Once Epic shipped AI charting of its own, the ambient seat could no longer be sold at the old price: Ambience put its fees behind a 3x ROI and a burnout rate falling from 45% to 31%, Abridge dug down into the revenue cycle and the nursing station — while the one peer-reviewed study across five academic systems and 8,500 clinicians puts the gain at $167 per clinician per month, about a sixth of what the vendor case studies claim

On February 5, 2026, Epic released AI Charting inside Art, its clinician-facing platform. It listens through the visit, drafts the note and queues orders from the conversation, and clinicians can reformat by voice — "make the current conditions a bulleted list" — with diagnosis-aware notes tying assessment and plan to specific diagnoses following in March. The chart-summary feature inside Art already passes 16 million uses a month, roughly triple its level the previous November. Epic's three agents are Art for clinicians, Penny for revenue cycle and Emmie for patients; Penny is in use at more than 200 organisations and is credited with a 20% reduction in denials.

Epic holds more than 42% of the U.S. acute-care hospital market and close to 55% of acute-care beds, on KLAS's 2024 figures. When "patient audio never leaves the building" is itself a security argument and the price is lower too, an independent scribe vendor has only two moves left: go deeper than Epic will bother to go — specialty, revenue cycle, nursing — or tie the price to the outcome. Both routes have now been taken, in the second half of this year.

Epic has published neither pricing nor an adoption count for AI Charting, and Penny's "20% reduction in denials" is Epic's own figure with no independent audit. The 16 million is a count of uses, not of distinct users, and cannot be converted into a clinician adoption rate.

Healthcare Dive, 2026-02-05 · HIT Consultant, 2026-02-05

On August 19 Ambience published "The Ambience Standard," tying platform fees directly to verified clinical, operational and financial outcomes instead of per-seat or per-token billing — which moves the financial risk back from the health system to the vendor. It has three parts: value-at-risk performance contracts whose fees depend on jointly agreed metrics (documentation time, coding accuracy, margin improvement); forward-deployed teams of clinicians, engineers and transformation specialists embedded inside the health system; and Chorus, an agent-orchestration harness covering more than 200 medical and surgical subspecialties. The company published verified numbers from five customers alongside it: 3x ROI at Ardent Health; 70% sustained utilisation across 4,000 clinicians at Cleveland Clinic with an NPS of 60; $24,000 net annual value per physician at Onvida Health; 92% adoption at MultiCare with an NPS 63 points above competitors; and physician burnout falling from 45% to 31% at St. Luke's, with turnover intent down from 31% to 18%.

Co-founder Mike Ng put it plainly: "There's a million miles between a better model and better care. Models don't improve healthcare on their own." This is the first time a leading ambient AI vendor has said out loud that model capability is no longer the differentiator and delivery capability is. It also runs exactly parallel to where the payer is going — CMS's ACCESS model likewise pays on outcomes rather than on services rendered.

All five customer figures are disclosed by Ambience itself; who verified them, by what method and on what sample is not published. Nor is the share of fee actually at risk — 5% or 50% — and without that number, "the risk moves back to the vendor" is only a sentence. The St. Luke's burnout change comes with no control group.

HIT Consultant, 2026-08-19 · Health IT Answers, 2026-09-16

Abridge's revenue-cycle product turns the encounter itself into audit-ready, billable documentation at the point of care rather than retrospectively, in three pieces: inpatient CDI, surfacing documentation specificity in-line during care instead of after discharge; Ambulatory Encounter Integrity, generating ICD-10 and HCC codes in real time, auto-calculating E&M levels and updating the record before the note is signed; and Care Signals, flagging risk gaps before the visit and tracking which conditions were actually discussed. The product page cites Sutter Health, where 78% of clinicians report improved work satisfaction. Nursing is the second line: on May 6 Abridge made the nursing AI platform it built with Mayo Clinic generally available across 250-plus health system partners. Mayo patients opt into ambient recording at rates in the high eighties, and the highest-adopting nurses at Corewell Health save close to 30 minutes of documentation a shift.

Neither line is about a better note; both are about a new budget line. Time saved on charting is hard to locate on a financial statement, whereas CDI and coding accuracy hit revenue directly and nurse turnover hits labour cost directly — which moves the buying decision off the CMIO's desk and onto the CFO's and the chief nursing officer's. Abridge's own framing in Newsweek names the real obstacle on the nursing side: "a lot of nurses have lost trust in tech… trust is earned in drops and lost in buckets."

Sutter's 78% and Corewell's 30 minutes are vendor- or partner-reported, not peer-reviewed. "250-plus health system partners" describes availability, not deployment: in the same report Mayo has seven units fully deployed and 170 nurses activated enterprise-wide, and only two of Corewell's 21 hospitals are using it. Reading "available" as "live" overstates the reality by an order of magnitude.

Newsweek, 2026-05-06 · Abridge Platform — Revenue Cycle

The five-site study led by Lisa Rotenstein at UCSF and published in JAMA covers more than 8,500 clinicians at Mass General Brigham, Emory Healthcare, UCSF, UC Davis and Yale New Haven, roughly 1,800 of whom actually adopted an AI scribe. The result: 13 fewer minutes of EHR time and 16 fewer minutes of documentation per eight-hour shift, weekly visit volume up 1.7%, and about $167 per clinician per month in additional evaluation-and-management billing; high-frequency users above 50% adoption saw twice the EHR-time reduction and three times the documentation reduction. Set that against the vendor case studies: the Suki data KLAS validated in January shows $1,004 per provider per month at McLeod Health and $862 per user at FMOL Health, with McLeod's Level 3 codes down 18.2%, Level 4 and 5 up 7.3% and 5.0%, and encounters up 18.3%.

A sixfold gap need not mean anyone is lying; more likely two things are being measured. Vendor case studies tend to come from the most committed adopters, organisations that deliberately rebooked schedules to fill the time freed up, whereas none of the five JAMA sites asked clinicians to see more patients — the authors call $167 a conservative lower bound. But that is exactly the stress test for Ambience's value-at-risk contracts: sign one, and what is at risk is not whether the model is good but whether the organisation converts saved time into money. It is also why the coding shift — Level 3 down, Level 4 and 5 up — needs independent audit, because the same curve can be called documentation finally reflecting true complexity, or upcoding.

Health Exec, 2026-04 · HIT Consultant, 2026-01-21 · Fierce Healthcare

On September 16 Oracle Health launched, in the U.S., a nursing clinical AI agent embedded in the Oracle Health Foundation EHR, combining three things: voice-driven navigation, acute nursing summaries and voice-enabled charting. Nurses can search the record in natural language, pull an AI summary of history and current condition, and complete structured documentation in near real time at the bedside. Oracle says its clinical note AI tools have saved physicians more than 400,000 hours across U.S. health organisations in the nearly two years since launch. BayCare Health System is the early adopter; its CIO Lynnette Clinton says it has the potential to reduce after-shift charting. Oracle bought Cerner for $28 billion in June 2022 and rebranded it Oracle Health; more healthcare AI is due at its Health and Life Sciences Summit in Orlando next week.

The point here is not the feature set but the unit of account. Both major EHR vendors now build ambient AI themselves, and both still score it in hours saved — the very yardstick Ambience says it is walking away from. A built-in EHR feature can be scored in hours because it was never a separately priced line item; an independent vendor cannot, because no CFO signs a purchase order for hours. That dividing line is the spine of today's report.

The 400,000 hours is Oracle's own figure, with no denominator given — how many organisations, how many physicians, how many notes — and no third-party verification. BayCare is the only named adopter so far, with no figure for how much of it is actually deployed.

Fierce Healthcare, 2026-09-16

On September 16 CMS extended its ACCESS model to heart failure, COPD and other chronic conditions, with officials saying they are weighing an extension into Medicaid. A day later the smart-ring maker Oura and the AI-native primary care company Counsel Health said they would enter the model's early cardio-kidney-metabolic track, reporting that 36,000 Oura members completed onboarding in the first nine weeks, with 70% satisfaction on onboarding, a 4.6-out-of-5 care experience rating, and 15% of members receiving symptom alerts — over 10,000 people — connecting to care. ACCESS is a ten-year federal payment initiative starting in early 2027 that pays on patient outcomes rather than services; for full payment in 2026–2027 at least 50% of aligned beneficiaries must hit every required outcome target, and the rates published in March came in below industry expectations. Oura has 5 million paid members.

Read this beside the Ambience item and the same turn is visible at both ends of the supply chain: the payer wants outcomes, and the vendor has started selling outcomes. The difference is that CMS has written its threshold down — 50% of beneficiaries hitting every target — while Ambience's share-at-risk is still undisclosed. Whoever commits to a number first sets the negotiating baseline.

The 36,000 is people onboarded, not outcomes improved; ACCESS does not formally start until early 2027, so every figure so far is a pre-period engagement metric and none of the health outcomes that payment will hinge on yet exist. Oura and Counsel Health's numbers are self-reported.

Fierce Healthcare, 2026-09-17 · Fierce Healthcare, 2026-09-16

On September 17 Sentara Health said it would pilot GW RhythmX's precision care AI platform, pushing AI past administrative automation into decision support inside the visit itself: the platform combines history, guidelines, care gaps, social determinants and payer requirements into personalised recommendations at the point of care. Phase one runs 90 days across 30 physicians and advanced practice providers in family and internal medicine in two markets. Sentara runs 12 hospitals and over 400 care sites with roughly 473 primary care clinicians. The reference case is Presbyterian Healthcare Services: 17,000 hyper-personalised treatment responses for 1,100 patients in six months, $1.1 million of condition capture validated through adjudicated claims, and expansion to 200 primary care physicians after the pilot. GW RhythmX says its platform spans more than 150 health systems and 85 million patients.

Note the shape of the Presbyterian number: $1.1 million of condition capture — that is, risk-adjustment coding value, validated through adjudicated claims. It points at the same pocket as Abridge's Encounter Integrity and Epic's Penny. In 2026 the value of clinical AI that a finance department most readily understands is still getting care that already happened correctly onto the bill. That is legitimate in itself — under-documented conditions should be recorded — but it is also the industry's largest open regulatory exposure right now.

A 30-clinician, 90-day, two-market pilot has produced no results yet. Presbyterian's $1.1 million and GW RhythmX's 85-million-patient reach are both vendor-reported.

Fierce Healthcare, 2026-09-17

On January 21 OpenEvidence closed a $250 million Series D at a doubled $12 billion valuation, bringing three rounds in a year to roughly $700 million, on annual revenue that has passed $100 million. In December 2025 alone it served about 18 million clinical consultations, against 3 million a month a year earlier; over 40% of U.S. physicians use it daily, across more than 10,000 hospitals and medical centres. It is free to doctors and funded by advertising. CEO Daniel Nadler credits a copyright-friendly approach and early partnerships with the AMA and NEJM. Surveys put OpenEvidence as the most-adopted AI tool among physicians at 45% — more than nine competing platforms combined.

This is the comparison worth holding alongside everything above. Ambient AI sells to a hospital CFO and must justify its price with ROI; OpenEvidence skips procurement entirely and sells to pharmaceutical advertisers, with physician attention as the product. Both models proved viable in 2026, but the moats are nothing alike: one is built on delivery and workflow embedding, the other on physician mindshare and content licensing. Microsoft's Dragon Copilot already carries OpenEvidence as a third-party content partner, so the line will only blur further.

The "40% of U.S. physicians daily" and the 45% adoption figure come from company-supplied or company-commissioned surveys with no independent audit, and the 18 million is consultations, not distinct users. An advertising-funded clinical decision tool carries a structural conflict of interest for which no regulatory framework yet exists.

Fierce Healthcare, 2026-01-21 · Sacra — OpenEvidence