Guidehealth

A hybrid model where the AI is real but the delivery is human. The company describes a proprietary AI built for real world care alongside Healthguides, virtually embedded medical assistants trained in data science working under licensed clinical supervision, and generative AI risk predictive models used by those Healthguides to engage patients and coordinate care. The AI predicts which patients need support; people then do the outreach and coordination. A buyer is purchasing a staffed service augmented by models, not software, and should size the engagement accordingly.

Oversight is structural and unusually clear because a licensed human sits between the model and the patient by design. Healthguides operate under licensed clinical professionals and extend the reach of the care team rather than substituting for it, and the predictive output informs prioritization rather than making clinical decisions.

The tradeoff is that the model's influence is indirect and therefore harder to audit: a risk score that quietly shapes who gets outreach is consequential even though no clinical decision is formally delegated to it.

The company names a proprietary AI and describes a combination of its own models with commercially available AI point solutions under one contract, plus a stated intent to move from predictive toward prescriptive analytics. That is more architectural candor than most services businesses offer, and acknowledging reliance on third party models rather than claiming everything is proprietary is to its credit. No model documentation, validation methodology, or performance figures are published.

One statement here is unusually honest for a services organisation and it is what places this above the floor. The company describes its offering as a combination of its own models with commercially available artificial intelligence point solutions delivered under one contract.

Acknowledging reliance on third party models rather than presenting everything as proprietary is a real disclosure, and it is the opposite of the pattern this index records repeatedly, where a vendor's technology is described as proprietary in a way that forecloses the question. It tells a buyer that a chain exists and that the contracting party is assembling rather than building. What it does not do is name anyone in it.

No point solution is identified, no model or provider is named, no hosting arrangement is published and no sub processor list was located, so a buyer knows there are third parties and cannot enumerate them or read their terms. Three data streams meet here and widen what those unnamed parties may touch: embedded personnel working inside partner clinical workflows with access to the record rather than an extract, predictive models running over population data at panel scale, and a conversational agent generating new patient statements that existed nowhere before.

The multi partner structure adds the last question: whether models improved from one partner's population are served to another. Ask which point solutions are in use, for a sub processor list, and how partner data is separated.

The evidence is institutional rather than statistical, and it is substantial. A large academic health system reported significantly improved CMS star ratings following its collaboration with the company, then invested 10 million dollars in it, which is a costly form of endorsement. Another major health system participated in its seed round. The platform is reported to support more than 500,000 lives. What is missing is published outcome or total cost of care data with methodology, so buyers have strong signals of customer satisfaction and weak evidence of measured effect.

Converted from Not Rated. No published stewardship framework, retention position, training use statement or de identification posture was located across two passes. The prior note's characterisation stands and can be stated more concretely: the data surface here is continuous and operational rather than transactional.

Three distinct streams meet in this company. Embedded personnel work inside partner clinical workflows, which means access to the record itself rather than to an extract. Predictive models run over population data to identify patients needing support, which requires claims, clinical and likely social data at panel scale. And a conversational agent interacts with patients directly, which generates a new record of what a patient said that did not previously exist anywhere.

That third stream deserves particular attention because it is the newest and least governed. A patient describing symptoms, circumstances or barriers to an automated agent is producing clinical information outside a clinical encounter. Whether that conversation is retained, whether it reaches the patient's record, who at the practice or health system can read it, and whether it informs model development are all unaddressed.

The multi partner structure raises the fourth question. A services organisation serving several provider networks accumulates data across them, and whether models are improved from one partner's population and served to another is a commercial question as much as a privacy one.

Ask for retention across each stream, a written position on model training, and how partner data is separated.

Converted from Not Rated. No business associate statement, availability, scope or subprocessor list was located across two passes. The prior note was right that such arrangements are structurally necessary and certain to exist contractually; the gap is publication.

What makes this more than a routine absence is that the company occupies several regulatory positions at once, and which one applies determines the instrument.

As a services organisation supplying embedded staff and administrative functions to provider organisations, it is a business associate in the ordinary way. Where it performs utilisation management and prior authorisation, it may be acting on behalf of a payer rather than a provider, which is a different principal with different permitted uses. Where two sided risk arrangements are involved, the entity bearing risk may itself be performing functions that look like health plan activity, and a managed services organisation structured that way can fall into hybrid territory of the kind this index has recorded elsewhere. And the conversational agent speaking to patients raises the same telephone consumer protection questions this index has framed for outbound healthcare contact, which turn on message type rather than on the vendor.

None of that is unusual for the segment and none of it is published, which leaves a provider organisation to establish it in diligence rather than read it.

Ask which entity signs and on whose behalf it acts for each function, whether any component operates as a covered entity or hybrid entity, what the agreement permits by way of data use across partners, and what consent supports automated patient outreach.

Converted from Not Rated after a second search that again failed to reach any security material from the company. No SOC 2, HITRUST, ISO 27001 or equivalent attestation was located, and no trust centre or security page was found across two passes.

The exposure is broader than for a software vendor and worth stating precisely. This company does not sell a tool that a provider operates. It supplies embedded personnel who work inside partner workflows on identifiable patient data, operates a conversational agent that speaks with patients directly, and runs administrative services including prior authorisation, utilisation management, care management and claims payment. Each of those is a continuous operational relationship with access to clinical and financial records, not a bounded data transfer.

That changes what an attestation would need to cover. The relevant scope is not only a platform but a workforce: how remote staff access partner systems, what credentials they hold, how access is provisioned and revoked as staff change, whether activity is logged in a way the partner can audit, and what separates one partner organisation's data from another's when the same operational teams serve several.

Those questions are answerable and none is addressed publicly.

Ask whether an attestation exists and what it covers, how embedded staff access is controlled and logged, how partner data is segregated across a shared services operation, and what the partner can independently audit.

No FDA pathway applies. Risk stratification and care coordination for value based contracts sit outside Software as a Medical Device. The regulatory regime that actually governs this vendor is the CMS accountable care and shared savings program rules its customers operate under, which is a payment regime rather than a device pathway.

Converted from Not Rated. The prior note identified the right concern: a model predicting which patients most need support is allocating a scarce human resource, and if the underlying data reflects historical access disparities it can reproduce them while appearing to optimise. That holds, and a second pass sharpens it considerably.

The company's service range extends well beyond care support. Its managed services organisation offering includes outreach, referral management, prior authorisation, care management, utilisation management and claims payment, alongside two sided risk arrangements in value based care. Two of those are not care coordination. Utilisation management and prior authorisation are coverage decisions, determining whether a requested service is approved, and they sit under a body of state and federal regulation that care management does not touch, including recent rules on automated decision making in coverage determination.

The structural point follows from combining them with the risk position. A company holding two sided risk carries financial consequence for utilisation, and the same company is performing utilisation management. That is a recognised arrangement in managed care and it is not improper. It does mean the organisation has a direct financial interest in the outcome of decisions its models inform, and where a model contributes to a coverage determination the governance expectations are different from those applying to a prioritisation tool.

The conversational agent adds a third surface, since it speaks with patients directly.

Ask which decisions models inform, what human review applies to any coverage determination, how appeal rights are preserved, and whether prioritisation has been examined across patient populations.

The route into this band is an externally audited accreditation specific to artificial intelligence in health care, which this index has recorded as one of four distinct routes and which very few vendors hold. An accrediting body examining an organisation's artificial intelligence practices against a published standard is a falsifiable claim with a named third party behind it and a scope statement a buyer can request, and it differs from a general security attestation because the subject of the examination is the artificial intelligence itself.

Being among the first to hold it also means the standard was applied recently rather than inherited. The limit is the one this index applies to every certification and it should be stated plainly rather than buried: an accreditation examines whether processes exist and operate, not whether outputs are correct. It creates no obligation to the customer when a prediction is wrong, no accuracy commitment and no remediation duty, and none of those was located.

Nothing measures the models either, with no validation methodology or performance figures published, which matters because the predictive layer identifies which patients receive proactive support and the failure direction that hurts is the patient who was never surfaced. Ask for the accreditation scope statement, for model validation and performance, and for what the company commits to when a patient is missed.

The company states its platform and staff integrate directly into primary care workflows and that it handles referrals, scheduling, prior authorization, and utilization management, which implies working inside customer systems. The data analytics foundation came from an acquired value based care services division with a licensing agreement for that vendor's analytics platform. No named EHR integrations, HL7, or FHIR capability were located, so depth is asserted operationally rather than documented technically.

Described as a cloud based solution. No tenancy, hosting, or data residency detail was located beyond that characterization.

No published pricing. The company describes an affordable operating model and offers technology plus staffing plus AI point solutions under a single contract, which is a real simplification for a buyer managing multiple vendors, but the economics are undisclosed. The central commercial question, whether compensation is fee based or tied to shared savings performance, is not addressed publicly and materially changes the risk a health system takes on.

Scoped to value based primary care for health systems, provider networks, and clinically integrated networks, spanning accountable care arrangements across lines of business. Coverage extends through partnerships rather than internal build, including a virtual cardiology program for at risk heart failure, hypertension, and arrhythmia patients, and an advance care planning partner. Specialty depth is therefore partner dependent, and the core competency is primary care panel management.