RapidClaims

The model is the engine of the core module, inside a platform that carries other value. RapidCode's coding agent reads the encounter's documents and assigns ICD, CPT and evaluation and management codes, and RapidVBC and RapidRecovery apply similar agents to risk adjustment and denials. Around them sit parts that are not learned: the evaluation and management level is set by an "E&M MDM Solver" running more than 10,000 rules, RapidRules maintains payer rule sets and customer built rules, claim scrubbing checks NCCI, modifier and bundling edits, a same day billing module files claims, claim resubmission is driven by robotic process automation, and dashboards report denial trends, productivity and payer behavior. The company calls the whole "the OS for your mid revenue cycle".

Without the model, coding assignment and documentation review would stop, but the rules engine, scrubber, billing module and analytics would still be products a revenue cycle team pays for.

The oversight structure is described and the threshold is missing. The home page states the principle as "Autonomous where it scales. Human where it matters", lists human in the loop escalation among RapidCode's capabilities, and describes confidence scored decisions, rule based validation and an explainable trail from each code to its evidence. Customers set the operating envelope through RapidRules, building custom rule sets for their own claim types and payer requirements.

What is not published is the confidence level at which a chart goes to a person, or the share of charts that reach billing untouched: the company publishes 98 percent accuracy, a 98 percent clean claim rate and 1,000 charts a minute, but no automation or direct to bill rate. The 14 day proof of concept on the customer's last six months lets a buyer measure that share before buying.

The company's marketing claims: "Capture every dollar of earned revenue" (a revenue outcome); "Every claim is pre-validated against known denial triggers" and "Nothing skipped" at ingestion (process); "100% Guidelines Coverage" and "100% Audit-ready" (basis claims backed by the code to evidence mapping); "consistent, defensible E&M levels" (a precision claim).

The approach is named without the version and update practice behind it. RapidClaims describes pre-trained models, trained on millions of encounters across more than 25 specialties, customized to each customer with about 500 of its charts ("few-shot learning"), against a stated industry norm of 10,000 or more, and six weeks to production. The coding agent is described as clinical reasoning across diagnoses, procedures, documentation and payer requirements rather than keyword matching, with confidence scored decisions. Beside the learned parts sit named rule components: the E&M MDM Solver with more than 10,000 rules that sets evaluation and management levels, and RapidRules, which mines payer rule sets and adapts "from outcomes".

No foundation model, model class or version is named, training data provenance goes no further than the encounter count, and nothing says when models or rules change or how customers are told. The 98 percent accuracy figure covers all code types together.

A chain the vendor states is short: the privacy policy (last updated 5 June 2026) says client data is "never shared with any third party", is hosted on AWS servers in the United States inside a private virtual network, and that no protected health information enters its models. Read together, that describes a deployment where chart data does not travel to an outside model provider.

The parties are still not enumerated. The pre-trained models are not attributed: nothing says whether they are built in house or adapted from a third party foundation model, or whether any model runs outside RapidClaims' own environment. No subprocessor list is published, AWS is the only party named, and RapidRules' payer policy data feeds are unattributed.

Vendor reported percentages and anonymous testimonials, with no method behind them. The product pages publish 98 percent coding accuracy, a 98 percent clean claim rate, 70 percent cost reduction and 1,000 charts a minute, and the home page reports steady state outcomes of 27 percent fewer denials and 5 percent more revenue. Customer results are attributed by role and organization type only: a health system chief financial officer (A/R days down 30 percent in a quarter, $2.5 million in accelerated cash), a federally qualified health center's HIM director (revenue up 5 percent, coding costs down 40 percent), an accountable care organization's medical director (risk adjustment scores up 15 percent), and a chief medical officer named as Brian Jonson at an unnamed health system (22 percent fewer readmissions and $2.2 million in additional reimbursement with RapidVBC). None names the institution.

Outside recognition is of the award and survey kind: Black Book ranked it first for AI powered claims automation in 2026 (9.70 of 10, a buyer survey), and Frost & Sullivan named it a 2025 technology innovation leader. Neither measures coding accuracy. The headline claim is 98 percent accuracy, and no audit method, sample or reference standard is published for it.

Categorical commitments are published without a retention schedule behind them. The privacy policy (last updated 5 June 2026), which covers chart data clients upload to the portal, says "No PHI data enters our models", that patient chart data is collected exclusively for billing, that data is never shared with any third party and only concerned staff have access, and that uploaded data cannot be downloaded again, only chart identifiers with their ICD and CPT codes. Data is encrypted at 256 bits at rest and in transit.

The commitment on models sits beside a product claim that needs reconciling: the home page says RapidClaims needs just 500 charts to customize its pre-trained models to a customer. Whether those charts are de-identified first, or whether the policy means that protected information never trains the shared models, is not stated. No retention period for uploaded charts, codes or logs is published, and nothing describes de-identification. Safety engineering, meaning how a model fault is detected and handled, is not addressed.

A privacy document reaches the service and no business associate status is stated. The privacy policy (last updated 5 June 2026) describes the treatment of chart data clients upload to the portal: US hosting, collection for billing only, no sharing with third parties, role based access, an incident response plan and download restrictions. Nowhere does it or any page say RapidClaims acts as a business associate, publish an agreement, or state a tier.

The HIPAA statements disagree with each other. The security block on every page says "Full adherence" to HIPAA, while the privacy policy says the company is "Currently holding a provisional HIPAA certificate, with full certification expected shortly"; HIPAA itself has no certification regime, so the certificate is a third party's attestation, unnamed.

One recognized attestation named with its type, one badge that is not a certification. The security block on every page lists SOC 2 Type II, "Independently audited security controls", and the privacy policy says the company is "SOC 2 and ISO certified" without naming the ISO standard. The same block shows a HITRUST badge captioned "Aligned with healthcare's most rigorous framework": alignment is a self description, not a HITRUST certification, which is issued at named tiers by authorized assessors.

No report date, audit period, auditor, trust center, report request route, penetration testing cadence or vulnerability disclosure policy is published. The privacy policy names a chief information security officer as the compliance contact.

No device claim is made and the products are scoped as revenue cycle work. Assigning billing codes from documentation is an administrative determination, so no clearance is expected and none is claimed.

The exposure sits in claims submission and in risk adjustment. Codes on a claim are representations to a payer, and errors fall under federal false claims law, landing on the billing provider. RapidVBC carries a sharper version: it offers HCC suspecting and risk adjustment gap closure, and the home page says it surfaces "missing diagnoses and quality opportunities at point of care". Each coding decision is said to map to the clinical evidence behind it, which gives the clinician something to check, but no documentation standard, provider attestation step or audit control for suspected conditions is published.

Case level explainability and rules governance, with no evaluation of model behavior published. Every coding decision is said to map back to the clinical evidence, guideline and rule that supports it, decisions carry confidence scores, evaluation and management levels come from a rule based solver described as producing "consistent, defensible" levels, customers can build their own rule sets, and the platform is described as "adaptive learning from outcomes". The home page describes the approach as "governed autonomy".

The population view is missing, and this vendor advertises the outcomes that make it matter: RapidVBC's proof points include a percentage lift in evaluation and management levels and a percentage improvement in HCC capture, and an accountable care organization's medical director is quoted on risk adjustment scores up 15 percent. Those are the intended commercial effect, and from outside they are indistinguishable from drift. No distribution of assigned levels against benchmarks, breakdown by specialty, payer or physician, external audit of coded output, or bias or validation method is published. Safety engineering is not addressed.

Mechanisms to challenge an output exist, and the one risk transfer published ends before the contract starts. Every coding decision maps back to the evidence, guideline and rule behind it, the platform keeps audit trails, and RapidRecovery builds appeals with "explainable appeal logic". Accuracy is published as 98 percent with a 98 percent clean claim rate, with no study, sample, period or reference standard behind either.

The entry offer is a 14 day proof of concept run side by side on the customer's last six months of data with no production cutover, on the stated terms "If we don't beat your baseline, you walk." That lets a buyer test performance on its own charts before buying, and it binds nothing once the contract starts. No service level, warranty, indemnity or remediation commitment for wrong codes after go-live is published. The allocation question is sharpest on RapidVBC, where HCCs captured and later found unsupported create program integrity exposure that lands on the provider.

A long list of record systems and standards is claimed, with no published evidence behind it. The home and RapidCode pages say RapidClaims integrates with all major record systems and show Epic, Cerner, athenahealth, eClinicalWorks, NextGen, ModMed, Meditech, Greenway, DrChrono and Allscripts, and list FHIR native design, bidirectional sync, HL7 support and an API first architecture. RapidCode's ingest step reads every document tied to the encounter through "bi-directional EHR integration", and RapidVBC lists EHR writeback.

There is no record system marketplace listing, no integration or API documentation, and no customer named on any system (customer quotes carry roles and organization types only), so nothing shows the integrations running in production. The names and standards amount to a claim of reach.

A single hosted service with residency committed. The privacy policy (last updated 5 June 2026), which covers chart data clients upload to the portal, says all data is hosted on AWS cloud servers located in the United States and that "All patient data is stored exclusively within US borders". Services run inside a private virtual network behind firewalls and intrusion detection, with role based access.

What stays open is isolation and the processing path: nothing says whether each client's data sits in a separate environment, whether any processing (a model call, support access or engineering work) happens outside the United States while data is stored within it, or where backups sit.

No price or pricing basis is published, and the buying terms are discoverable. The public ROI calculator is a "revenue leak calculator": it asks for the buyer's monthly encounter volume, coder headcount, clean claim rate and reimbursement per claim, states "No RapidClaims math yet", and sends the full breakdown only after a contact form. It sizes the buyer's savings, not RapidClaims' fee, and no fee input or unit of charge appears in it.

What is published is the shape of the sale: a 14 day side by side proof of concept on the last six months of data with a walk away if the baseline is not beaten, six weeks to production, 500 charts to customize the models, and a stated 70 percent lower cost than manual coding. Four products (RapidCode, RapidVBC, RapidRecovery and the RapidRCM platform) and RCM services are offered, with no statement of how they are priced together.

Coverage is named in detail, with validation behind part of it. The company names more than 25 specialties and singles out hard cases: multi-modality oncology with infusion, radiation and drug administration; neurosurgery across spine, cranial and neuromodulation; interventional cardiology with catheterization, device and bundling rules; and complex orthopedics. It sells to multi-specialty physician groups, hospitals and health systems, accountable care organizations, federally qualified health centers, community health centers, ambulatory surgery centers, management services organizations, and dental and behavioral health groups, with value based care work in RapidVBC. It also draws its own boundary: facility and inpatient coding is being built with design partners and is not yet generally available.

Validation reaches part of this. The customer results published come from a health system, a federally qualified health center and an accountable care organization, each unnamed, and none addresses the complex specialties the company leads with. Nothing addresses coding regimes outside the United States.