AvodahMed

The model is the entire product and the company exists to build it. AvodahMed is the digital health division of an AI company whose sibling division builds language translation technology, which is an unusually direct explanation for a multilingual capability that most scribes assert without provenance. This is the counterpart to the C graded on Veradigm Ambient Scribe: same intelligence, but here it belongs to the vendor rather than being licensed in.

Not documented for the direct product, and the earlier assessment was right to refuse the shortcut available here.

Oversight behaviour is described in the distributor's materials for the integrated deployment, where codes arrive as suggestions for provider review. This index does not transfer a claim from an integration partner's materials onto the builder's own record, and the same second search that examined this vendor's privacy architecture found a case where a distributor's characterisation and the builder's own documentation did not align. That is a concrete reason to hold the line rather than a procedural one.

For the direct product nothing was located: no review gate description, no confidence threshold, no acceptance or edit rate, no statement of behaviour on uncertain output, and no description of what is flagged for closer attention.

The question has become more consequential since, because of where this output now goes. The vendor has partnered with an autonomous coding provider positioned downstream of its documentation. Where a scribe's output feeds a process explicitly described as autonomous, the oversight model of the scribe itself carries more weight, since it may be the last point at which a person is positioned to intervene before a claim is produced.

Ask what the review gate looks like when the product is bought direct rather than inside a record system, whether it differs between purchase routes, and where the human sits once the coding partnership is in the path.

No accuracy figure, no model card, no named models, no evaluation methodology. Specialty aware documentation is claimed but neither enumerated nor substantiated. The disclosure asymmetry is worth noting on its own: more is publicly documented about how this model performs by its distribution partner than by the company that built it.

Nothing published by this company identifies any party in its chain: no foundation model provider, model class or version, no hosting arrangement and no sub processor list was located in two passes. The distinctive feature of this record is a disclosure asymmetry that a buyer needs to understand before relying on anything, because this engine reaches most of its users through a record system distributor rather than directly.

More is publicly documented about how this model behaves by the company that resells it than by the company that builds it, and on one point the two accounts do not sit comfortably together. The distributor describes the integrated product as producing real time transcripts without generating audio recordings, which would be a materially better privacy architecture than the category norm.

This company's own release documentation describes work to support longer audio recordings, and its own testing guidance instructs users to verify audio alongside transcription. The reconciliation may be innocent, since the distributor may be describing a configuration specific to its deployment in which audio is discarded at the point of transcription.

The lesson generalises and should be applied wherever this index grades a resold engine: a privacy architecture claim made by a distributor about a product it did not build is a claim about a deployment, not about an engine, until the builder confirms it. Ask the builder directly whether audio is created, where it is held and for how long, and whether customer content trains models.

Better than pure self report, still short of evidence. The most useful figure is partner reported rather than vendor reported: Veradigm's vice president of technology stated that structured pilots run in September and October 2024 saved an average of seven to ten minutes per encounter, which is a specific operational number from the party doing the integrating.

Against that, AvodahMed's own site carries a customer claim of chart closure rates up by nearly 70 percent with no denominator, and one testimonial appears under both brands with only the product name changed, the same six months of consistent use quote running as AvodahMed on one site and as Veradigm Ambient Scribe in partner coverage. Recycled attribution is not evidence of anything except shared marketing. No study, no controlled comparison, no independent evaluation.

The earlier assessment declined to transfer a claim from the distributor's materials onto this vendor's record, and the second search shows that caution was correct.

The distributor described the integrated product as producing real time transcripts without generating audio recordings, which would be a materially better privacy architecture than the category norm. This vendor's own release documentation describes work to support longer audio recordings, and its own testing guidance instructs users to verify audio alongside transcription and workflows. Those two accounts do not sit comfortably together.

The reconciliation may be innocent. The distributor may be describing a configuration specific to its own deployment, where audio is discarded at the point of transcription. But the engine as documented by the company that builds it handles audio recordings, and a buyer purchasing this product outside that one integration should not assume the no recording characterisation applies to them. Nor should a buyer inside that integration assume it without confirming which layer implements it.

That is the substantive lesson and it generalises: a privacy architecture claim made by a distributor about a product it did not build is a claim about a deployment, not about an engine, until the builder confirms it.

The remaining questions are unanswered. Whether audio is created, where it is held and for how long, whether transcripts persist separately, and whether customer content is used to train or improve models. None was located in this vendor's own material. Those are the first questions to put, and the answer should come from the builder rather than from a partner's brochure.

Describes itself directly as a HIPAA compliant ambient scribe on its own materials. Specific business associate agreement terms are not published for inspection before contracting.

No SOC 2 report of either type, no HITRUST certification, no ISO 27001 and no trust centre were located in a second pass. Health privacy compliance is asserted, which is a self description rather than an examination, and there is no certification for that rule in any case.

Two routes a buyer might mistake for assurance should be named, because both are available here and neither is one.

The first is the marketplace. This vendor's product is listed in a major electronic record supplier's application catalogue, which reads as vetting. That catalogue's own notice states that listed applications have either self reported production readiness or, where indicated, been tested through a certification process. Two different things in one list, and the catalogue says so.

The second is the brand. The scribe is sold under an established record system vendor's name, and a buyer may reasonably assume the larger organisation's security posture extends to it. It does not follow. Whatever attestations the distributor holds cover the distributor's own systems, and the encounters are processed by this company.

The absence matters in proportion to position. This vendor sits mid chain, receiving encounter content from a distribution partner and, under a further announced partnership, passing documentation output to an autonomous coding provider. An organisation in that position is exactly where a buyer most needs an independent examination, because it is furthest from the buyer's own contract and closest to the data.

Ask which entity holds a report, of which type and period, and whether its scope covers the engine rather than a partner's estate around it.

No clearance claimed and none required for ambient documentation. No United States device pathway attaches to a note the clinician reviews and signs.

The scope question here runs through a partnership rather than through a feature list, and it is the sharper for it. This vendor has announced a collaboration placing an autonomous medical coding provider downstream of its documentation output, describing the combination as uniting ambient scribe technology with coding automation to remove bottlenecks in billing.

The word doing the work is autonomous. Across most of this category the coding boundary involves suggestions, which presume a human selects among them. Autonomy presumes one does not. A documentation engine feeding a coding system that operates without a human in the selection path produces a claim built from a conversation, and the clinician's signature sits on the note rather than on the code.

That is worth separating carefully, because the coding is the partner's product rather than this vendor's. What this record establishes is that the output of this scribe is designed to flow into an autonomous coding process. A buyer should establish where the human review gate sits in that combined path, whether it exists at all once both products are deployed, and which party is accountable for a code that cannot be traced to a clinician's decision.

Commercial framing points the same way. Customer material cites a large increase in chart closure rates as enabling more rapid reimbursement, which describes the documentation product in revenue terms.

A published policy analysis treats whether better coding reflects better documentation or more intensive coding of the same care as the open question for this category. Autonomy sharpens it, because it removes the person who would otherwise be asked.

This organisation is better placed than almost any vendor in the category to characterise how its documentation performs across languages and accents, and it has published none of it. No fairness statement, no subgroup analysis, no accuracy breakdown by accent, dialect or speaker population, and no evaluation methodology were located.

The absence was already noted as more striking here than elsewhere because the parent company's other division builds language technology. The second search sharpens that considerably, because the sibling division is not a peripheral line of business and its claims are clinical.

That division markets clinical language services across more than sixty languages, describing the purpose as clinically safe communication. So one part of the organisation makes an explicit clinical safety claim about handling many languages, while the part generating clinical documentation says nothing about how its own performance varies across accents, dialects or language backgrounds.

That combination cuts against the vendor rather than for it. The common defence of silence on this axis is that measuring performance across speaker populations is difficult and beyond a small company's reach. An organisation operating a clinical language division across sixty languages cannot make that argument. It holds the linguistic expertise, the evaluation problem is already familiar to it, and the two divisions share a parent and a stated mission about being understood.

The documentation product also serves an integration to which multilingual support has since been added, so the exposure is live rather than theoretical.

This remains the single most valuable disclosure this organisation could publish.

Two passes located no accuracy or error figure, no published limitations, no warranty, indemnity or remediation commitment. Read this alongside the record for the distributor that resells this engine, because together they show what an original equipment manufacturer arrangement does to accountability, and neither side of it is visible from the other. The health system contracts with the distributor, which does not control model behaviour.

This company controls model behaviour and has no relationship with the health system, no obligation to it and no route through which it could be asked anything. Neither party publishes a document that covers the combined product, so a question about what happens when the note is wrong has no single addressee, and a buyer who sends a model questionnaire to its counterparty is sending it to a company that must forward it to one the buyer has no agreement with.

That structure is not unusual in this lane and it is rarely made explicit at the point of sale. The unresolved audio question compounds it: whether recordings are created at all is described differently by the builder and the reseller, so a buyer cannot even establish what artifact exists to be wrong about.

Establish which entity carries the performance representations, whether any supplier warranty passes through the reseller, and get the audio architecture confirmed by the builder rather than the brochure.

Two documented routes with real depth rather than a long list of shallow ones. It is listed on the athenahealth Marketplace, and it holds the deepest possible position inside Veradigm, which selected it specifically for integration depth over lightly integrated alternatives and certified it through the Veradigm Connect programme. That is a meaningful third party judgement about integration quality, made by a party with its own EHR to protect.

No hosting region, residency option or subprocessor list was located.

The more useful finding on this record is the shape of the chain rather than the missing detail, because this vendor sits in the middle of one and most buyers will never see the whole of it.

At one end a practice contracts with its electronic record supplier and buys an ambient scribe carrying that supplier's brand. This company builds and operates that scribe, so it is already a party the practice has probably not contracted with directly. This company has in turn announced a partnership placing an autonomous coding provider downstream of its documentation output. That is at least three organisations touching one encounter, and the practice's paperwork reaches the first of them.

The product is also sold directly, not only through that integration, so the chain differs by purchase route. A direct buyer has a shorter chain and a different set of terms from a buyer acquiring the same engine under another company's brand, and neither can infer their position from the other's.

None of the ordinary questions is answered anywhere: where processing runs, whether a third party model service is involved, what is retained and for how long, or which entity holds what.

Ask for the full chain rather than the immediate counterparty. Specifically, which organisations process or receive encounter content, in what order, under what terms with each other, and what changes if the distribution or coding partnerships end.

No published rate card located for the direct product. Pricing runs through a demo request, and the embedded route is priced inside a Veradigm EHR contract, so neither path exposes a comparable number.

Ambulatory, with specialty aware documentation claimed but no specialty count, no note type list and no published set of supported capture languages. Coverage cannot be compared against competitors in this category that enumerate both.