Abridge

Abridge builds healthcare AI that turns patient-clinician conversations into clinical documentation and combines that information with patient records to support clinical and administrative work. Founded in 2018, it expanded from a patient conversation app into enterprise documentation, then into clinical decision support and revenue cycle workflows.[8][2][3][10][15]

A consequential expansion came in August 2026, when Abridge made its clinical intelligence agent available to clinicians at partner organizations who did not use its ambient documentation product. Access could run through EHR workflows or chart-connected mobile and web experiences. Capabilities included pre-visit summaries, chart queries, calculators populated from the EHR, cited medical evidence, and clinician-directed referral letters or handoffs. Abridge reported adoption of its context-aware clinical decision support by more than 300 enterprise health systems since its April launch; the systems’ represented patient populations were not counts of individual Abridge users.[15]

These clinician-facing products followed earlier work helping patients understand and revisit medical conversations. The patient app and the subsequent enterprise launch establish how Abridge’s use of conversation data broadened over time.[2][3]

Abridge officially launched its patient app on October 6, 2020. Patients and loved ones could record health conversations, receive selective transcripts of medical portions, replay audio bookmarks, look up medical definitions, and identify medication instructions or follow-up appointments. The company reported helping 50,000 people before launch. Its machine learning foundation involved staff led by CTO and co-founder Sandeep Konam, alongside Carnegie Mellon University professors and their laboratories.[2]

On August 10, 2022, Shiv Rao announced Abridge’s enterprise solution for physician documentation. It structured and summarized doctor-patient conversations in real time, mapped medical information to ontologies and codes, and connected extracted insights to their originating conversation passages. Abridge also made its core machine learning available through an API for telemedicine and call center integration. Patient-facing interactive transcripts, definitions, and next-step reminders remained part of the offering.[3]

At the enterprise launch, Abridge reported a proprietary dataset derived from more than 1.5 million medical encounters, more than 2,000 clinicians helped, and 200,000 patients helped. These described separate historical measures of data and reach, rather than paid customers or current active users.[3]

Abridge’s documentation pipeline separates clinical speech recognition from note generation: speech recognition produces a transcript, and generation turns that transcript into a draft. Clinicians verify and edit the note before signing it in the EHR. The company’s 2023 Series B announcement described deeper integration of structured summaries into Epic workflows through Epic’s Partners and Pals program.[11][4]

Linked Evidence gives clinicians a concrete way to inspect generated documentation. Highlighting note text reveals the corresponding transcript passage; double-clicking the transcript or selecting Play replays the original audio. This connects a generated statement to the conversation used to produce it.[7]

Announced in February 2025, the Contextual Reasoning Engine extended the inputs beyond the current conversation. Its product description includes previous patient encounters, health system guidelines, and clinician preferences. It also describes grouping medical problems in language aligned with billing codes and producing EHR-integrated outputs, such as orders, for clinician review.[8][9]

In July 2024, Abridge, Mayo Clinic, and Epic announced collaborative development of documentation for nurses. Abridge supplied generative AI, Epic supplied development and integration, and Mayo supplied nursing practice expertise, with nurses helping prioritize workflows, design, and test the solution. The work addressed continuous nursing activity across multiple patients and tasks, a different documentation setting from a doctor’s visit. The announcement described development rather than a completed rollout.[6]

By June 2025, Abridge was expanding into revenue cycle workflows, seeking to capture documentation and billing requirements during the clinical conversation. In January 2026, it announced a collaboration with Availity to develop real-time prior authorization, combining Abridge’s conversation-based clinical context with Availity’s FHIR-native Intelligent Utilization Management solution. The announcement described a development initiative and did not report measured authorization turnaround times or completed authorization volumes.[10][12]

Abridge describes model evaluation that combines blinded head-to-head comparisons by licensed clinicians, automated metrics, clinician spot-checks, anytime-valid sequential hypothesis testing, and staged releases. After deployment, it monitors edits, ratings, and qualitative feedback. Its technical account identifies practical blind spots, including newly released medication names and distinguishing a speaking parent from an infant patient.[11]

In an August 2026 engineering article, Catherine Chen, Samir Khan, Michael Oberst, and Alex Chouldechova described encounter-specific evaluation rules written by clinicians from de-identified encounters and checked against generated notes by LLMs. The approach uses expert knowledge to build scalable assessment criteria. Evaluation also checks for regressions outside the dimension a change was intended to improve.[14]

For clinical decision support, Abridge described clinical content rubrics, adversarial boundary tests, and safety evaluations with clinician review of flagged cases. On March 31, 2026, it reported an internal safety evaluation score of 99.5% across more than 1,000 cases. That figure describes its test set, rather than diagnostic accuracy or a real-world patient safety rate. Clinicians choose whether to act on or incorporate the system’s responses.[13]

A UW Health randomized trial published in NEJM AI studied documentation-only Abridge over 24 weeks with 66 practitioners across eight ambulatory specialties. It found reduced work exhaustion/interpersonal disengagement and a reduction in note time of 0.36 hours per eight hours of patient time; professional fulfillment did not significantly improve. The work-outside-work result lost statistical significance when the highest 3% of daily observations were excluded. UW Health commercially procured the licenses and received no funding from Abridge. The findings concern that historical documentation product and study setting, not the later decision-support system.[16]

At its October 2020 launch, Abridge announced $15 million across Seed and Series A financing, with Union Square Ventures and UPMC named as leads. The August 2022 enterprise announcement disclosed a further $12.5 million from investors including Wittington Ventures, Whistler Capital, and Yoshua Bengio. Abridge then announced a $30 million Series B led by Spark Capital in October 2023 and a $150 million Series C led by Lightspeed, with Redpoint as co-lead, in February 2024. The Series C announcement also included an enterprise agreement giving thousands of Yale New Haven Health clinicians access to the product.[2][3][4][5]

On February 17, 2025, Abridge announced a $250 million Series D co-led by Elad Gil and IVP and reported surpassing 100 health system deployments. On June 23, it announced the successful close of a $300 million Series E led by Andreessen Horowitz, with Khosla Ventures participating, and reported partnerships with more than 150 enterprise health systems. Its projection of more than 50 million medical conversations during 2025 was a forecast.[8][10]

In January 2026, Abridge reported more than 200 health systems and projected more than 80 million patient-clinician conversations that year. By August, it reported clinical decision-support adoption across more than 300 enterprise health systems. These measures describe different scopes—organizational reach, projected conversation volume, and adoption of a particular product—and do not supply an absolute active clinician count.[12][15]