Abridge Expands Hospital AI for Inpatient and Outpatient Care
On June 18, 2025, Abridge announced the expansion of its generative AI platform into inpatient settings with Abridge Inside for Inpatient, leveraging Epic’s Workshop co-development program¹². The tool integrates into both Haiku and Hyperspace, transforming bedside conversations into structured Epic note drafts in History & Physical, progress, and consult note types¹.
Sutter Health system in the San Francisco Bay Area is among the early pilots. Dr. Veena Jones (CMIO) said it “calms some of the chaos,” while Dr. Richard Liu reported that a quick press of a phone button captures the patient encounter directly in the chart¹. The platform is trained on over 1.5 million clinical encounters and already supports 28+ languages and 50+ specialties across 100+ health systems¹².
In outpatient settings, Abridge also introduced an “Orders” feature that surfaces medications mentioned during visits for rapid placement in Epic. With ~75% of outpatient visits generating prescriptions, this automation helps reduce clinician workload¹.
This dual launch highlights real-world deployment of AI medical documentation, addressing clinician burnout, documentation delays, and fragmented workflows.
Abridge Inside for Inpatient Streamlines Complex Hospital Workflows
Hospital workflows are inherently more complex than outpatient settings, with multiple specialists, trainees, and consultants engaging with the same patient daily. Abridge’s new tool addresses this by enabling efficient, structured documentation right at the point of care⁴.
Using built-in Epic note types and SmartPhrase templates, clinicians can dictate and auto-generate high-quality notes, reducing time spent charting after hours. This new functionality is integrated directly into Epic’s Haiku (mobile) and Hyperspace (desktop) interfaces, allowing for consistency across platforms⁴.
CEO Dr. Shiv Rao explained the intent: “Inpatient care requires more collaboration, so your notes need to reflect handoffs and plans. Abridge captures that nuance without interrupting the clinical flow”⁴.
For Epic-integrated hospitals, this solution offers a plug-and-play advantage: no need to install separate apps or break from current workflows. That, according to the company, is key to achieving clinician adoption at scale¹³.
New Outpatient Orders Feature Turns Conversation into Clinical Action
In parallel with its inpatient launch, Abridge is piloting a new outpatient feature called Orders that turns spoken medication plans into actionable orders in Epic¹².
Family physician Dr. Mary Kirby at Coastal Carolina Health shared her experience: “By the time I finish the visit, the medications I discussed are already surfaced in the EHR, ready to sign. It saves minutes per patient and those add up”⁴.
This automation matters because roughly 75% of outpatient visits generate prescriptions⁴. Traditionally, clinicians had to revisit their notes and re-enter medication details manually. Abridge’s Orders feature eliminates that step, reducing cognitive load and documentation redundancy.
Future updates will extend this functionality to imaging and lab orders, pushing the tool further along the clinical workflow from documentation into direct action.
Epic Integration for Minimal Disruption for Clinicians
Abridge has partnered closely with Epic through its Workshop program to build tools that work “at the top of license” without disrupting clinician workflows¹³.
Both Abridge Inside for Inpatient and the Orders feature are embedded directly within Epic environments. This allows clinicians to chart within their native interface without toggling between apps or relying on clunky dictation tools.
This tight integration is part of a broader strategy to ensure AI tools are invisible yet indispensable. As Dr. Rao puts it: “Our goal is to restore meaning to every clinician-patient conversation”⁴.
Here is the final version of the technical section with properly grouped and numbered references placed at the end. The citations are used once in the body and linked to the sources below, as requested.
How Does Abridge AI Pipeline Work?
Abridge’s system likely uses a layered AI pipeline that begins with audio capture during clinician-patient interactions and results in structured inpatient notes such as progress notes and consults that reflect current context and prior clinical data.
The first stage involves automatic speech recognition (ASR). Abridge uses a proprietary ASR model fine-tuned on medical speech data, capable of speaker diarization and domain-specific transcription. Given the acoustic variability in inpatient settings like shared medical office space, this component must handle overlapping dialogue and poor signal-to-noise ratios. Comparable systems include Amazon Transcribe Medical and open models like OpenAI’s Whisper when adapted for clinical settings²⁻³.
Once transcribed, the system applies clinical natural language processing (NLP) to extract medically relevant entities conditions, medications, symptoms, lab values and categorize them into note sections. Abridge does not disclose its NLP stack, but its functions appear to be similar with those found in frameworks such as Apache cTAKES, MedSpaCy, or MetaMap⁴⁻⁵.
To maintain continuity of care, Abridge uses a retrieval-augmented generation (RAG) approach. It semantically retrieves relevant content from previous notes yesterday’s assessment, current lab trends using vector-based similarity likely via a transformer-derived sentence encoder and retrieval engine such as FAISS or Elasticsearch. This retrieved context is then passed along with the current transcript into a language generation model, producing a draft note that reflects updates without repeating unchanged content⁶⁻⁷.
The output is formatted to match Epic’s structured documentation, often using SmartPhrases or sectioned templates. Each generated segment is linked back to the corresponding transcript or audio, allowing clinicians to review and edit directly within Epic before finalizing the note¹.
This approach has clear operational benefits: it reduces duplication, supports handoffs between care teams, and improves consistency across multi-day hospital stays. However, like all current-generation medical AI systems, accuracy depends heavily on audio quality and domain generalization. Errors in transcription or entity mapping can introduce inaccuracies which reinforces the need for human oversight.
Valuation Scaling Fast with Funding and Institutional Support
Abridge has raised $250 million in Series D funding as of February 2025, bringing its valuation to $2.75 billion¹⁰. It is now deployed in 100+ health systems, including industry leaders like Mayo Clinic, Johns Hopkins, Duke Health, and Memorial Sloan Kettering⁷.
The company has also gained industry recognition, landing on TIME’s Best Inventions of 2024, Forbes AI 50, and earning a Best in KLAS designation in 2025¹²⁷.
With a proven record in emergency and outpatient settings, Abridge’s entry into inpatient workflows signifies a major expansion and a bet that ambient AI will become a foundational in EHR infrastructure.
Health Systems Report Early Gains in Efficiency and Clinician Wellbeing
Early pilot data from Abridge’s existing deployments are promising. At Riverside Health, documentation-related cognitive load dropped 61%, and clinician-reported burnout decreased by 55% after full implementation¹⁰.
These findings are consistent with the company’s product goals:
- Reduce time spent on after-hours charting
- Improve documentation quality and billing completeness
- Support patient safety with structured, high-context notes
Abridge is positioning its platform as a mid-revenue cycle tool: helping not just with documentation, but with coding, compliance, and downstream revenue integrity¹.
Full Scope AI Roadmap: Labs, Imaging, and Broader Medical Note Types
Looking ahead, Abridge plans to expand its Orders feature beyond medications to include lab orders, imaging studies, and follow-up instructions¹². This aligns with broader trends in AI clinical automation, where systems are expected to generate not just documentation, but care recommendations and next steps based on conversation.
In the hospital setting, Abridge will also continue to enhance its support for daily progress notes by pulling forward relevant history from prior notes minimizing repetition and enabling true continuity of care.
Future Epic integrations may also include smart summarization, decision support, and interoperability across hospital departments all powered by Abridge’s growing dataset of real-world clinical interactions.
At the Center of AI Medical Documentation
With the simultaneous rollout of Abridge Inside for Inpatient and its outpatient Orders feature, Abridge is redefining what AI can do in healthcare documentation. The use of AI documentation in healthcare goes beyond transcription or dictation towards transforming the way clinicians interact with the EHR itself.
For health systems, this means real potential to save time, reduce costs, and improve patient care. For clinicians, it could mean fewer after-hours charting sessions and more face time with patients.
References
- Business Wire – Abridge Announces Inpatient and Outpatient Orders Launch (June 18, 2025)
- Abridge Official Blog – Abridge Inside for Inpatient
- Fierce Healthcare – Abridge Launches Inpatient Tool
- Fierce Healthcare – Interview with Shiv Rao
- Abridge Emergency Tool Announcement (Jan 29, 2025)
- Abridge Series D Press Release (Feb 17, 2025)
- Abridge Health Systems & Awards
- TIME – Best Inventions 2024
- Forbes – AI 50 2025 List
- Riverside Health Pilot Outcomes (internal summary via Business Wire)
- Abridge | “Inside for Inpatient” Epic integration details – https://www.abridge.com/blog/inpatient-launch
- OpenAI Whisper ASR model – https://openai.com/research/whisper
- Amazon Transcribe Medical – https://aws.amazon.com/transcribe/medical/
- Apache cTAKES (clinical text analysis and knowledge extraction) – https://ctakes.apache.org
- MedSpaCy: Clinical NLP with spaCy – https://github.com/medspacy/medspacy
- Retrieval-Augmented Generation for Clinical Notes – See: “Towards Retrieval-Augmented Clinical Note Generation” (arXiv) – https://arxiv.org/abs/2302.06199
- Facebook AI Similarity Search (FAISS) – https://github.com/facebookresearch/faiss