Infinx

The models carry a core module, and the company also sells the people. The agent platform names three kinds of agent: AI agents built on multiple generative models, large language models and supervised machine learning for reasoning work such as document classification and extraction, prior authorization determination, medical necessity review and a scheduling voice agent; automation agents for repetitive work such as eligibility checks, authorization follow-up and claim status; and human agents, a staff of more than 6,000 specialists onshore and offshore who handle complex claims, coding and exceptions.

Engagements can be platform only, specialists only or a blend. With the models removed, a buyer would still hold rules-based automation, integrations, payer connections and a large outsourced workforce, so the models are the engine of the reasoning work rather than the whole product.

The routing structure is described, and the company also promises the automated work is error free. The structure: repetitive, high-volume work goes to automation agents, reasoning work to AI agents, and exceptions and complex cases to the company's human specialists under a human-in-the-loop model, with the operating system routing exceptions to people.

The promise: the agent platform page says the automation agents, which verify insurance eligibility, submit claims and reconcile payments, 'ensure tasks are completed quickly and without error'. A buyer who believes eligibility checks and claim submissions come back without error will not build the review step that catches a wrong benefit or a wrong claim.

The same page publishes a 98 percent prior authorization accuracy rate, which concedes the rate is not perfect and carries no method or denominator. No threshold at which an agent stops and routes to a specialist, and no share of work completed without a person, is published.

The approach and the infrastructure are named; the models and their versions are not. The company describes its AI agents as built on multiple generative models, large language models and supervised machine learning trained on revenue cycle data, working beside rules-based automation agents inside an operating system with named layers for workflow execution, human-in-the-loop routing, monitoring and auditability.

Its product material names a managed foundation model service from its cloud provider for summarization and workflow understanding, and the platform stack down to Keycloak for identity, MongoDB and Amazon RDS for data and Amazon S3 for files. A 2025 strategic investment in Maverick Medical AI brought an outside autonomous coding engine into the coding offering. What is not published is which models run which agent, their versions, or when the system changes, and nothing separates the coding work done natively from work done by the partner engine.

Substantial partial disclosure. The trust center names Amazon Web Services for product hosting and Microsoft Azure for IT infrastructure, and says their controls are monitored through annual SOC reports or ISO/IEC 27001 certificates. Product material names a managed foundation model service from the cloud provider for summarization and workflow understanding, and the data services (MongoDB, Amazon RDS, Amazon S3).

The coding offering includes an outside engine through the company's 2025 investment in Maverick Medical AI. The chain also runs through people: offshore staff in India and the Philippines may access customer data for permitted work, though the company says data is not stored outside the United States. Missing: the model vendor behind the managed service, a subprocessor list, and the terms that govern data once it reaches the model service or the coding partner.

Named customers and vendor-reported figures, with no method behind them. Case studies name Ivy Rehab (99,000 reverifications in five weeks and a 41 percent reduction in bad debt) and describe a multi-region rehabilitation provider processing more than 200,000 patient access cases in the 2026 reverification season, a physical therapy provider automating more than 100,000 eligibility cases in 60 days, and an ENT group that cut denials.

The agent platform page adds a 98 percent prior authorization accuracy rate, more than 15,000 data entry transactions automated daily and 200 hours saved per provider each year, and the dental page claims 20 percent more revenue and 25 percent lower overhead. The company reports serving more than 900 provider organizations, which is scale of use rather than evidence of benefit. None of the figures carries a baseline, denominator or measurement method on the pages that state them, and none is independently evaluated.

The training position is stated as a categorical commitment. The trust center says customer protected health information is not used for AI model training, that testing and prompt tuning use synthetic and sample documents rather than customer faxes, and that retention is tied to operational needs and the contract.

Offshore staff may access data for permitted work but do not store it outside the United States, AI processing runs on US servers, and local storage on laptops and portable media is prohibited. What is missing is the retention schedule itself and any de-identification position, and an article on the site describes attended automation that learns from staff inputs in payer portals, which sits beside the no-training commitment without explaining what that learning uses.

A stated business associate status, with an independent assessment behind it. The trust center says Infinx operates as a business associate for covered workflows, supports a business associate agreement as part of contracting, and expects to comply with the HIPAA Security, Breach Notification and Privacy Rules.

A third-party auditor has assessed the design and implementation of its HIPAA controls for a defined audit period, covering the security, breach and privacy domains, and that assessment's scope includes locations in the United States, India and the Philippines. The agreement itself is not published, so its terms, the tier it applies at and whether it binds subcontractors cannot be read before contracting. The HITRUST certification is security assurance, separate from business associate status.

Three recognized frameworks are named as the company's own, and the artifacts are shared only in the sales process. A trust center linked from the main navigation lists HITRUST certification (reported as Implemented one-year, framework version 11.5.1, attained April 2026 for the in-scope revenue cycle and patient access platforms and environments), a SOC 2 Type II report covering the Security trust category, ISO/IEC 27001 certification for the information security management system, and a third-party HIPAA controls assessment.

It also describes third-party penetration testing with remediation tracked to closure, regular vulnerability scanning, AES-256 encryption at rest, TLS 1.2 or above in transit, multi-factor authentication and a documented incident process, and states a 99 percent uptime service level for the platform in scope. Security documentation is shared 'during the sales process as part of a qualified evaluation', so nothing can be retrieved without asking.

No device claim is made, and the work is administrative and sits on the provider's side of the payer relationship: eligibility and benefit checks, determining whether a procedure needs prior authorization under payer rules, assembling and following up authorization requests, document intake, coding, billing and denial work.

The nearest boundary is the Medical Necessity agent, offered in beta, which reviews clinical notes against payer guidelines to establish that an imaging procedure is medically necessary before a request goes in. That is coverage documentation for the provider rather than a clinical decision about the patient, but it shapes which orders go forward. Codes and claims are representations to a payer, so errors carry false claims exposure for the billing provider.

Governance is named as architecture and not published as results. The operating system lists human-in-the-loop routing, monitoring, auditability and security controls as distinct layers, and the trust center describes security governance across onshore and offshore locations. Nothing from those layers is disclosed: no exception or escalation rate, no evaluation method for the AI agents, no bias or fairness testing, and no external audit of agent output.

The 98 percent prior authorization accuracy rate on the agent platform page has no method or denominator. Two workflows carry governance weight: denial risk prediction decides which claims get scrutiny before submission, and a model that misjudges risk for particular payers, service lines or patient groups would shift attention invisibly; the Medical Necessity agent decides which imaging requests look supportable.

Mechanisms exist to catch a wrong output, and no fee is shown to be at risk. The platform routes exceptions to the company's human specialists, the operating system names auditability and monitoring layers, and the trust center describes post-incident root cause reports and a 99 percent uptime service level, which answers for availability rather than for results. The 98 percent prior authorization accuracy figure has no method or denominator.

The pricing page offers a contingency fee for recovery programs, where the customer pays an unpublished percentage of the dollars recovered on accounts receivable and denied claims. The same page says pricing 'can include a fixed monthly platform license' depending on the solution, and that annual engagements start at $5,000 a month, without saying that the contingency share is the whole fee on recovery work; a fee paid whatever is recovered would make the contingency a bonus rather than a fee at risk. Eligibility, prior authorization, intake and scheduling work, and every contract on a monthly or per-transaction fee, carry no commitment on results.

Record systems and standards are named; none is verifiable in a marketplace. The integrations menu lists Allscripts, athenahealth, Centricity, Cerner, eClinicalWorks, Epic, eRad, Greenway, ModMed, NextGen, Salesforce and a range of radiology, laboratory and specialty systems, connected through HL7, X12, APIs, robotic process automation and FHIR, and the agent platform page claims two-way compatibility with most record, laboratory and radiology systems, with the document capture agent creating referrals directly in the record system.

On the payer side the company claims more than 2,800 payer, clearinghouse and benefit manager connections, naming Availity, Evicore, AIM, Optum, NIA, Magellan and HealthHelp, with coverage varying by workflow. No marketplace listing, integration documentation or named production deployment on a specific record system is published.

Residency and the cross-border path are stated; isolation is described only as access control. The trust center says customer data is processed and stored in a US-hosted application environment on Amazon Web Services and is not stored outside the United States, AI processing runs on US servers, and IT runs on Microsoft Azure.

It states the people side plainly: offshore staff in India and the Philippines may access data for permitted work without storing it, and onshore-only delivery can be scoped for select workflows. Delivery can be platform only, specialists only or blended. Tenancy is described as users limited to their own tenant through Keycloak, with internal services in private subnets. Each statement is qualified 'for scoped services', and AI processing location is 'confirmed during solution scoping'.

A starting price and the pricing bases are published. The pricing page, linked from the main navigation, says annual engagements start at $5,000 a month and names five ways to price: per transaction (per CPT code, case or page, with volume tiers), dedicated specialists priced by capacity, a subscription retainer for platform access with usage overages, managed services under agreed service levels, and a contingency fee for recovery programs, where the customer pays a percentage of what is recovered.

It lists what drives the quote (workflow complexity and exception rate, integration and reporting needs, volume by work type), says standard support and upgrades are included, and that a platform license fee and implementation charges apply depending on scope. Prior authorization and A/R recovery return calculators are public. No rate card, tier figures or self-serve route are published, and the contingency percentage is not stated.

Its patient access work covers eligibility and benefits verification, insurance discovery, patient pay estimates, prior authorization, referral and document intake that reads faxed or scanned documents and creates the referral in the record system, and patient scheduling by voice agent and contact center.

Coverage is named broadly and specifically: acute care, rural hospitals, physician groups, ambulatory surgery centers, long-term care pharmacies and dental practices, and more than thirty specialties from radiology, the company's origin, through cardiology, pathology, orthopedics, physical therapy, behavioral health and specialty pharmacy. Validation sits behind part of it, mostly in rehabilitation and physical therapy (Ivy Rehab and two unnamed providers) and an ENT group. Most specialties on the list carry no case study, and nothing addresses work outside the United States.