Fabric

The grade describes the mechanism rather than the quality of the product. Conversational artificial intelligence is one component of an assembled platform, and the company is explicit that it is paired with physician authored clinical logic and customisable decision trees under the name Hybrid AI. Decision trees are rules, not models, and a customisable one is configured by the customer.

The company's own history supports the reading. The conversational capability arrived by acquiring GYANT in 2024, the asynchronous care engine arrived by acquiring Zipnosis, and the clinical network arrived by acquiring a physician group's virtual service line. What is being sold is consolidation of a fragmented technology estate, which is a real and valuable proposition, and the artificial intelligence is a feature of it rather than the thing itself.

Oversight is structural and comes from the design rather than from a published policy. Symptom gathering and triage produce a routing recommendation, and a clinician delivers the care, including through the company's own 50 state provider network in the virtual pathway. Nothing is treated autonomously.

The pairing of conversational models with physician authored logic is itself a constraint on model behaviour, since the clinical pathway is authored rather than generated. What is not published is where the boundary sits: which parts of a patient interaction the model drives freely, which are constrained to the authored tree, and what happens when a patient describes something the tree does not cover. For a product that routes patients to an acuity level, that boundary is the safety question.

No model, architecture or training description is published. Hybrid AI is a marketing name for the combination of conversational models with clinical logic and it is not elaborated technically anywhere in public material.

Nothing quantifies triage performance, which is the measurement that matters for this product. There is no published accuracy for acuity assignment, no undertriage or overtriage rate, and no description of how the clinical protocols were validated beyond the statement that physicians and clinical informaticists wrote them. A buyer is asked to trust the authorship of the protocols without seeing how the resulting system performs.

Nothing identifies any party in the chain: no model or model family, no foundation model provider, no hosting arrangement and no sub processor list was located in two passes, and no position on whether patient conversations contribute to model development was found.

The term standing in for a technical description is a marketing name for the combination of conversational models with clinical logic and is not elaborated anywhere in public material, so a buyer cannot establish whether a third party conversational model is involved at all, which is the first question for any product of this shape. The published security page addresses access control in general terms without encryption detail or a retention schedule.

One structural feature widens what this axis covers here. Because the company both operates the software and, through an acquired network, employs or contracts clinicians delivering virtual care, the parties handling patient content include a clinical workforce as well as systems, and nothing states where those clinicians sit, how their access is scoped, or whether the same entity holds both roles for a given patient.

Ask whether any third party conversational model is invoked, for a sub processor list, for the training position on patient conversations, and for the entity structure across the software and care delivery pathways.

Operational claims are specific and the customer list is named and substantial: OSF HealthCare, MUSC Health, Luminis Health and Intermountain, with stated coverage of more than 100 million lives.

The figures are vendor published and carry no baselines. Clinicians working two to ten times faster is a range wide enough to describe almost any result, and the qualifier that it depends on setting is honest but removes the number's usefulness. A 15 percent call centre volume reduction is more concrete and still lacks a denominator or period. The inherited asynchronous claims of 99 percent administrative automation and 89 second treatment describe a highly constrained low acuity pathway rather than the platform. No peer reviewed evaluation was located.

The published security page addresses access control in general terms, stating that unauthorised access to patient data is prevented through logical and physical restrictions, but the specifics this axis looks for are absent: no encryption detail, no retention schedule, no position on whether patient conversations contribute to model development.

One structural feature deserves naming because it is unusual. The company both operates the software and, through its acquired network, employs or contracts the clinicians who deliver virtual care. That means it is not only a processor of a health system's patient data but in some pathways the treating provider, and those are different legal and stewardship positions that public material does not separate.

Compliance with United States health privacy law is stated on the published security page, which is more than an unlocated claim, and it remains an assertion because no body certifies that compliance.

No business associate agreement posture, notice or supporting documentation was located. The dual role complicates the usual analysis: a company acting as a technology supplier to a health system and separately as a provider organisation through its own clinician network occupies two different positions under the same law, and nothing published explains which applies to which pathway.

A dedicated security and compliance page exists and names SOC 2 Type 2 with the type stated rather than an unqualified SOC 2 claim, alongside health privacy compliance and, unusually, conformance with the WCAG 2.1 accessibility standard across user interfaces.

One inconsistency is recorded rather than resolved in the vendor's favour. The integrations page states the platform is held to standards including both SOC 2 and HITRUST, while the dedicated security and compliance page names SOC 2 Type 2 and does not mention HITRUST at all. Where a marketing page claims more than the security page, this index grades on the security page, which is the more specific and self limiting document. Held below A for that reason, and because there is no trust centre, no statement that the report can be requested, and no penetration testing disclosure.

No clearance and none claimed. Symptom gathering and care routing generally sit outside device regulation where the software supports rather than replaces clinical judgement, and the architecture here, with authored clinical pathways and a clinician delivering care, is consistent with that position.

The boundary is nearer than the marketing suggests. Software that assigns an acuity level and directs a patient to emergency care, an urgent visit or self care is making a consequential recommendation, and the exemption it relies on generally depends on the clinician being able to review the basis for it independently. Nothing published addresses which exemption the company considers applicable or how the basis is surfaced.

Nothing published on subgroup performance, model monitoring or triage error rates, and the scale makes the omission consequential: this system is stated to reach more than 100 million lives and its job is deciding where a patient enters the health system.

Triage is one of the better documented sites of disparity in medicine, with differences in acuity assignment recorded across race, language and presentation style. A conversational system inherits whatever is in its training and its authored pathways, and it applies that uniformly at a scale no triage nurse operates at. The accessibility commitment to WCAG 2.1 is a genuine and rare counterweight, because a digital front door that cannot be used by patients with visual, motor or cognitive impairment excludes exactly the people with most need. It answers the interface question and not the model question.

Two passes located nothing quantifying triage performance, which is the only measurement that matters for this product. There is no published accuracy for acuity assignment, no undertriage or overtriage rate, no description of how the clinical protocols were validated beyond the statement that physicians and clinical informaticists wrote them, and no warranty, indemnity or remediation commitment.

A buyer is asked to trust the authorship of the protocols without seeing how the resulting system performs, and authorship and performance are different things: a well written protocol implemented by a system that misreads presentations produces confident wrong answers. The two error directions are not symmetric and only one of them is visible. Overtriage sends someone to a higher level of care than they needed, which costs money and is noticed.

Undertriage sends someone home or to a lower acuity pathway when they needed more, and the failure surfaces later as a deterioration nobody connects back to the triage decision. One structural feature changes the liability picture entirely and deserves naming, because it is unusual in this index.

The company both operates the software and, through its acquired network, employs or contracts the clinicians delivering virtual care, so in some pathways it is not a processor of a health system's data but the treating provider. That is a different legal position with different duties, and public material does not separate the two. Ask which pathways are which, and for undertriage rate.

Integration is treated as a named product surface rather than an afterthought. The company publishes an integrations page citing HL7, FHIR and application programming interface connectivity into a health system's record and other systems.

Deployments at large systems including OSF HealthCare, MUSC Health and Intermountain corroborate that this works at enterprise scale, since intake, scheduling and asynchronous visit documentation all have to land in the record for the workflow to make sense. Held at B because no specific record vendor certification or partnership is named and the write back mechanism is not described at the standards level.

Not described. No hosting model, region, retention schedule or customer controlled option was located.

The question has an extra dimension here because the platform carries recorded patient interactions, asynchronous visit records and, in the virtual pathway, clinical encounters delivered by the company's own network. Those are different data classes with different retention obligations, and where a virtual visit record legally lives, with the health system or with the company as treating provider, is not addressed in public material.

Nothing published. No price, no pricing mechanism and no unit of sale, with a demonstration request as the only route.

The structure is unusually hard to price from outside because the company sells three suites plus access to a clinical network, so a buyer may be purchasing software, clinical labour, or both, and those have entirely different cost curves. The stated value proposition is consolidating a patchwork of point solutions, which is a claim about total cost that cannot be evaluated at all without a number. Buyers should ask for software and clinical network pricing separated.

Broad by construction, since each acquisition added a setting. Coverage spans the emergency department, in person ambulatory care, asynchronous and synchronous virtual care, and the engagement layer before a visit begins, with a stated reach of more than 100 million lives.

The buyer set is correspondingly wide: health systems, health plans, employers and brokers, which is three different purchasing motions for one platform. Clinical scope is deliberately shallow rather than broad, concentrated on low acuity and undifferentiated presentations where routing and asynchronous treatment are appropriate, which is the right scope for the product and does bound it. United States only.