Abridge just hit 100M doctor visits and cut prior auth from days to minutes — here's what production healthcare AI actually looks like
Abridge processed 100M patient visits, saves clinicians 10-20 hours per week, and turned prior authorization from a 3-day ordeal into minutes. Real numbers from a real deployment.
Abridge just crossed 100 million patient visits processed. That's not a marketing number — it's live clinical encounters where their ambient documentation system listened to the conversation, generated structured notes, and fed them directly into EHR workflows.
The company's co-founders Janie Lee and Chai Asawa walked through the numbers on Latent Space this week. Clinicians using Abridge are saving 10-20 hours per week on documentation. Prior authorization requests that used to take 3 days now complete in minutes. The system is running in hundreds of health systems across the US.
This is what production healthcare AI looks like when it actually ships.
The prior auth problem
Prior authorization is the insurance approval dance that happens before a patient can get a procedure, imaging study, or specialty drug. A clinician submits a request with supporting documentation. The insurer reviews it. Days pass. The request gets approved, denied, or bounced back for more information.
Abridge built an agent that pulls the relevant clinical notes, lab results, and imaging reports from the EHR, structures them according to the insurer's specific requirements, and submits the request. What used to take a medical assistant 45 minutes of chart review and form-filling now takes the agent under 5 minutes. The insurer gets a complete packet. The approval comes back same-day or next-day instead of 3-5 days out.
That's not a demo. It's in production at multiple health systems.
Ambient documentation at scale
The core product is still ambient clinical documentation. A clinician wears a badge or uses a phone app. The system records the patient encounter. It generates a structured SOAP note (Subjective, Objective, Assessment, Plan) and pushes it into Epic, Cerner, or whatever EHR the practice runs.
The 100M visit milestone matters because healthcare AI has a credibility problem. Lots of vendors announce pilots. Fewer ship to dozens of sites. Almost none hit 8-figure encounter volume. Abridge did.
Clinicians report 10-20 hours per week saved on documentation. That's 2-4 hours per day. It means they finish their notes during the workday instead of spending evenings on pajama time charting. It means they see one or two more patients per day without extending hours. For a practice with 10 clinicians, that's 100-200 hours per week of recovered capacity.
What makes this different from Whisper + GPT-4o + a prompt
The technical stack isn't public, but the interview revealed a few details. Abridge runs custom speech models trained on medical conversations. They fine-tune language models on clinical note structures. They handle multi-speaker diarization, background noise, interruptions, and the specific rhythm of a patient visit.
More importantly, they handle EHR integration. The note doesn't land in a generic text field — it populates the structured fields Epic or Cerner expects. Billing codes get suggested. Orders get drafted. The note becomes part of the workflow, not a thing the clinician has to manually transfer.
That's the production gap. A demo that generates a pretty note in a web UI is 20% of the work. The other 80% is EHR integration, compliance, error handling, and making the output something a clinician will actually sign.
The prior auth agent is the next layer
Once you have structured clinical notes in the EHR, you can build agents that act on them. Prior authorization is the obvious first target because it's pure administrative overhead. No clinical judgment required — just chart review, form-filling, and submission.
Abridge's prior auth agent reads the clinical notes their documentation system generated, pulls supporting data from the EHR, fills out the insurer's forms, and submits. The clinician reviews the packet before it goes out. The whole loop takes minutes instead of days.
That's the agentic healthcare stack in practice. Ambient documentation generates structured data. Agents consume that data and automate administrative workflows. Clinicians review and approve. The system handles the grunt work.
What this means for healthcare AI
Healthcare AI has been stuck in pilot purgatory for years. Vendors announce partnerships. Health systems run 6-month trials. Nothing scales. Abridge scaled.
The 100M visit milestone and the 10-20 hour time savings are real. The prior auth agent is real. The EHR integrations are real. This is what it looks like when a healthcare AI company actually ships.
If you're building in this space, the bar is now clear. Ambient documentation that integrates with EHRs and saves clinicians double-digit hours per week. Agents that automate administrative workflows end-to-end. Production scale measured in millions of encounters, not dozens of pilot sites.
That's the benchmark. Anything less is still a demo.