Arintra

The model is the product. Arintra's platform page calls autonomous coding "the core of the engine": a generative AI engine reads the whole chart when the provider closes the encounter, assigns specialty specific codes and writes them back into the record system. The company reports that 82 to 86 percent of charts go to billing with no person involved, while the rest reach the customer's coders with an explanation.

The other modules, documentation improvement, denial intelligence and appeals, and revenue analytics, are built on the same coding output: the documentation gaps, denial root causes and appeal letters all come from what the coding engine read and cited. Remove the model and nothing is left to sell.

The routing is described: the home page says 82 to 86 percent of charts "flow through automatically, claims out the door", and the rest route to the coders' existing work queues with a full explanation for review. Every code traces to the passage in the note that justifies it, shown inside the record system. The threshold that decides which charts go to a person is not published, and nothing describes review of the charts billed automatically.

Two of the company's claims are ones a review step cannot answer. The home page says "Arintra makes sure every claim is right before it goes out", a claim that every output is correct, and the platform page says the engine's intelligence runs across the revenue cycle "so nothing falls through and every dollar is defended", a completeness claim. Taken at face value, either one removes any reason to review what goes out automatically, and no review shows a code the system never assigned.

Other claims are made as well: "accurately enough to stand up in an audit" and "the same accuracy" wherever care happens, which evidence would have to support, and "smarter every cycle", a learning claim.

The architecture is described in general terms with nothing identified. Arintra calls its product "a GenAI-native engine" and "an agentic platform" that codes the whole chart "with one model" across ambulatory, emergency, diagnostic and inpatient care, and every code traces to the words in the note that justify it. Output scope is described as complete, specialty specific codes written back to the record system, across 23 specialties. The about page adds nothing on the technology beyond coding "every chart explainably".

No model family, foundation model provider, version, training data or update practice is published on the home, platform or about pages, so a buyer cannot tell what the "one model" is, what it was trained on, or when it changes. No accuracy figure is published at all, in aggregate or by code type.

No party in the chain is named. Arintra describes a generative AI engine and "one model" across care settings, without naming a foundation model provider or saying the model is built in house. The website privacy policy says website data sits with third party cloud database services in the United States, unnamed, and does not describe the coding service.

No subprocessor list, hosting provider or model provider is published for the product, and nothing states whether chart text leaves Arintra's own systems for a model provider, or the terms that govern it if it does. The data crossing this chain is the complete clinical note for every encounter coded.

Named deployments with outcome figures, answering revenue and throughput rather than accuracy. The company's case study summary says Mercyhealth, with upwards of 130,000 claims a month, processed more than 112,000 charts through Arintra in five months, cut work queue aging days by more than half and saw a 5.1 percent revenue uplift; the full study sits behind a download form. Other named customers include Med First (over 6 percent revenue uplift, October 2025), Vanova Health, Meritus Health (live within two months), UC Davis Health (audits about 50 percent faster than manual review) and Rochester Regional Health.

Company wide figures, a 5.1 percent increase in compliant revenue capture, 32 percent lower cost, 43 percent fewer coding related denials and 88 percent direct to billing automation, carry no baseline or period, and a 2026 release gives a 64 percent reduction in pre accounts receivable days. The company reports a score of 93 out of 100 in a KLAS Research Emerging Company Spotlight (April 2026), a customer experience measure.

The headline claim is accuracy ("accurately enough to stand up in an audit"), and no accuracy figure, method or reference standard is published, so the evidence answers a different question.

General assurances that do not answer the questions AI raises, plus one retention statement of uncertain reach. The April 2026 certification release frames HITRUST e1 around cybersecurity and information protection, and the chief technology officer says customers trust Arintra with highly sensitive data. The website privacy policy says Arintra retains service data for the duration of the customer's business relationship and a period after, "to analyze the data for our own operations, and for historical and archiving purposes", with no period stated; the policy is scoped to information gathered through the website, so how far that sentence reaches the charts the product codes is not clear.

Nothing states whether customer charts train or tune the models, whether data is de-identified first, or how the engine's coding knowledge is maintained. Safety engineering, meaning how a coding fault is detected and handled, is not addressed. The product reads the complete note for every encounter it codes, and the company says it processes more than $5 billion of claims a year.

HIPAA compliance is claimed and business associate status is not stated. The April 2026 certification release says HITRUST e1 "builds on Arintra's existing SOC 2 and HIPAA compliance posture". The only privacy document published is the website privacy policy (effective 1 November 2022), which covers information gathered through the website.

It mentions service data, saying it is retained for the duration of the customer relationship and for a period after, but it does not address protected health information, business associate obligations or the coding service itself. No agreement, tier or subcontractor position is published, and the footer's data protection link does not lead to a page.

A recognized certification named with its date, without the artifact. Arintra announced HITRUST e1 certification on 17 April 2026, with a HITRUST executive quoted and the assessment described as independent third party testing of implemented security controls. The e1 is the entry tier of HITRUST's three, a foundational control set below the i1 and the risk based r2, and it runs for one year.

The same release refers to an existing SOC 2 and HIPAA compliance posture without a report type, auditor, period or date, so SOC 2 is asserted rather than named as an attestation. No trust center, report request route, penetration testing cadence or vulnerability disclosure policy is linked from the site, and the footer's data protection link leads nowhere.

No device pathway applies and none is claimed. Assigning billing codes from documentation is an administrative determination rather than a clinical one, so the absence of a clearance is correct and not a gap.

The regulatory exposure sits in claims submission rather than device law. Codes submitted to a payer are representations, and where they are wrong the framework is federal false claims enforcement, which lands on the billing provider rather than on the software vendor. The company positions the product around compliant reimbursement and states the audit trail supports compliance and clinician education, which is the right posture, and a posture is not an allocation of liability.

Nothing published addresses how responsibility is apportioned when an automated code is later found incorrect, while the regulatory position itself is represented accurately and without overstatement.

Explainability and parity language with no evaluation behind it. The platform page describes a system that is "transparent, not a black box", with every code traced to the passage in the note that supports it, and says Arintra codes every care setting with one model and "the same accuracy". That is a parity claim across ambulatory, emergency, diagnostic and inpatient settings, and no results by setting or specialty are published to support it. The analytics module reports revenue patterns by specialty, provider and revenue line, and the documentation module surfaces documentation gaps at the organization, specialty and provider level; those are reports about the customer's documentation and revenue, not about how the model behaves.

The governance question specific to coding is systematic drift in assigned evaluation and management levels, which raises or lowers revenue and compliance exposure across a population without showing in any single chart. No distribution of assigned levels against benchmarks, bias testing or model validation method is published. Safety engineering is not addressed.

Mechanisms exist to challenge an output, with nothing standing behind it. Every code traces to the exact words in the note that justify it, shown inside the record system so coders and auditors can review the reasoning line by line, and the company's denial and appeals module uses that audit trail to build appeal letters when a payer denies a claim. A customer quoted on the home page describes pulling up the reasoning behind a code after a denial.

No accuracy figure, error rate, warranty, indemnity, service level or remediation commitment is published on the home, platform, certification or funding pages. The audit trail is therefore a tool for the customer to contest a payer, not a commitment from Arintra when its own code is wrong. Codes on a claim are the provider's representations to a payer, and the false claims exposure stays with the provider.

Named integrations that read the chart and write codes back, listed on the record systems' own marketplaces and running in production. The home page says Arintra is featured in the Epic Toolbox and the athenahealth Marketplace for autonomous medical coding and links both listings (on Epic Showroom and athenaConnect), and names Oracle Cerner, eClinicalWorks, Meditech, NextGen and Allscripts among supported systems. The product picks up the chart when the provider closes the encounter, assigns codes and writes them back into the record system, from which the customer's system submits the claim.

Customers attest to production use: Meritus Health's chief information officer cites the depth of the Epic integration and a go live within two months, and a customer's chief financial officer says the product ran within athenahealth from day one. No interface standard (FHIR or HL7) is stated, and production use on the systems beyond Epic and athenahealth is not shown.

A single hosted service, with location implied rather than committed. The product runs in the back end and writes codes into the customer's record system, which the company describes as working "invisibly" inside Epic, athenahealth and other systems. The website privacy policy says personal data is stored on Arintra's servers and those of cloud database services located in the United States, but that policy covers information gathered through the website, and no hosting provider, region, tenant isolation or on premises option is stated for the coding service.

Complete clinical notes for every coded encounter move to the service for processing, so where they rest is a real question.

The economics are described only in outcomes, and there is no published price. The company says customers see 32 percent lower cost, a 5.1 percent increase in compliant revenue capture and 43 percent fewer coding related denials; those are savings and revenue claims, not the vendor's charge, and no unit of charge, whether per chart, per encounter or per volume band, is stated.

How the product is sold is clear before contact: an enterprise platform with autonomous coding at its core and documentation, denial and analytics modules on top, deployed inside the record system, with a go live of about two months cited by a customer. A demo request is the route in.

Coverage is named with validation behind part of it. The company says it codes ambulatory, emergency, diagnostic and inpatient care with one model across 23 or more specialties, for health systems, academic medical centers and large physician groups. Named customers back the ambulatory and physician group core: Mercyhealth, Meritus Health (now expanding into surgery and inpatient), Rochester Regional Health (which began with one high volume specialty), UC Davis Health, Med First and Vanova Health, an independent physician practice on athenahealth.

Emergency, diagnostic and inpatient coverage carry no named customer or figure on the pages that sell them, and a customer describes surgery and inpatient as areas still to come. Nothing addresses coding regimes outside the United States, which matters because coding systems are national.