RapidClaims

The models do the work people previously did, across four named products. RapidCode assigns the codes, RapidVBC handles risk adjustment and documentation, RapidRecovery works denials and appeals, and the rules engine governs all of it. Remove the models and no workflow layer, network or platform remains that a customer would pay for.

The company's own framing is autonomous agents that code charts, improve documentation, prevent denials and recover revenue. Human escalation exists for low confidence work, which is the correct design in this category rather than a dilution of it. Note for the record that this vendor also markets coding outsourcing prominently in its content, and the screening pass established the platform as the business rather than the delivery model, with independent profiling across four databases describing software. That ruling is what admitted this record and it bears on this axis directly.

The oversight half of this axis is well answered and the autonomy half is not.

On oversight, governed autonomy is an explicit product principle rather than a caveat: the company states it is autonomous where it scales and human where it matters, pairs that with named human escalation for low confidence charts, rule based validation, an explainable audit trail, and a configurable rules engine through which the customer sets the workflows and payer logic the models operate under. A buyer therefore controls the operating envelope.

On autonomy the number is missing. Accuracy is published at 98 percent, throughput at over a thousand charts per minute, and clean claim rate as a percentage, but no direct to bill or automation rate appears anywhere, so a buyer cannot tell what share of charts reaches billing untouched. That is the inverse of Arintra, which publishes the automation share and no accuracy figure, and both gaps leave the same question open from opposite directions.

The validation mechanism is the strongest in this lane. The company offers a fourteen day proof of concept run side by side against the customer's own last six months, with no production cutover and an explicit walk away if it fails to beat the existing baseline. That is a performance conditional entry rather than a free trial, and it lets a buyer establish the missing automation figure on their own data before committing.

The mechanism is described at the level of approach and the calibration claim is specific enough to test. Pre trained models are customised per customer, trained across more than 25 specialties and, per the company, millions of encounters. Alongside them sits a named rules engine maintaining current payer policy, denial patterns and regulatory changes, and rule based validation is listed as a distinct capability from the learned components. Explainable output is a stated design principle across coding, appeals and denial logic.

The differentiating disclosure is calibration cost: 500 sample charts to customise the models against a stated industry norm of 10,000 or more, six weeks to production. That is a falsifiable operational claim rather than an adjective, and the proof of concept gives a buyer a way to check it.

What is absent is everything under the word proprietary. No foundation model, model class or version is named despite large language models being cited in the company's own material, no training data provenance beyond the encounter count, and no accuracy breakdown by coding element. The single 98 percent figure covers evaluation and management levels and procedure codes together, and those are not equivalent judgements.

Pre trained models are acknowledged as the starting point, which concedes that something upstream exists, and nothing about it is named. No foundation model provider, model class or version, no hosting arrangement, no cloud platform, and no sub processor list was located.

The training position is the specific gap here, and it is more consequential for this vendor than for most. The product is customised on the customer's own charts, which the company presents as its central advantage, so customer documentation demonstrably enters model preparation. Nothing published states whether that customisation stays isolated to the customer's instance or contributes to the shared pre trained base, whether data is de identified first, or whether permission is sought. CombineHealth answers all three questions plainly; this vendor answers none of them while running a workflow that raises them more directly.

A rules engine ingesting current payer policy implies external policy data feeds that are also unnamed. Ask for the base model, the sub processor list, whether customer charts used for calibration are isolated from the shared model, and where the payer policy data originates.

The broadest third party recognition in this lane, though of the ranking and award type rather than the measurement type. Three independent bodies are cited: a 2026 first place ranking in artificial intelligence powered claims automation with a published score of 9.70 out of 10 from a buyer survey organisation, a 2025 technology innovation leadership recognition from a research firm, and an outperformer placement in an analyst matrix for automated and assisted coding assessed against fifteen companies including much larger vendors.

Customer logos include a major academic health system alongside a urology group, a regional hospital, a neighbourhood health centre and others. Quantified references cover a 30 percent reduction in accounts receivable days unlocking 2.5 million dollars in accelerated cash flow, a 5 percent revenue increase with coding costs down 40 percent, and risk adjustment factor scores improved 15 percent.

Two limits hold this at B. Every quantified reference is anonymised by role and organisation type rather than attributed to a named person at a named institution, which is weaker than XpertDox's attributed quotes. And no measurement study exists: no peer reviewed publication, no independently audited accuracy assessment, and no parallel coding study of the kind CombineHealth published. Buyer surveys and innovation awards reflect satisfaction and analyst judgement, not verified performance.

Two protection facts are published and the stewardship questions are untouched. Encryption is stated at 256 bit covering data at rest and in transit, and an independently audited controls report covers the security programme in general. Both are real.

Nothing addresses how this product handles the data it ingests. There is no retention schedule, no statement on whether customer documentation contributes to model development, no de identification position, no data ownership or deletion statement, no access control detail, no data minimisation commitment, and no incident or breach disclosure.

The omission matters more here than for a batch coding vendor because customisation on customer charts is the company's headline differentiator. A model customised on 500 of a customer's own charts has ingested that customer's documentation into model preparation by design, and the absence of any statement about isolation, consent or de identification leaves the most basic question about the product's central feature unanswered.

Held at C rather than lower because encryption and an audited controls report are genuine, and rather than higher because every question specific to stewardship is open. The contrast within this lane is instructive: CombineHealth answers all of them on a dedicated page.

Health privacy appears as an assertion with no substance behind it. The security block states full adherence to the framework and stops there. No control enumeration, no de identification position, no access control description, and nothing comparable to the itemised programme CombineHealth publishes.

The health specific credential problem is more serious and is treated in full on the security certifications axis. In short, a recognised healthcare security framework is displayed as a badge with the caption stating alignment rather than certification. Alignment is not certification, that framework awards certification at three named tiers, and no tier, assessor or date accompanies the badge.

What is real underneath is a general independently audited controls report and stated encryption at rest and in transit, both of which overlap substantially with the health security rule. So the controls are externally assessed and not against a health specific framework.

No business associate agreement posture, template, negotiation stance or execution requirement was located, despite the company contracting with health systems and an academic medical centre. Graded C because real assessed controls sit underneath an unevidenced health specific claim.

One credential is properly disclosed and one badge overstates what is held, and a buyer scanning the security block will not distinguish them.

The real one is a controls report specified at the second type, the one covering operating effectiveness over a period rather than design at a point in time, described as independently audited. Type plus an assurance statement passes the credential test. Encryption at 256 bit at rest and in transit sits alongside it.

The problem is the healthcare security framework badge. It is displayed in the same row, at the same size, in the same style as the audited credential, and its caption states alignment with the framework rather than certification against it. That framework issues certification at three named tiers through authorised assessors, and two competitors in this lane hold it and state their tier. Alignment is a self description that requires no assessor and confers no assurance. Presented as a badge it reads as a credential, and it is not one. This is recorded as a disclosure quality finding rather than an accusation: the underlying controls may well merit certification and none has been evidenced.

No trust center exists as a standing page, and no report availability process, audit period, auditor, certification date, penetration testing disclosure or vulnerability disclosure policy was located. Graded B on the strength of the audited report, held below A by the missing trust center and the overstated badge.

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 is not a gap in the record.

The exposure sits in claims submission, where codes are representations to a payer and error is governed by federal false claims enforcement, landing on the billing provider rather than the software vendor. The company's rules engine tracking current payer policy and regulatory changes is aimed at that risk and is the right posture.

One product line carries regulatory weight the company does not address. Risk adjustment coding that improves risk adjustment factor scores and closes hierarchical condition category gaps operates in the area of federal programme integrity that has drawn the most sustained enforcement attention in recent years, where the question is whether captured conditions are supported by documentation or generated by suspecting logic. The company advertises a percentage improvement in capture as an outcome. Nothing published describes the documentation standard, provider attestation or audit controls governing suspected conditions. Graded C because the regulatory position is otherwise correctly represented.

The case level instruments are real. An explainable audit trail runs under every coding decision and appeal, the company claims fully audit ready output, rule based validation is a distinct layer, and dashboards surface denial trends, productivity and payer behaviour in real time. A reviewer can therefore interrogate any individual assignment.

What is missing is the population view, and this vendor needs it more than most because of what it advertises. Two headline outcome metrics run in exactly the direction the governance question concerns: a stated percentage lift in evaluation and management levels, and a stated percentage improvement in hierarchical condition category capture with risk adjustment scores up 15 percent at a named customer type. Both are the intended commercial effect. Both are also indistinguishable from drift when viewed from outside.

No instrument is published that would separate them. There is no distribution of assigned billing levels against an expected benchmark, no breakdown by specialty, payer or physician, no external audit of coded output, and no bias, fairness or validation methodology. Arintra and XpertDox both publish the distribution reporting that makes this checkable; this vendor publishes the lift figure without the monitoring. Graded C rather than B for that gap, and rather than D because the audit trail and rules governance are genuine. Ask what evaluation and management distribution looks like before and after, and who audits it.

A stated performance level exists and the pre purchase risk transfer is the best in this lane, and neither becomes recourse once the contract starts.

Accuracy is published at 98 percent with a clean claim rate and throughput alongside it, so there is a figure to hold the vendor to. The fourteen day proof of concept is genuinely unusual: it runs against the customer's own last six months with no production cutover, and the company states that if it does not beat the existing baseline the buyer walks. That places evaluation risk on the vendor and lets a buyer establish real performance on their own data rather than accepting a published number, which is stronger than XpertDox's free first month because it carries an explicit performance condition.

After go live nothing carries over. No service level agreement, warranty, indemnity or remediation commitment was located. No confidence threshold governing autonomous handling is disclosed, no accuracy breakdown by coding element exists, and no denial or reversal rate for assigned codes is published as a vendor figure.

The allocation question is sharpest on the risk adjustment product, where captured conditions later found unsupported create programme integrity exposure that lands on the provider. Ask what contractual commitment attaches to the 98 percent figure after the proof of concept ends, and how liability is apportioned on a suspected condition that fails audit.

The strongest standards position in this lane. The company states the platform is natively built on the modern healthcare interoperability standard, with bidirectional synchronisation, support for the older messaging standard, and an interface first architecture. No other vendor in this category claims native support for that standard, and the distinction matters: it is an architectural statement about how the product connects, not a description of bespoke work done per customer.

Ten record and practice management systems are named, covering both dominant enterprise vendors and a substantial ambulatory tail. One of them serves community and rural hospitals and is named by no other vendor in this lane, which matters given the company's stated focus on federally qualified health centers, community health centers and critical access facilities.

Bidirectional synchronisation with writeback into the record system is stated explicitly for the risk adjustment product, so the round trip is addressed rather than assumed.

Graded A on the strength of the native standard claim, four named mechanisms, ten named systems and stated bidirectionality. What would make it unimpeachable is what Arintra and XpertDox have and this vendor lacks: a listing in a record system vendor's own programme or marketplace, which converts an integration claim into an externally verifiable credential. Ask which integrations are certified by the record vendor and which are direct.

The hosting position is undescribed. No cloud provider, region, residency commitment, tenancy model or customer controlled deployment option was located across a dedicated pass of the published material.

The one adjacent fact is encryption at 256 bit covering data at rest and in transit, which establishes that data is stored somewhere under the vendor's control and says nothing about where. The company operates from New York, with engineering presence reported in India by third party profiling, and nothing published addresses whether documentation or support access crosses a border.

The volume claim sharpens the question. Processing over a thousand charts per minute means substantial quantities of clinical documentation move for processing continuously, and the customer segments the company targets most explicitly, federally qualified health centers and critical access facilities, are among the most constrained on data handling obligations relative to their technical capacity to evaluate them.

Ask for the processing and storage regions, the tenancy model, and whether personnel outside the United States hold access to production data.

No price is published and a buyer can nonetheless model the economics further here than anywhere else in this lane, because the company provides the instrument rather than the number.

A public return on investment calculator specific to the coding product lets a prospective customer enter their own volume and derive an estimate without contacting sales. Alongside it sit a stated 70 percent lower cost than manual coding, a customer reported 40 percent reduction in coding costs, return on investment inside 30 days, six weeks to production and a 500 chart calibration requirement. Each of those is a cost driver a buyer can check against their own operation.

The entry terms are the strongest part. The fourteen day proof of concept runs on the customer's last six months with no production cutover and an explicit walk away if the existing baseline is not beaten, which places the performance risk of evaluation on the vendor.

Held at B rather than A because the unit of charge is never stated. Whether the fee is per chart, per claim, per provider or a volume band is unknown, and that determines how cost behaves as volume grows, which is precisely what a calculator is supposed to resolve. Ask for the pricing mechanism, the minimum commitment, and how the four products are priced relative to one another.

The most complete coverage disclosure in the autonomous coding lane, and it is differentiated in the opposite direction from its competitors.

Most vendors in this category concentrate where charts are high volume and documentation is repetitive. This company states the reverse explicitly and names the hard cases: multi modality oncology with infusion, radiation and drug administration coding; neurosurgery covering spine, cranial and neuromodulation with complex modifier application; interventional cardiology including catheterisation, device coding, bundling rule navigation and structural heart; and complex orthopedics across joint replacement, trauma, fusion and sports medicine. Naming the specific coding difficulties within each specialty is a substantive claim rather than a list.

Buyer segments are documented individually with dedicated material for physician groups, hospitals and health systems, accountable care organisations, federally qualified health centers, ambulatory surgery centers, community health centers and management services organisations. Value based coverage is a distinct product spanning concurrent, prospective and retrospective review, hierarchical condition category suspecting, risk adjustment gap closure, and quality measure capture.

Graded A partly for the boundary the company draws on itself: facility and inpatient coding is stated to be in a design partner programme rather than generally available. Disclosing what is not ready is what makes the rest of the claim usable. The claims outrun evidence in the complex specialties, where no reference is attributed to a named institution, and nothing addresses coding regimes outside the United States.