The ambient AI scribe market raised roughly $733 million across five companies in 2025 and 2026. Suki's $90 million Series E, announced October 13, 2026, was the most recent round. Each company in the group started by replacing the medical scribe, and each is now telling health systems that documentation was only the beginning.
The expansion pitch follows the same logic everywhere. A note is a recurring subscription tied to a clinician seat. Seat pricing gets squeezed when competitors multiply, and right now they are multiplying. Orders, referrals, coding, and prior authorization connect to money moving through the revenue cycle. Vendors want to price on outcomes instead of seats.
That pitch may be correct as a business strategy. It tells you nothing about what the technology does in a clinical setting. The evidence base for scribes is thin and still forming. The evidence base for agents acting in systems of record barely exists. If you run a health system, an FQHC, a radiology group, or a benefits team that pays for this category of software, the studies available today are the right place to start.
1. What Suki Is Building and Why It Goes Beyond Documentation
On October 13, 2026, Healthcare Dive reported that Suki raised a $90 million Series E to develop AI agents that operate after the visit ends. Those agents are described as placing orders, scheduling follow-ups, handling referrals, and processing prior authorizations.
The round brought total Suki funding to more than $250 million across five rounds, according to Healthcare Dive. New investors include Healthier Capital, Crosslink Capital, and Hercules Capital. One outlet credited Healthier Capital with leading the round; Healthcare Dive did not name a lead. Treat the lead attribution as unsettled.
Suki reports serving more than 450 healthcare organizations and working with more than 30 technology partners. Healthcare Dive named Optum, Zoom, and athenahealth among those partners. Athenahealth's Ambient Notes feature is powered by Suki, as is the ambient documentation feature on Zoom's telehealth platform. Revenue doubled in 2025 and is tracking for similar growth in 2026, per company statements. One outlet put 2026 revenue at $60 million to $80 million, citing an unnamed source. That number is unconfirmed.
The strategic statement from CEO Punit Soni, quoted in Pulse 2.0, is worth reading carefully. He said documentation was the entry point and that healthcare needs an intelligent operating system rather than a collection of disconnected AI features. That framing is a business case, not a capability claim. It tells you where Suki wants to go. It does not tell you whether the route is proven.
A scribe drafts text that a clinician reads and then signs. An agent takes an action in a system of record. The review step shrinks, because the value of an agent is that the human does not have to handle each task individually. That difference changes what a reasonable evidence bar looks like, and it changes what risk looks like when the system makes an error.
2. Where the $733 Million Went: A Crowded Expansion Pattern
Suki is the latest in a pattern of ambient AI companies raising large sums and then announcing plans to move upstream into revenue cycle, coding, and administrative work.
The funding rounds shown span 2025 and 2026. Healthcare Innovation reported the 2025 rounds: Ambience Healthcare's $243 million Series C co-led by Oak HC/FT and Andreessen Horowitz, Abridge's $300 million Series E led by Andreessen Horowitz, Nabla's $70 million Series C in June 2025, and Freed's $30 million Series A led by Sequoia. Healthcare Dive reported Suki's $90 million round. These rounds happened at different times and stages, so the total of roughly $733 million (our arithmetic across the five) shows scale, not a precise market total.
The expansion announcements are consistent. Abridge told Healthcare Innovation it was moving upstream to automate back-office work and support revenue cycle teams. Fierce Healthcare, writing before Epic's February 2026 launch, cited a Stat analysis reporting that investors had put nearly $1 billion into ambient AI companies in 2025 up to that point.
When every well-funded vendor makes the same move at the same time, that is a unit economics signal, not a product differentiator. Scribes are commoditizing. The next layer is where pricing pressure has not yet arrived. For buyers, this matters because vendor incentives have shifted. A company optimizing for seat revenue behaves differently than one optimizing for outcome-based contracts on prior authorization approval rates. Know what your vendor is optimizing for. That understanding should drive how you write the contract.
3. The JAMA Study: Modest Savings Concentrated Among Heavy Users
The strongest real-world evidence on scribes published to date is a multisite study by Lisa Rotenstein and colleagues, co-led by UCSF and Mass General Brigham. The Mass General Brigham newsroom described it as covering five US hospitals over more than two years, comparing more than 1,800 clinicians using AI scribes with 6,770 control clinicians. It was published in JAMA.
The headline results are real and also modest. Daily EHR time fell by 13 minutes, a 3% relative decrease. Documentation time fell by 16 minutes, a 10% relative decrease. Scribe users completed an average of 0.5 additional patient visits per week. There was no significant difference in EHR time outside working hours. The Mass General Brigham release estimated revenue impact at $167 per month per clinician who adopted the tool, calling it statistically significant but nominal.
The adoption pattern matters more than any single average. Only 32% of licensed users relied on the scribe for more than half of their visits. Clinicians in that group saw roughly twice the reduction in total EHR time and roughly three times the reduction in documentation time. The benefit is concentrated. Most licensed users are not heavy users.
Rebecca Mishuris, chief health information officer at Mass General Brigham and a senior author on the study, said the modest time reductions are unlikely to fully account for changes in burnout. That is an important sentence. Earlier research links ambient documentation to lower burnout rates, and this study suggests something beyond minutes saved is driving that effect. A quality-improvement study reported by Textbook of Digital Health found burnout fell from 51.9% to 38.8% after 30 days among 263 clinicians in six health systems. I read that via a secondary source and have not confirmed the methodology against the primary paper.
The Peterson Health Technology Institute reached a similarly mixed picture in a 2025 analysis. As reported by Becker's Hospital Review, PHTI found roughly 60 scribe solutions in use across the market, some health systems reporting up to a 40% reduction in burnout among active users, mixed time savings, and no definitive proof that scribes improve financial outcomes or justify their cost at scale.
For contract structure, the implication is direct. License count is a poor proxy for value. Active use by clinician is the metric that predicts savings. Your agreement should require reporting of active use as a condition of renewal, not as an optional output.
4. The Randomized Trial: Vendor Differences Are Real and Measurable
If the JAMA study asks whether scribes work on average, a randomized controlled trial published in NEJM AI asks whether two specific products work equally well in the same setting. The answer is that they do not.
The trial enrolled 238 outpatient physicians across 14 specialties. Physicians were randomized to Microsoft's Dragon Ambient eXperience Copilot, to Nabla, or to usual care. Enrollment ran from November 4, 2024 to January 3, 2025. DAX was used in 33.5% of 24,696 visits and Nabla in 29.5% of 23,653 visits.
Time in note fell 9.5% with Nabla compared with usual care (95% confidence interval, 1.8% to 17.2%; P = 0.02). With DAX, the change was a 1.7% reduction, which was not statistically significant. Both products improved scores on the Mini-Z 2.0 burnout instrument: plus 2.83 for DAX and plus 2.69 for Nabla, both with confidence intervals excluding zero. Physicians rated both products as producing inaccuracies occasionally, scoring 2.7 for DAX and 2.8 for Nabla on a five-point scale.
Two limitations apply. I read this trial through its PubMed abstract, not the full paper, and one burnout measure in the abstract appeared internally inconsistent. The authors noted that burnout and task-load gains need confirmation in larger, multicenter trials. Neither result should be treated as final.
The practical implication is still clear. Both products improved burnout and cognitive load measures. Only one produced a measurable reduction in documentation time. If you assume that a scribe is a scribe because both carry the ambient AI label, you may pay for savings that your own data does not support. The only way to know which result your organization gets is a controlled pilot in your specialties, against your own baseline.
Suki was not in this trial. That is not a criticism of Suki. It is a statement that no peer-reviewed controlled comparison has placed Suki against another product in a randomized design. And no trial at all has evaluated the agent capabilities that Suki is now raising money to build. The evidence base for the next product layer does not yet exist.
5. Epic's Native Tool and What It Means for the Buying Decision
The largest competitive threat to every independent ambient AI vendor is not another well-funded startup. It is the EHR vendor.
On February 5, 2026, MedCity News reported that Epic rolled out an AI charting tool that listens during appointments, drafts clinical documentation, and prepares orders. HLTH described it as part of an Epic AI suite branded Art for clinicians, Penny for revenue cycle, and Emmie for patients, and reported that it draws on the patient's full longitudinal record. KLAS data cited by MedCity puts Epic's share of the acute care EHR market at 42%. Oracle Health is next at 23%. The year behind those numbers was not stated in the source.
Abridge took the top ambient spot in KLAS's 2026 Best in KLAS awards, per MedCity, and the report noted it was unclear how much that distinction would matter once Epic's tool was embedded in the workflow. HLTH reported that health system leaders were already reassessing separate contracts with companies such as Abridge, Ambience, and Nabla, and that startups would need to demonstrate capabilities that significantly exceed Epic's offering.
Pricing is unconfirmed. Before launch, Fierce Healthcare relayed a LinkedIn post from a health tech founder who estimated Epic's tool would cost roughly $80 per provider per month. Epic has not confirmed a price in any source I read. Treat that figure as a rumor and obtain a direct quote from Epic for your specific contract terms.
The vendor responses reflect the situation. DeepScribe CEO Matthew Ko told MedCity that many Epic customers run his product alongside Epic. Heidi CEO Tom Kelly said demand exists for platforms that sit alongside the EHR and reduce clinician burden more broadly. Both statements come from sellers with a stake in the outcome. The buyer question they point toward is still the right one: what does the specialized tool do, measured on your outcomes, in your specialties, that the native tool cannot do at its price?
For any organization already on Epic, the native tool gives you a baseline for every negotiation. Use it in one group for 90 days and measure the result. That result is the floor every independent vendor must beat, in your data, before you extend or expand a contract.
6. Agents Raise the Stakes. The Evidence Gap Is Large.
The agent layer is where the evidence gap is largest and where the stakes for error are highest.
A scribe error is usually caught before it causes downstream harm. A clinician reads the draft note and signs it. An agent error can send an incorrect referral, queue the wrong medication order, or submit a prior authorization with the wrong clinical justification. The human review step shrinks, because the value of an agent is precisely that nobody has to handle each task by hand.
I found no peer-reviewed study of agent performance on referrals or prior authorization in the sources I read. The one outcome figure I found in the market is a secondhand report from HLTH that more than 200 organizations use Epic's Penny coding tool, with a reported 20% reduction in claim denials. That describes a different product, and HLTH attributed the figure without naming a primary source. I have not confirmed it.
Prior authorization deserves the most scrutiny before you give an agent access to it. The payer decides on what it receives. If an agent builds a prior authorization request from an ambient note that contains an inaccuracy, that error travels into a coverage decision and can surface months later as a denial or a payer audit finding. A sensible pilot should measure first-pass approval rate and the reasons for denials, not only the hours saved on the submission task.
Order entry belongs in a different risk tier than appointment reminders. Before activating any agent capability, define who reviews its output, what error rate triggers a halt, and which actions the agent may never take without an explicit human click. Those parameters belong in the contract, not in a vendor FAQ.
For FQHC leaders, the staffing risk compounds the clinical risk. A denial on a prior authorization at a busy FQHC means manual rework that pulls from a team already operating thin. The cost of an agent error is not only the time spent correcting it. It is the delay in care and the staff capacity consumed fixing a machine output instead of handling the next patient.
Before you sign a contract that includes agent capabilities, ask for answers to six questions. Ask for active-use reporting by clinician, not licenses purchased. Ask for a controlled pilot result from your own organization before you scale. Ask what the vendor measures for accuracy and who catches errors in the current workflow. Ask what the product does better than the native EHR tool on your specific metrics, if you run Epic or another major EHR. Ask for outcome-linked terms on any agent capability, such as prior authorization approval rates or order accuracy rates. Ask where your patient data goes, and who can use it.
What This Means for Lung Cancer Screening and Underserved Communities
Health systems and FQHCs share a constraint that venture-funded vendors do not feel: every clinician minute is scarce and every operating margin point is thin. A scribe that saves 16 minutes a day for the 32% of users who adopt it heavily is worth real money over a year. A scribe that most licensed users touch for fewer than half their visits is not.
The lesson the scribe data teaches applies directly to any AI tool purchased for a screening or diagnostic program. Software does not change a workflow by being installed. It changes a workflow when clinicians use it often enough to produce an effect, and when someone in the organization measures the result against a baseline.
At Oatmeal Health, we build LungAI for lung cancer diagnostics from CT scans, focused on finding cancers earlier in underserved communities. LungAI is investigational and not FDA cleared. We apply the same standard described in this article: pilot in a real setting, report active use, and publish what the data shows, including results that are modest.
Suki's $90 million round is a useful moment to ask harder questions of every AI vendor in your stack. Capital will keep flowing into this category. Your job is to make the evidence flow too.
Jonathan Govette is Co-Founder and CEO of Oatmeal Health, which builds AI lung cancer diagnostics to catch cancers earlier in underserved communities. He writes The Oatmeal Bite for health system leaders, FQHCs, radiologists, and benefits leaders.
Key References
- Healthcare Dive, "Suki raises $90M in Series E round to create AI agents," by Anuja Vaidya, October 13, 2026. - Pulse 2.0, "Suki Secures $90 Million Series E Financing To Scale Healthcare AI Agents," by Amit Chowdhry, October 20, 2026. - AI Weekly, "Suki Raises $90M Series E Led by Healthier Capital at 450-Customer Scale," October 12, 2026. - Mass General Brigham newsroom, "AI scribes linked to modest reductions in EHR documentation time," April 1, 2026. Underlying study: Rotenstein et al., JAMA, DOI 10.1001/jama.2026.2253. - NEJM AI, "Ambient AI Scribes in Clinical Practice: A Randomized Trial," December 2025, PubMed abstract (PMID 41497288). - Becker's Hospital Review, "The fastest-adopted tech in healthcare: Report," by Naomi Diaz, March 26, 2025. - MedCity News, "Do Ambient Scribe Startups Have a Future Now That Epic Launched Its Own Tool?" by Katie Adams, February 5, 2026. - HLTH, "Epic Releases Native AI Charting for Integrated Clinical Documentation," February 6, 2026. - Healthcare Innovation, "AI Scribe Platforms Get More Venture Funding to Move Further Upstream," by David Raths, July 30, 2025. - Textbook of Digital Health. Source for the secondhand report of a 30-day burnout quality-improvement study.