Vivian Health

Vivian says this plainly itself, and the candour is worth crediting rather than punishing: its own materials lead with the line that AI is only as strong as the data and humans behind it, and describe the AI Assistant as built on top of a network and data layer that came first. That is an accurate description of the business.

The marketplace ran for years without AI and would still run without it, and the durable asset is 2.8 million registered clinicians plus the hiring signal generated by roughly one million job searches and 2.7 million applications a year. This is the moat is the dataset pattern the index applies across the credentialing lane and to Reveleer.

What lifts it above a token layer is depth of use: roughly half of all applications on the platform are now screened by AI Assistant, which is real operational dependence rather than a marketing feature. Graded B. The grade describes the mechanism, not the quality of the product.

Vivian publishes the disclosure that decides this axis and that most of its peers withhold. Its employer facing material states that role requirements are evaluated with flexibility and that candidates are not disqualified unless the job explicitly requires it.

That is an explicit statement of the elimination principle, which is the consequential direction of error in automated screening: a wrong pass costs a recruiter time, a wrong rejection costs a clinician a job they were qualified for.

It also names the screening criteria rather than leaving them generic, listing current clinical activity, years of experience, board certification, active licensure in the required state and DEA registration for advanced practice roles, all of which are objective and job related. Outbound is gated too, with AI Proposals requiring a recruiter to review, edit and approve every message before it sends, and screened candidates handed to a human rather than advanced automatically. Graded A. The residual gap is what happens in the case the disclosure carves out, when a job does explicitly require a criterion the candidate lacks, and whether that rejection is ever reviewed.

The inputs are described but the model never is. Vivian names its data layer as detailed job requirements, clinician intent and behavior, and placement outcomes, and it names what the assistant screens for. Nothing published identifies the model family, whether matching is a learned ranking model or a rules engine over structured credential fields, whether any large language model is used for the conversational and drafting functions, or what was measured to validate it.

The provenance detail offered is organizational rather than technical, namely that the assistant was built over two years with two travel nurses on the team. That is useful signal about domain grounding and it is not a substitute for a model description. Publishing the model family and the feature classes behind match ranking would move this to B.

Nothing identifies any party in the chain: no model or model family, no statement of whether matching is a learned ranking model or a rules engine over structured credential fields, no indication whether a language model performs the conversational and drafting functions, no hosting arrangement and no sub processor list.

The data layer is named at input level, covering detailed job requirements, clinician intent and behaviour, and placement outcomes, which tells a reader what the system consumes without telling them who handles it. The clinical form of this axis does not apply, since the holding is clinician employment data rather than patient records, and the grade is for the domain equivalent. That equivalent is substantial.

Licensure, certification, controlled substance registration, work history and behavioural signal on a population approaching half the active nursing workforce is a labour market wide dataset held by a single company, and behavioural signal in this context means how a person searched, what they applied for and what they declined.

Nothing states how long a profile persists after a job search ends, whether it can be deleted rather than deactivated, or whether one employer's candidate interactions inform models serving a competing employer, which is both a privacy question and a commercial one in a market where employers compete for the same nurses. Ask all three, plus a sub processor list.

Every figure is vendor generated and most are framed as ceilings rather than averages. Vivian reports candidates responding up to 4 times more often with AI Assistant, up to a 15 percent lift in placements, 20 percent more complete candidate profiles, and 8 in 10 recruiters reporting positive workflow impact. Beta figures for AI Proposals report proposals sent nearly 12 times faster, up to 3 times higher response rates and up to 1.7 times higher proposal to application rates.

Named customers appear as logos and testimonials rather than as study sites, and no independent or peer reviewed evaluation was located. The figures are at least internally consistent and tied to named mechanisms, which is more than several vendors in this category manage, but up to framing without denominators or a control comparison is a marketing claim rather than evidence. Publishing one customer cohort with a before and after denominator, or a holdout comparison against unassisted recruiting, would move this to B.

Vivian handles clinician employment data rather than patient data, so PHI stewardship does not apply in the provider sense. The C is not for the absent PHI posture.

It is for the domain equivalent, which is substantial and unanswered. Vivian holds licensure, board certification, DEA registration, work history and behavioural signal on 2.8 million healthcare professionals, a population approaching half the active United States nursing workforce, and its AI Assistant generates screening assessments about named clinicians that are shared with prospective employers.

Nothing located states whether a clinician can see or contest the screening assessment attached to them, how long a profile persists after a job search ends, or whether one employer's candidate interactions inform models serving a competing employer. Those are the three disclosures a buyer and a clinician should both want, and they are the employment data analogue of what this index requires on PHI everywhere else. The asymmetry worth naming is that the person being assessed is not the customer and has no route to the assessment.

No patient data is in scope, so a HIPAA posture is not applicable in the provider sense and is rated accordingly rather than penalized. No BAA language was located and none would ordinarily be required, since a marketplace handling job applications is not receiving PHI from a covered entity.

The governing instrument is the platform's terms and privacy policy together with whatever data protection terms sit in the employer agreement, and those are where the credential retention and secondary use questions raised in the stewardship row have to be answered.

No SOC 2 Type II, ISO 27001 or equivalent attestation was located and the site carries no trust center, with the footer linking only to a privacy policy and terms of service. Applicability is not in question: Vivian stores identity, licensure and DEA registration records on millions of clinicians, which is a credential dataset attractive to anyone attempting credential fraud or impersonation of licensed practitioners, and it exchanges candidate data with employers and agencies.

A published SOC 2 Type II would move this to B. Given that the platform now sells into large health systems, procurement at that tier will ask for one regardless of what the public site shows, so the absence may reflect marketing rather than posture, and buyers should ask directly before assuming either way.

Vivian is not a medical device and makes no clinical claim, so no FDA pathway applies and the missing clearance is not the reason for this grade.

The C reflects a regime that squarely applies and that nothing published addresses. A system that screens candidates against role requirements and ranks them for recruiter attention is an automated employment decision tool under New York City Local Law 144, which requires an annual independent bias audit with published results, and it sits inside the scope of the Illinois law on AI in video interviews and the Colorado framework for consequential decisions. No such audit or equivalent filing by Vivian was located.

The complication specific to this vendor is worth raising in contracting. It sells to staffing agencies as well as to direct employers, so the party carrying the audit duty depends on who actually makes the hiring decision. That allocation should be settled in the agreement rather than assumed by either side.

The architecture is defensible and the disclosure is not. What protects Vivian here is that its published screening criteria are objective and job related, namely licensure, board certification, DEA registration, recent clinical activity and years of experience, which is the strongest structural defense against disparate impact, and that its stated design principle is not to disqualify candidates unless the job explicitly requires it.

Those are governance relevant choices and they are credited. But nothing published names a bias mitigation method, reports fairness testing, gives an adverse impact ratio, or shows an independent audit, and no Local Law 144 filing was located. The comparison inside this category is unflattering: Arena Analytics names its technique, states that it strips protected characteristics and their proxies, and commits to the EEOC four fifths threshold. Vivian ranks and screens at far greater scale and says nothing equivalent. Naming a fairness method and publishing an adverse impact ratio by role would move this to B.

Two passes located no model family, no accuracy measurement, no validation and no warranty, indemnity or remediation commitment, and the output is an assessment about a named individual. The assistant generates screening assessments about named clinicians that are shared with prospective employers, and nothing states whether the clinician can see the assessment attached to them, understand what it was based on, or contest it.

The asymmetry is the finding and it should be stated plainly: the person being assessed is not the customer and has no route to the assessment. Every mechanism this axis normally credits, traceability, an appeal path, a published error rate, exists to let the affected party act, and here the affected party is structurally outside the relationship. Scale makes it consequential rather than theoretical.

The company holds licensure, certification, registration, work history and behavioural signal on millions of healthcare professionals, a population approaching half the active nursing workforce in the country, so an assessment practice applied at that scale shapes access to work across a labour market rather than at one employer.

The provenance detail offered is organisational rather than technical, that the assistant was built over two years with two travel nurses on the team, which is useful signal about domain grounding and is not a substitute for a model description or an error rate. Ask what the assessment is based on, whether a clinician can view and dispute it, and the accuracy of the screening determination.

No EHR integration exists and none is needed, since this product writes to hiring systems rather than clinical ones, so the axis is assessed on the interoperability surface that applies. On that surface the published detail is thin.

One named integration was located, a partnership announced in October 2025 embedding Vivian's hiring technology into Hallmark Health Care Solutions' vendor management and flexible workforce products, enabling one click submission of travel, per diem and local candidates. Beyond that, applicant tracking system integration is asserted in third party descriptions rather than documented by the vendor, and no list of supported systems appears on the employer site.

That is a material gap for a health system buyer whose recruiters live in an existing ATS, and it contrasts directly with Arena Analytics in this same category, which names nine ATS and HRIS connectors outright. Publishing the supported integration list would move this to B.

Nothing published was located on hosting model, tenancy, region or residency, and the earlier assessment was right that the deployment model is not genuinely in question for a consumer facing marketplace while residency, retention location and subprocessor disclosure are.

The second pass answers the subprocessor question partially, and the partial answer is the finding.

The vendor discloses that it shares information with recruiters and employers, its own service providers, advertising partners, event participants and sponsors, and social media services, and that certain third party data collection tools used by advertising partners may target advertising based on activity on the site. Its mobile application separately declares that data may be used to track users across applications and websites owned by other companies.

So third parties receiving data are acknowledged in category but not named, and one category has no connection to healthcare employment at all. That is the subprocessor position for practical purposes, and it means a nurse's activity on the platform reaches organisations they have no relationship with.

That sits alongside what the platform holds. The privacy policy contemplates users supplying a full social security number or driving licence, and commits not to share those unless the user instructs it for each application requiring them. The commitment is a good one and worth crediting; per application consent on identity documents is more protective than most marketplaces offer. It also confirms that government identity numbers are held at a scale of roughly two million professionals, and where those sit is a residency question with real weight.

Ask for the hosting region, the named subprocessor list including advertising partners, retention for identity documents specifically, and what happens on a change of business control.

There is no pricing page. Both employer paths, direct healthcare employer and staffing agency, resolve to a demo request, and no rate, unit or charging basis is published. Third party summaries describe an employer paid model billed per hire or by subscription, which is recorded here as secondary sourced and unconfirmed.

The irony is worth stating plainly, because it is the sharpest commercial observation in this category: Vivian's entire brand proposition to clinicians is pay transparency, and it publishes salary data, employer reviews and mandatory rate breakdowns on job postings precisely because opacity disadvantages the worker. It extends none of that to the buyer on the other side of its own marketplace. Publishing the charging unit, without a figure, would move this to B.

This is the broadest coverage in the hiring segment of the category. Vivian spans nursing, allied health, therapy and advanced practice roles including nurse practitioners, physician assistants and certified registered nurse anesthetists, across travel, permanent, per diem, local contract and advanced practice employment types, with more than 250,000 job opportunities live.

The buyer base is equally broad and covers both sides of the labor supply chain, with named large health systems including Trinity Health, Tenet Health, Adventist Health, Memorial Hermann, University of Miami Health and Sanford Health alongside staffing agencies and a group purchasing workforce organization.

Because AI Assistant was extended in stages, starting with travel roles and later covering permanent, per diem, local and advanced practice work, buyers should confirm that the AI capability and not merely the job listing coverage reaches the employment type they are hiring for.