OmniMD

This is a platform rather than an algorithm, the same position as ModMed, Elation Health, Oracle Health and NextGen. OmniMD is a two decade old EHR, practice management and revenue cycle company; the AI suite arrived on top of that and the vendor says so directly, describing the AI tools as part of the platform rather than add ons negotiated separately or features behind an extra fee tier. The moat is the installed base and the integrated stack, not the model.

One reason to revisit: unlike a pure module, the scribe is also marketed as working with Epic, Cerner, eClinicalWorks and athenahealth, which is closer to the separately saleable position that earned Sunoh.ai an A.

The stated posture is explicit and conservative: doctors review and approve notes, and the vendor writes plainly that AI scribes are designed to assist rather than replace clinical judgment. That is more than several better funded competitors say. Two things weaken it. Notes push directly into the record with no copy and paste, which removes the natural pause that transfer based products preserve.

And the AI Clinician layer goes beyond documentation into risk flags, medication contraindication alerts and care suggestions, for which no confidence threshold, escalation path or accuracy figure is published.

One useful architectural disclosure sits alongside inconsistent performance claims. On the useful side, OmniMD states that all audio processing and note generation occurs within its own HIPAA compliant infrastructure and that no patient audio or clinical data routes through third party consumer AI services. That is a direct answer to the question of whose model it is, and it rules out the consumer API wrapper case even though it does not say what is used instead. Against that, accuracy is claimed at 98.65 percent on one page and 99 percent on another, with no methodology, reference standard or denominator for either, and no model card is published.

One statement here answers whose model it is, and a qualifier is doing considerable work in it. The vendor states that all audio processing and note generation occurs inside its own compliant infrastructure and that no patient audio or clinical data routes through third party consumer artificial intelligence services. That is a direct response to the whose model question and it rules out the worst case, a clinical product quietly forwarding encounters to a general purpose consumer service.

The word to notice is consumer. A commitment that content does not reach consumer services does not exclude enterprise model interfaces, which is very likely where a healthcare vendor would send content if it were sending it anywhere, and those are contractually different products from the same providers rather than a different technology. The claim as written is therefore narrower than it reads. Nothing is named affirmatively either: no model, model family or provider, no hosting arrangement and no subprocessor list is published.

Deployment scale is real at more than 600 clinics, and there is a long operating history behind it, but neither is evidence of benefit. Every published performance figure is vendor generated and several use the unfalsifiable up to construction: documentation time reduced by up to 72 percent, charting accelerated 30 to 40 percent, more than five hours saved weekly, denial rates falling from an average 21 percent to under 4 percent. No study, controlled evaluation, accuracy benchmark or third party assessment is published.

The vendor commits that all audio processing and note generation occurs inside its own HIPAA compliant infrastructure, and specifically that no patient audio or clinical data routes through third party consumer AI services. No retention schedule for audio or transcripts, no deidentification practice and no statement on whether customer data trains models is published.

States that the HIPAA business associate agreement is included at no extra cost, so it is not gated behind a higher tier or a fee.

Two elements are missing. The agreement still has to be executed through a sales conversation and its terms are not published, so what will govern protected health information cannot be read before contracting. And nothing states that its obligations extend to subcontractors, which matters here: the same platform runs documentation, coding, claims and a front desk agent handling calls and insurance verification, so the range of processing covered by one agreement is wide.

The infrastructure commitment sits alongside: the vendor states that audio processing and note generation occur inside its own compliant environment. That describes where content is handled rather than what the vendor has agreed to.

The company states that it holds HITRUST certification and SOC 2 Type II alongside HIPAA compliance. Those are independent examinations rather than self assertions, and it is the right posture for a platform holding production records for a stated twelve thousand clinicians across six hundred facilities.

Three gaps remain. The claim appears in the company's own blog material rather than on a compliance page or trust center, so there is no dated report, no scope statement, and no mechanism through which a counterparty could request evidence. It is written as a compound, which leaves unresolved which certification covers which systems, and for a vendor selling a record system, practice management, revenue cycle services and a scribe, scope is the whole question. And the ONC certification this vendor also holds is a separate matter: that is a functional certification of record system capability required for federal quality programs, a different kind of object from a security attestation, and a compliance list mixing the two reads as more than it contains.

What the vendor publishes on infrastructure is unusually specific. Data is stated to be encrypted in transit and at rest with access logs and audit trails available for compliance review, and audio processing and note generation are stated to occur inside its own infrastructure, with no patient audio or clinical data routed through third-party consumer AI services. That last commitment answers the first question any ambient scribe faces.

No clearance claimed and none required for the documentation component. The base position holds for the scribe itself: it drafts, the clinician reviews and signs.

Beyond the scribe sits a much larger decision support layer. AI Clinician is described as surfacing risk flags, medication contraindications and care suggestions during the encounter, delivering evidence based recommendations at the point of care, surfacing relevant differentials from patient history and symptoms, supporting prescribing through contraindication alerts, anticipating complications and readmissions, and driving context aware ordering of labs, imaging and referrals. Differential generation, contraindication alerting, readmission prediction and order preparation are four distinct decision support functions and none of them is documentation.

The exemption that keeps clinical decision support outside United States device regulation turns partly on whether a professional can independently review the basis of a recommendation rather than relying on it. Nothing published describes what these recommendations are grounded in, how they are sourced, or what the clinician is shown to support them. That question should be put per function rather than about the platform as a whole, because a contraindication alert checked against a medication list and a readmission prediction produced by a model are different objects carrying different review requirements.

Coding is the second boundary and the vendor states it plainly. Documentation is described as billing ready with coding suggestions at the point of care, under the line that what you document is what gets coded. That is the closed loop from encounter to claim, presented as a feature.

A third surface sits alongside both: an AI front desk handling patient calls and intake, which is patient facing rather than clinician facing and carries a different safety posture again.

This is an ONC certified electronic record platform rather than a standalone scribe, so which modules are in contract matters as much as the scribe on its own.

The issue is incentive structure, and here it is tight. Because the EHR, practice management and revenue cycle are one platform, the vendor describes documentation feeding the coding, the coding feeding the claim, and denial data feeding back into the workflow with no manual work between vendors. That is a closed loop from spoken word to submitted claim with no vendor boundary anywhere in it, and coding suggestions are delivered at the point of care rather than after the fact.

The integration is valuable and it also removes every natural checkpoint where a separate party might question a code. No fairness statement, subgroup analysis or performance disclosure by language is published, although multilingual processing is marketed.

No published limitations, and no warranty, indemnity or remediation commitment is published. The accuracy position is undermined by the vendor's own inconsistency.

Accuracy appears as 98.65 percent on one page and 99 percent on another, with no methodology, reference standard or denominator for either. Two different figures for the same product is a worse position than one undefined figure, because it establishes that at least one published number is not being maintained and a buyer has no basis for choosing between them. The decimal places on the lower figure matter too: a number stated to two decimal places implies a measurement precise enough to warrant them, and no measurement is described.

The exposure is wider than a note. The platform also generates coding suggestions, risk flags and medication contraindication alerts, and runs a front desk agent handling calls, scheduling and insurance verification. A contraindication alert that fails to fire is a different kind of wrong from a clumsy sentence in a note, and nothing published distinguishes the two.

The infrastructure commitment on data stewardship is real but does not answer for liability, since keeping processing inside your own environment protects the content rather than standing behind the output.

Deepest inside its own stack, where notes push straight into the record with no copy and paste and documentation flows into billing without reentry. That single vendor path is the real strength here, and it removes a handoff where fidelity can be lost.

Beyond that, compatibility is claimed with Epic, Cerner, eClinicalWorks and athenahealth, which would make it more portable than a pure record system module. That cross platform depth is not documented: no integration architecture, write back mechanism or certification program membership is published for any of the four, so nothing says whether compatibility means an embedded application, a delivered note, or an export.

The distinction matters commercially as well as technically. A practice already running one of those four systems would be buying this as a scribe rather than as a platform, which is the configuration the published material says least about.

Fully cloud based with no on premise component, accessible from any browser, and the vendor states where processing happens: within its own HIPAA compliant infrastructure rather than routed to external AI services. No hosting region, residency option or subprocessor list is published, so the commitment is about which organization processes the data rather than where.

Partial. A seven day free trial is published and the vendor states clearly that the AI tools are part of the platform rather than add ons behind an extra fee tier, which tells a buyer the shape of the commitment, and the BAA is confirmed as carrying no extra cost. What is missing is any number: there is no rate card, tier structure or per provider figure for either the platform or the scribe. Managed billing services are also offered alongside the software, so a full quote may combine license and service fees.

Ambulatory and multi specialty, with specialty tuned templates named across cardiology, orthopedics, ENT, obstetrics and gynecology, internal medicine, radiology, pediatrics and pulmonology, plus urgent care and family practice among documented users. Speaker diarization separates provider from patient voice, multilingual processing is supported, and an AI chart comparison feature summarizes changes between visits side by side, which is a longitudinal capability. No specialty count or language list is published.