Enzo Health

Built AI native by design rather than retrofitted, with the company describing an EHR constructed around the model rather than a model added to an EHR. Every stage it covers, from referral analysis through documentation to claim preparation, is model driven.

The design intent is stated plainly and it is to act first. The company says Enzo does not just record what happened, it takes the next step before anyone has to ask, with the team only needing to review and approve. Across the episode that means analyzing referrals, verifying eligibility, building schedules, coordinating supplies, generating assessment documentation, checking codes and preparing clean claims automatically. Its own materials describe referrals analyzed before staff review is needed, schedules built the moment frequency is submitted, and every chart reviewed automatically so a person steps in only when something needs them.

A review and approve gate is asserted but not described. No confidence threshold, accuracy rate, abstention behavior or escalation rule is given for any of those steps. Acting ahead of the human across an entire episode with one undefined approval step is a wide surface for a single gate to cover, and the wider the surface the more the gate's design matters.

One mechanism disclosed in the mobile application listings is the clearest example of the product acting rather than proposing. On Android the application uses the operating system's accessibility service to automate medication data entry into a separate home health record system, reading screen content to identify input fields and simulating user input to fill medication names, dosages and other details. That is software driving another application's interface as though a person were typing.

The disclosure itself is candid, calling the screen automation what it is rather than an integration. The mechanism raises specific questions. Writes performed by simulated input carry the signed in clinician's identity, so the receiving system's audit log will attribute them to a person rather than to software. The company states the service is active only during medication writeback, that it does not collect, store or transmit personal data, and that a user must explicitly enable it. Establishing how the first of those is enforced matters, because an accessibility service that can read screen content in one application can generally read it in others.

The accuracy claim of 95 percent or better carries no methodology, reference standard, denominator or definition of what was measured, and it is the only performance figure published. Capability is otherwise described in behavioral terms, such as knowing the difference between small talk and a clinical finding, which is a useful way to explain the problem and not a measurement of how often it succeeds. There is no model card or evaluation protocol. Enzo describes OASIS-E1 fields while the current instrument is OASIS-E2, and nothing published says which version the product actually produces.

Nothing identifies any party in the chain: no model or model family, no foundation model provider, no hosting arrangement and no subprocessor list is published. Encryption and a full audit trail are stated, which record and protect access rather than naming who has it. The content types make the silence broader than for a capture only product.

This platform ingests referral documents, photographs of medication labels and recorded visits across an entire episode of care, so the unnamed chain is handling images and documents originating outside the practice as well as speech captured inside it.

Photographs in particular matter, because an image of a medication label is taken in a patient's home and may incidentally capture surroundings, other labels or people, and image handling is rarely covered by a retention policy written for audio.

One named customer testimonial from a registered nurse at a Utah home health and hospice agency describing improved OASIS coding, plus claims of 75 percent less charting time carrying no denominator or method. No study, controlled evaluation, accuracy benchmark, third party rating or deployment count is published.

Encryption and a full audit trail on every note are stated, and an audit trail is a real control.

Nothing else is disclosed: no retention schedule for audio or transcripts, no deidentification practice and no statement on whether customer content trains models, despite the product ingesting referral documents, medication label photographs and recorded visits across the whole episode.

HIPAA compliance and audit ready records are stated consistently across product materials. Business associate agreement terms are not published for inspection.

Enzo publishes no named or dated attestation, no SOC 2 report of either type, no HITRUST certification, no ISO 27001 and no trust center. The vendor describes HIPAA compliant infrastructure, which is a self description rather than an examination, and there is no certification for the health privacy rule in any case.

One published statement deserves close reading, because the qualifier inside it is doing a great deal of work. The company states that patient health information is processed inside its own platform and is never sent to external AI services in a way that would compromise protected health information. That is not a statement that protected health information is never sent to external model services. It is a statement that it is never sent in a way the vendor considers compromising, which leaves the vendor as the judge of precisely the question the buyer is asking. The phrasing neither names a model provider nor states flatly that no patient audio or clinical data routes to third party services. Whether encounter audio or transcript reaches any external model provider, which ones, and under what retention terms is left unsaid.

The absence matters more than a small vendor's ordinarily would. This is now a full electronic record holding the complete patient episode, for agencies the company states collectively support more than five hundred thousand patients annually, following a substantial funding round. Custodial scale of that kind is normally where an independently examined posture becomes the procurement baseline rather than a nice to have.

None of this means the controls are absent; it means they cannot be verified before contracting.

No clearance claimed and no United States device pathway attaches. A real and demanding regulatory framework does govern the product. It simply is not device regulation.

The product completes OASIS, the federally mandated patient assessment instrument for home health. That instrument is not documentation in the ordinary sense. The same answers drive three things simultaneously: the clinical care plan, the payment case mix under the patient driven groupings model, and the publicly reported quality star ratings the agency is measured against. Failures surface as survey findings, claim denials and audit exposure rather than as adverse event reports, and CMS assessment and payment policy is the governing regime.

That sharpens the incentive question. A scribe suggesting billing codes shapes reimbursement for care already delivered; here the assessment instrument, the payment instrument and the quality instrument are the same artifact, so a system populating its items determines all three at once. The vendor markets accuracy and increased reimbursement together and quotes a coding accuracy figure alongside revenue outcomes. Nothing says which direction the system resolves ambiguity in when an OASIS item could defensibly be scored two ways.

The stated control is human in the loop by design, with the software drafting, the clinician reviewing and the clinician signing. For a mandated assessment the clinician attests to, that is the right control.

Scope note: this is now a full electronic record for home health rather than a scribe alone, and the company states it is expanding into skilled nursing and hospice, which carry their own mandated instruments. United States only.

Its coding automation, built for home health, leans defensive. PDGM relevant diagnoses are generated, codes are checked and clean claims prepared automatically, and the QA guardrails are described as catching documentation issues BEFORE they impact reimbursement or compliance, which is downside protection rather than revenue maximization. What is absent is any independent check: nothing here describes a human with coding credentials reviewing before submission. No fairness statement, subgroup analysis or accent disclosure is published.

A full audit trail on every note is a real control, and it is the main point in the product's favor. An audit trail is what makes a later dispute answerable: it lets an organization reconstruct what the system produced, what a clinician changed and when, which is the evidence any correction or complaint would need.

Against that, the accuracy position is weak. The published figure of 95 percent or better arrives with no methodology, reference standard, denominator or definition of what was measured, and it is the only performance number offered. A separate claim that charting time drops by up to 75 percent across agencies is an efficiency measure rather than a correctness one.

Fields for a superseded revision of the standardized home health assessment instrument have been attributed to the product. That is not supported by current materials, and the actual position is different rather than better. No revision is specified anywhere on the company's own pages, which describe OASIS-ready documentation and OASIS assessment items without naming which version is produced. So a buyer cannot tell from published material whether the output matches the revision currently in force.

That matters more here than a missing version number normally would. In home based care the assessment instrument is not documentation: it drives payment grouping and quality reporting, revisions carry hard effective dates, and the agency submitting carries the exposure rather than the vendor generating. Yet for a product whose entire proposition is producing this instrument, neither the revision it produces nor the release cycle for updates is stated.

Nothing published says how quickly instrument updates ship after a regulatory effective date, or what the company commits to if output is behind.

The vendor makes two claims that sit awkwardly together and names no system to support either. One page states that Enzo connects directly with your existing EHR so notes flow without manual re entry; another states that Enzo REPLACES your legacy EHR and every point solution running on top of it. Those are different products with different implementations, and the first comes with no named integration, architecture or certification. Nothing published says which of the two propositions is on offer.

One useful property, no residency detail. Enzo states the scribe works in low connectivity, rural locations and spotty signal, which is a real requirement in home health where visits happen in patients' houses rather than clinics and where most cloud dependent tools simply fail. Delivered across web, Android and Android tablet. No hosting region, residency option or subprocessor detail is stated.

A figure exists but says very little. A software directory populated by the vendor itself lists subscription pricing ranging between $500 and $8,000 a month, a sixteenfold spread with no stated basis, no unit and no tier definition, and no rate card is published on the vendor's own site. A free trial is referenced. The range cannot be budgeted from.

Deliberately narrow and specialized at the level of the regulatory instrument rather than the note template. The product generates start of care and visit documentation structured for the standardized home health assessment, payment grouping relevant diagnoses and care plan goals, with assessment gaps flagged as the note builds rather than caught later in review.

Depth in one segment is the proposition here, and it is a legitimate one. This is not a general scribe with a home health template; the whole platform is built around a single care setting and its regulatory instrument, and the company states it should not be described as a hospital, outpatient or general medical record system.

Two capabilities read as designed from fieldwork rather than from an office. A clinician can photograph a medication label and have it documented automatically, which is how medication reconciliation actually happens in a patient's home. And the product works without connectivity, recording visits offline and syncing later, which matters in rural settings where cloud dependent tools simply fail.

Coverage is one care setting and one clinician type, no language coverage is published, and the assessment revision produced is not named anywhere in published material, which is the central specification in home health.