scribes will soon recommend billing and diagnostic codes while ensuring notes include the details necessary to justify billing. Ambience reported that its coding tool increases revenue by about $5 per visit, possibly enough for a positive financial return on investment.

Patient summaries: AI scribes can generate after-visit summaries, including patient instructions in English and other languages.

Discrete Data Capture: Clinicians and nurses continually enter data into discrete EHR fields within flow sheets, problem lists, clinical registries, and more. AI scribes could automate this painstaking work, transforming unstructured conversational data into structured text.

Clinical Decision Support: While offloading “administrative” work is a noble cause, supporting better clinical decision-making adds far greater value. Abridge has taken an initial small step by linking notes to UpToDate’s evidence-based recommendations. It is easy to foresee scribes nudging clinicians to close care gaps, recommending differential diagnoses, and more.

The point is that the “scribe” moniker is frequently becoming outdated. Some companies are rebranding their products as ambient and conversational “intelligence,” “assistants,” and “copilots.” Importantly, as I will explain, moving beyond note writing requires deeper EHR integration that only some AI scribes can achieve.

The Small Practice AI Scribe Market Will Remain Fragmented

In small practices, price-sensitive clinicians both buy and use the software. So, AI vendors typically use a product-led growth strategy, essentially giving clinicians a light version of their product for free, hoping they will later convert to long-term paying customers.

Yet integrating across a longtail of small practice EHRs may not be worth the effort. Consequently, we will likely see multiple AI scribes with relatively limited functionality competing to serve smaller practices. These scribes will likely differentiate by tailoring to specific practice types (e.g., physical therapy or behavioral health). Some vendors may give these products away for free as part of a larger bundle.

The Large Health System AI Scribe Market Will Consolidate

An AI scribe land grab is on as countless U.S. health systems test AI scribes in head-to-head pilots. These organizations prioritize solutions that integrate with their enterprise EHRs, satisfy their various constituencies, and meet high compliance and security requirements.

Relatively few AI scribes can clear this high bar. Suki CEO Punit Soni reports that “as ambient documentation becomes table stakes, users will become more demanding and expect that solutions meet their personal preferences and needs. This will force out some players and drive consolidation among others, leaving a few left standing.”

Will EHR Vendors Decide The AI Scribe Winners?

Basic EHR integrations (usually via FHIR APIs) allow clinicians to complete note-writing within their EHR. But other activities, such as pending orders or updating flowsheets, require deeper EHR integration.

EHR vendors may, therefore, shape outcomes by deciding which vendors gain access to privileged integration points. Epic (DAX Copilot and Abridge are “Workshop” partners), Meditech (Suki, Augmedix, and DAX Copilot), and athenahealth (Suki) have already formed partnerships with different AI scribes. Some are co-developing new capabilities.

Brendan Keeler (aka Health API Guy) believes these partnerships give select AI vendors a head start. However, “Information blocking rules (and for Epic, antitrust rabble-rousing) means any functionality created for one AI scribe should be made available thereafter for others to use.”

Conversely, Canvas CTO Andrew Hines thinks these partnerships confer a long-term advantage, explaining that “Information blocking rules ensure equal access to information (i.e., data reads) but not to write functionality, workflow modification capabilities, or engineering expertise.”

Regardless, while many AI scribes aim to become broader intelligence platforms, EHRs remain the platform and distribution channel upon which these products must sit.

AI scribes are quickly becoming a standard feature in clinical environments. But, as Elion’s Patrick Wingo explained, “It remains to be seen whether new vendors entering the market will be able to offer enough in the way of product features and usability or cost savings to make major headway as a standalone product.” Large EHR vendors like Epic and Oracle are building various AI capabilities.

Penguin AI CEO Fawad Butt adds, “The scribe business is highly vulnerable to competition from all sides, including hyper-scalers (e.g., AWS) and EHR vendors.”

So, AI scribes — or whatever we eventually call them — are here to stay. Over time, they will blend into the background, sometimes becoming part of the EHR. The market will undoubtedly look quite different than it does today.

Acknowledgment: I appreciate the following people who discussed AI scribes with me and helped shape this article: Ruben Amarasingham, Fawad Butt, Adam Carew, Andrew Hines, Brendan Keeler, Alex LeBrun, Yair Saperstein, Punit Soni, and Patrick Wingo.