Andor Health

The company calls itself artificial intelligence first and has genuinely repositioned around agents, and a real communications platform still sits underneath.

What the models do is consequential: ambient sensing in the room, conversational documentation, triage at the digital front door, and orchestration of virtual encounters. That is work people previously did and it is the part the company now leads with.

What remains without them is also a business. Virtual visits, device agnostic virtual rounding, virtual sitting, secure care team collaboration and remote specialty consults are video and communications infrastructure. Tele stroke and tele intensive care connect one clinician to another; the platform carries the encounter rather than forming the judgement inside it. A health system that bought this for virtual nursing coverage would retain most of the value with the models switched off.

Graded B on that split, consistent with how Optum Integrity One and Infinx were treated: the intelligence is real and growing, and a substantial workflow and connectivity layer would function independently of it. Ask what proportion of deployed value sits in the agents versus the virtual care substrate.

Agents are described and their boundary is not. The platform is presented as orchestrating virtual experiences through digital front door agents, virtual triage and conversational patient interaction, which implies agents acting on patient contact without a human in the loop for at least part of each interaction.

Nothing published establishes where that line sits. No automation rate, containment rate, escalation rate or accuracy figure appears for any agent. No description of what triggers handoff to a human, what a patient sees when the agent is uncertain, or whether patients are told they are speaking to a machine.

That last omission matters more here than in the back office products elsewhere in this index, because these agents interact with patients directly rather than with staff. Several jurisdictions the company operates in are moving toward disclosure requirements for artificial intelligence in patient facing interactions, and the platform spans three countries with different expectations.

The documentation side is better bounded by nature, since a clinician signs a note, though nothing states whether generated content is marked as machine originated. Ask for the containment and escalation rates, and the disclosure behaviour at first patient contact.

More is said about the model layer than most records in this index manage, and the best of it comes from outside the company.

Capabilities are named specifically rather than generically: generative capability for ambient sensing and conversational documentation, multimodal agents, real time translation of signals from records and biometric devices into actionable insight. The architecture claim is that intelligence is applied to data unlocked from source systems rather than to a summary, which is a meaningful statement about where the models sit.

The model provenance statement is the notable disclosure. A third party health system marketplace listing states that the Microsoft investment relationship optimises virtual interactions using that company's partnered large language model technology, which identifies the model family. This record credits that as a real disclosure while flagging its source: it appears on a marketplace listing rather than in the company's own material, and the company should be asked to confirm it directly.

What is absent is measurement. No accuracy, word error rate, containment rate or clinical concordance figure is published for any component, despite ambient documentation and triage both being measurable. Graded B for named capability and identified model family with no published performance.

Better party identification than most, arriving indirectly.

The corporate relationship is public and material: Microsoft's venture arm holds an investment position taken in 2020 and is described as working closely with the company on optimising virtual interactions. A third party marketplace listing identifies the underlying model technology as that company's partnered large language model family. Between them a buyer can form a reasonable view of who is in the chain, which is more than the proprietary label offered by most competitors here.

The sourcing caveat applies as on the transparency axis: the clearest statement is a marketplace listing rather than company material.

What is missing is the rest of the chain and the terms on it. No sub processor list, no statement on whether customer data or captured audio and video contribute to model development, no retention position, and no cloud platform named despite the solution being described as cloud based. The data crossing that chain includes continuous ambient capture from patient rooms, which is a more sensitive payload than documentation text. Ask for the model provider and version, the sub processor register, and the training position on ambient capture specifically.

Three independent assessments with a published number on one of them, which is the strongest external validation profile in this lane.

The platform was rated 2026 Best in KLAS for virtual care platforms outside the record system category, with a score of 95.9 out of 100 that the company positions as 15.7 points above the overall average. Black Book has rated it highest in the same category for three consecutive years, and Frost and Sullivan recognised it for acute care virtual health in 2025. Two national group purchasing agreements, with Vizient and Premier, are meaningful because those bodies conduct their own technical and commercial diligence before contracting.

Deployment is substantial and named: more than 500 hospitals and 70,000 providers across three countries, with Sentara Health across twelve hospitals, Providence, and the National Institutes of Health Clinical Center identified.

Held at B because the assessments measure customer satisfaction and analyst judgement rather than clinical performance, and because the clinical claims are unsupported. A stated 47 percent reduction in readmissions is a serious outcome assertion that would ordinarily rest on a published study, and none was located. The efficiency figures, ten to twelve minutes per visit and a 35 percent fall in call abandonment, are plausible and equally unsourced. Ask for the readmission study.

Nothing operational was located. No retention schedule, encryption statement, access control description, data ownership or deletion position, data minimisation commitment, or statement on whether customer data contributes to model development.

Graded conservatively for the same reason as the privacy axis, and the same ambient capture caveat is the substance of the gap. The platform ingests records, biometric device signals and, in smart room and virtual sitting deployments, continuous room sensing. That is the broadest and most intimate data footprint of any record built in this sweep, and the published stewardship material is nil.

The questions that follow are specific. Whether ambient audio and video are retained or processed transiently. Whether captured room data is used to improve models. How a patient who declines monitoring is handled. Whether staff in the room are treated as data subjects. Whether the answers differ across the three countries the platform operates in, where consent standards for recording differ materially.

A vendor selling continuous patient room sensing has a stronger obligation to publish here than a documentation vendor, and currently publishes less.

No health privacy material was located in a dedicated pass: no compliance statement, control enumeration, de identification position, or business associate agreement posture, template or execution requirement.

Graded conservatively rather than punitively, on the basis applied to TeleTracking, Alcidion and Optum Integrity One. Deployment across more than 500 hospitals including a federal research hospital means privacy review has been satisfied repeatedly and agreements demonstrably exist. What the pass established is that none of it is published where a prospective buyer would look.

One dimension here is more sensitive than in most records and is unaddressed. Virtual sitting and smart room deployments involve continuous ambient audio and video capture in patient rooms, which is protected information of a different character from documentation, and which implicates people who are not the patient, including families and staff entering the room. Nothing published describes what is captured, what is retained, whether recording occurs or only real time streaming, or how consent is handled for room occupants. Ask specifically about ambient capture, not just about data handling in general.

No credential was located. A targeted pass including compliance and security terms found no controls report, no information security certification, no health specific security framework certification and no cloud authorisation. There is no trust center, no security page, no report availability process, no penetration testing disclosure and no vulnerability disclosure policy in the material retrieved.

The absence is very likely publication rather than substance. Two national group purchasing organisations conduct technical diligence before awarding agreements, a federal research hospital does not deploy a platform handling patient interaction without security review, and deployment across more than 500 hospitals in three countries implies repeated third party assessment. The artefacts exist somewhere.

That does not change the grade, because this index measures what a buyer can verify before contacting sales, and the gap is wider here than the count suggests: a platform performing continuous ambient capture in patient rooms carries a higher expectation of published security posture than a documentation tool, and publishes less than several vendors a fraction of its size.

Graded D on the basis applied to TeleTracking and Alcidion. Ask for the controls report, its type and period, and any certification held in each market.

No device pathway is claimed and for most of the platform none applies. Virtual visits, care team collaboration and remote consults are communications infrastructure carrying clinician judgement rather than forming it, and documentation support is administrative.

Two components sit closer to the line than the company acknowledges. Ambient sensing in patient rooms that detects clinically relevant events, and virtual patient monitoring described as delivering actionable insight from biometric device signals, both edge toward the territory where software analysing patient physiological data for clinical purposes can meet a device definition. Whether these products stay clear of that depends on design detail this record could not establish, and the company publishes no position on device status in any of the three countries it operates in.

The published readmission claim compounds it. Asserting a 47 percent reduction in readmissions attributes a clinical outcome to the platform, which is a stronger claim than an operational efficiency figure and invites the question of whether the product is making clinical claims that carry regulatory weight.

Graded C because the position is probably correct and entirely undocumented across three regulatory regimes. Ask for the device status determination and its basis in each market.

Nothing is published. No bias or fairness testing, no model validation methodology, no monitoring output, no distribution reporting, no drift detection, no external audit and no responsible artificial intelligence statement was located for any component.

The governance exposure here is broader than in the administrative products this index usually grades, because the agents touch patients directly and make allocative decisions.

Triage at the digital front door determines who is routed where and how quickly. A model that systematically under prioritises particular presentations, accents, languages or communication styles produces a distribution of access rather than a single visible error, and only population level reporting would reveal it. Conversational agents interacting by voice carry known performance variation across accent and dialect that is well documented in speech technology generally. Ambient room sensing carries the same question for visual detection across patient appearance and room conditions.

None of it is addressed. There is no statement that fairness was considered for any of these, no performance breakdown by patient population, and no description of what the triage agent optimises for. Graded D because a platform making patient facing allocative decisions across three countries publishes nothing about how it makes them.

No performance figure is published for any intelligent component, so there is no stated level against which a shortfall could be measured. No accuracy, containment rate, escalation rate, transcription error rate or detection sensitivity appears for the agents, the ambient documentation or the room sensing.

No service level agreement, warranty, indemnity or remediation commitment was located, and no pilot or validation offer of the kind several competitors publish was found.

The outcome claims run the other way from recourse and are worth separating clearly. Ten to twelve minutes saved per visit, a 35 percent fall in call abandonment and a 47 percent reduction in readmissions are benefit assertions, not performance commitments, and the last of them is a clinical claim with no published study behind it. A vendor asserting a readmission reduction while publishing no accuracy figure for the agents producing it has inverted the usual order of proof.

The apportionment question is sharpest at the digital front door. If a triage agent mishandles a patient contact and care is delayed, responsibility sits between the vendor's model, the health system's configuration and the clinician who never saw the contact. Nothing published addresses it. Only a published performance measurement or a contractual commitment will move this grade; further outcome claims will not.

The integration description is directionally strong and entirely unspecific.

What is claimed is genuine bidirectionality, which many competitors leave vague: the platform pulls patient and clinical context out of source systems into the virtual encounter, and pushes recommended tasks and documentation back into the record. It also integrates biometric devices and other real time signals, which is a second integration class beyond the record system and a harder one, and it describes device agnostic virtual rounding, implying tolerance of varied endpoint hardware.

What is missing is every particular. No record system is named. No interface standard is described. No connection mechanism is specified. No vendor marketplace or programme listing was located. No biometric device or platform is identified despite device integration being a stated capability.

Graded C for credible bidirectional capability with no published specification, consistent with how the same absence was treated for Infinx and TeleTracking, and below the B given to vendors naming a dozen systems. Ask which record systems are supported, through what standards, which device ecosystems integrate, and whether any integration is certified by the record vendor.

The delivery model is stated at the highest level and nothing beneath it. The solution is described as cloud based, which distinguishes it from an on premise deployment and is the whole of what is published.

No cloud provider is named, no region, no tenancy model, no residency commitment and no customer controlled option. The Microsoft investment relationship makes an inference about the underlying platform tempting and this record does not draw it, consistent with the treatment of cloud logos elsewhere in this index.

Residency is a live question rather than a formality because the platform operates in the United States, Canada and the United Kingdom. Canadian provincial health privacy law and British data protection requirements both constrain where health data may be processed and who may access it from outside the jurisdiction, and a single platform serving all three either runs regional instances or moves data across borders.

The ambient capture dimension sharpens it further, since continuous room audio and video is a far larger and more sensitive payload to move than transactional data. Ask for hosting regions by country, the tenancy model, and where ambient capture is processed and stored.

No price is published. A dedicated pass located no pricing page, no unit of charge, no range, no implementation or onboarding fee position, no minimum commitment, no pilot terms and no return calculator.

One commercial route is genuinely more accessible than for most vendors here and deserves recording. National group purchasing agreements with Vizient from February 2025 and Premier from February 2026 mean pre negotiated terms exist and are reachable by any member of either organisation without an individual negotiation from zero. For a large share of United States health systems that is a real procurement shortcut, and it is the closest thing to published commercial terms on this record.

It is not disclosure. Group purchasing pricing is visible only to members behind their own portals, so a prospective buyer outside those bodies learns nothing, and nothing indicates the unit of charge, whether the five pillars are licensed separately, or how agent based capabilities price against the underlying virtual care platform. Ask for the licensing basis, whether pillars are modular, and how the group purchasing terms compare with direct contracting.

Wide across care settings, with named specialty programmes and a genuine multi country footprint.

Coverage is stated across outpatient, inpatient, post acute and at home, organised into five pillars spanning visits, the virtual hospital, monitoring and care management, care team collaboration and community collaboration. Specialty depth is named rather than implied: tele stroke, tele psychiatry and tele intensive care are each identified as supported remote consult programmes, and virtual nursing, virtual sitting and virtual rounding are distinct inpatient deployments.

The footprint spans the United States, Canada and the United Kingdom across more than 500 hospitals, which places it among the small group of records in this index operating beyond a single national system.

Held at B rather than A because scale outside the United States is not quantified, so the Canadian and British presence cannot be sized, and because the named specialty programmes are capability statements rather than evidenced deployments with references attached. Coverage of the agent based capabilities specifically is also unclear, since the newer conversational and digital front door components may not be live everywhere the virtual care platform is. Ask which pillars are deployed at the named reference sites.