Health System AI Platforms
P

Proximie

Proximie began as a way for a surgeon to virtually scrub in to any operating room and has become an operating system for the OR built on the data that capability generates. The original product is augmented reality telepresence: a remote expert joins a live procedure with full audio and video, annotates the surgical field, and observes, advises or teaches in real time, with every case recorded and indexed into a secure library for debrief, coaching and structured video review. On top of that sits the Intelligence Suite, described by the company as an ambient AI infrastructure layer that continuously captures intraoperative video, workflow activity imagery and instrument data across hundreds of facilities. In April 2026 Proximie joined NVIDIA's Project Rheo, building its Smart OR platform on NVIDIA foundation models while contributing real world surgical data back to inform NVIDIA's next generation healthcare models. The company reports deployment in more than 500 hospitals across 50 countries and five continents, more than 20 peer reviewed publications, and platform users seeing operating room productivity improve by up to 24 percent, which it frames as up to 300 additional procedures per room per year through better workflow visibility and more accurate prediction of surgery duration. Proximie was founded in 2016 by Dr Nadine Hachach-Haram, a practising NHS consultant plastic surgeon, and raised an 80 million dollar Series C with SoftBank participation. It is also lead author of a report on patient safety in surgery and the case for reform in the NHS.

Last VerifiedJuly 22, 2026
Compare Proximie with other vendors
Founded
2016
Headquarters
Boston, Massachusetts
Categories
health-system-ai-platforms, workforce-and-training
Indexed Products
Surgical Suite, Intelligence Suite
Assessment

Capability Axes

AI Capability
AI Centrality
B
Vendor Published

Proximie's founding product was connection, not intelligence. Augmented reality telepresence letting a remote expert join a live operation is a communications and collaboration capability that works with no model in it, and that is still what many customers buy. The AI arrived later, layered onto the data the capture footprint produces, and is now positioned as the Intelligence Suite and an operating system for the intelligent OR. The moat is the dataset in its clearest form: continuous intraoperative capture across more than 500 hospitals in 50 countries produces one of the largest surgical datasets in existence, and that asset, not a particular model, is what the NVIDIA collaboration values. Graded B. Rising rather than static, since the intelligence layer is where current investment is going, and worth re examining on the next verification pass.

Autonomy and Oversight Model
B
Vendor Published

Nothing acts autonomously. The telepresence layer is human to human by definition, a remote surgeon advising a present one, and the analytics inform scheduling and workflow decisions made by managers. The oversight question that does apply concerns prediction: the platform predicts surgery duration and those predictions drive scheduling, so a systematically wrong estimate produces either idle theatre time or cases running past staffed hours, which is a patient safety issue as well as an operational one. No accuracy figure, confidence interval or override behaviour is published for the duration model. Publishing prediction accuracy and how schedulers are shown uncertainty would move this to A.

Model and Technology Transparency
C
Vendor Published

The technology description spans augmented reality, machine learning and computer vision without specifying what performs which function. The most concrete disclosure is architectural rather than technical: the Smart OR platform is built on NVIDIA foundation models under Project Rheo, which tells a buyer something real about the underlying stack and is more than most peers offer. Below that there is no model card, no accuracy figure for any individual analytic, and no description of how workflow phases or instrument use are recognised from video. Naming which analytics are model driven, and publishing accuracy for the duration prediction that customers actually schedule against, would move this to B.

Clinical and Operational Evidence
B
Vendor Published

Better evidenced than most of this vein, on academic footing rather than case studies. The company reports being featured in more than 20 peer reviewed publications, and its platform appears in the independent literature as the recording environment for surgical research, including a published study on training a novice to label robot assisted radical prostatectomy video, conducted under Guy's and St Thomas' NHS Trust approval. Being the instrument other researchers publish with is a distinct and credible form of evidence. The operational claims are weaker and should be read as ceilings: productivity improvement of up to 24 percent and up to 300 additional procedures per room per year are vendor stated, without denominators, baselines or comparison sites. Publishing the productivity methodology, or naming the sites behind the 24 percent, would move this to A.

AI Safety and PHI Stewardship
C
Vendor Published

The compliance posture is evidenced and the secondary use disclosure is not, and for this vendor that is the gap that matters. On the first point, HIPAA and GDPR compliance are documented and were sufficient for Guy's and St Thomas' NHS Trust information technology approval in a published study, which is stronger corroboration than a marketing claim. On the second, the April 2026 NVIDIA arrangement involves Proximie contributing real world surgical data and insights to help inform NVIDIA's next generation AI models. That is patient derived intraoperative video and workflow data moving toward a third party's foundation model development, and nothing located states what is de identified, at what aggregation it leaves, whether patients or institutions consented to that specific use, or whether a hospital can opt out while remaining a customer. Publishing the terms of the data contribution would move this to B and is the single most important disclosure this vendor could make.

Regulatory and Compliance
HIPAA and BAA Posture
B
Third Party Estimated

Stronger than the category norm and independently corroborated. HIPAA and GDPR compliance are stated by the vendor and, unusually, appear in the peer reviewed literature: a published study describes Proximie as a GDPR and HIPAA compliant augmented reality platform approved by an NHS trust information technology department for recording robot assisted procedures. Third party corroboration of a compliance claim is rare in this index and is credited. No BAA template or scope description was located publicly, and for a platform where video routinely crosses institutional and national boundaries for telementoring, the scope questions are unusually consequential. Publishing the BAA and the cross border data terms would move this to A.

Security Certifications and Trust Center
C
Vendor Published

No SOC 2 Type II, ISO 27001 or equivalent attestation was located in retrieved material and no trust center was found. The absence is more surprising here than elsewhere in the category, because a vendor operating in NHS trusts and across European health systems would ordinarily hold ISO 27001 and complete the NHS Data Security and Protection Toolkit, and the documented trust level approval implies some assessment took place. The likeliest reading is that certifications exist and are not published, which is itself a finding for a buyer who has to evidence third party assurance. A published certificate list would move this to B without any change to the underlying posture.

FDA and Regulatory Status
Not rated

Proximie is collaboration, capture and analytics software rather than a diagnostic or therapeutic device, so no FDA pathway applies and the axis is rated accordingly rather than penalized. Two other regimes govern in practice. The first is discoverability, as across this whole surgical vein: recorded and indexed operations tied to named surgeons are what a malpractice plaintiff seeks, and state peer review privilege was not drafted with routinely recorded procedures in mind. The second is jurisdictional, and it is sharper here than for any peer: telementoring means a surgeon in one country advising on a live patient in another, which raises licensure, scope of practice and liability questions that no software vendor can resolve for its customers. Nothing published addresses who carries clinical responsibility when remote guidance contributes to an adverse outcome, and that should be settled contractually before any cross border use.

AI Governance and Bias Disclosure
C
Vendor Published

No bias assessment, fairness testing or governance framework was located. Two specific risks apply and neither is addressed publicly. The first is the unadjusted comparison problem recorded against Theator and Caresyntax: workflow analytics that surface variation between surgeons and sites, without published risk adjustment for case mix and acuity, penalise whoever takes the harder work. The second is distributional and particular to this vendor's stated mission of democratising surgery: models trained predominantly on data from well resourced hospitals with modern equipment may perform worse in exactly the lower resource settings the platform is meant to reach, and nothing published describes how performance is validated across the range of the 50 country footprint. Publishing site level validation across resource settings would move this to B and would substantiate the mission claim rather than merely asserting it.

Integration and Deployment
EHR and Interoperability Depth
C
Vendor Published

The integration story is about the operating room rather than the record. Proximie connects to OR video sources and captures instrument and workflow data, and it makes recorded cases available through its own library and content management tools. No EHR integration was located, so scheduling insight and case documentation live in the vendor's environment rather than flowing into the systems a hospital already runs, which is a material limitation for a product whose operational value proposition is scheduling and utilisation. The contrast inside this vein is direct: Theator writes structured output natively into the Oracle Health EHR, while Proximie's intelligence appears to stay in Proximie. Publishing an EHR or surgical scheduling system integration path would move this to B.

Deployment Model and Data Residency
B
Vendor Published

The footprint itself is the disclosure and it cuts both ways. Deployment across more than 500 hospitals in 50 countries on five continents, with cloud based recording and cross border telepresence, means the vendor demonstrably operates multi region infrastructure, and GDPR compliance plus NHS trust approval indicate European handling is addressed rather than ignored. What is not published is the detail: no regions are named, no statement describes where recorded procedures are stored relative to where they were performed, and the NVIDIA collaboration adds a further question about where contributed data is processed. For a platform whose core function is moving surgical video across borders, residency deserves explicit treatment rather than inference from compliance claims.

Commercial
Commercial Transparency
C
Vendor Published

No rate, unit or charging mechanism was located. The commercial model is also more layered than most, since Proximie sells to hospital systems and partners with medical device manufacturers, whose representatives are among the parties that join procedures remotely, so a buyer should establish whether device partners fund or subsidise deployments and what that implies about vendor neutrality in the room. The operational case rests on throughput, up to 24 percent productivity improvement and up to 300 additional procedures per room per year, which is the savings linked framing the index scrutinises because the vendor supplies both the promise and its measurement. Publishing whether pricing is per room, per case recorded or per user would move this to B.

Setting and Specialty Coverage
A
Vendor Published

The widest geographic reach of any vendor in this vein and among the widest in the index. More than 500 hospitals across 50 countries and five continents, spanning operating rooms and cath labs, with a buyer base including hospital systems and medical device manufacturers. Because the platform captures the room rather than analysing one procedure type, it is not confined to endoscopic or robotic cases, and specialty coverage follows whatever the institution does. The international spread is not incidental to the product: the founding use case, extending a surgeon's expertise to places that lack it, only works at that spread, and it is what distinguishes this record from the largely US bound peers in the category. The caveat recorded in the governance row applies, that reach across resource settings has not been shown to come with validated performance across them.

Commercial

Pricing

Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.

Entry Price Pricing Basis BAA Tier Implementation Source
Not published
Undisclosed. No published rate or unit. HIPAA and GDPR compliance stated and independently corroborated in the peer reviewed literature, including NHS trust information technology approval. No BAA template or cross border data terms located publicly. Not published. Deployment involves operating room capture hardware and network integration, so implementation is likely material, and no figure or timeline was located. Vendor Published

No rate, unit or charging mechanism was located. Two structural questions matter more than the figure. First, the commercial model is layered: Proximie sells to hospital systems and partners with medical device manufacturers whose representatives are among the remote participants joining procedures, so a buyer should establish whether device partners fund or subsidise deployments and what that implies for vendor neutrality inside the room. Second, the operational case rests on throughput, stated as up to 24 percent productivity improvement and up to 300 additional procedures per room per year, which is the savings linked framing the index scrutinises because the vendor supplies both the promise and the measurement of it. Ask whether pricing is per operating room, per case recorded or per user. Verified 22 July 2026.

AI Health Index

An independent reference for evaluating AI vendors in healthcare. No vendor pays for inclusion, placement, or rating.

Index Status
Last index update
July 23, 2026
The AI Health Index is an editorial reference, not a regulatory body. Vendor data is verified against published sources and public regulatory filings. Figures labeled “Estimated” have not been confirmed by the vendor. See the Methodology page for evaluation standards and limitations.
© 2026 AI Health Index
3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746