Machinify

Machinify presents its products as running on what it calls healthcare's AI operating system: a unified data layer joining claims, medical records, contracts and policies; foundation models the company says are trained on billions of healthcare data points to learn coding and payment; task specific agents that execute work independently; and feedback loops that learn from human reviewers and business outcomes. The learned component is real and sits inside every product.

The payment integrity work still stands without it. The products page offers each of the five products (Coordinate, Subrogate, Audit, Pay, Pharmacy) fully managed by the company's own team of experts, the home page leads with over 40 years of experience, and the businesses combined into Machinify sold subrogation, coordination of benefits and complex claim review as services. Remove the models and what remains is a recovery and review operation, a policy map and claims editing.

Autonomy is claimed and oversight is stated as a phrase. The home page describes task specific agents that execute work independently; the products page says the company combines autonomous actions with human expert oversight, that its systems do not just detect problems but fix them, and that Pay edits and reprices claims in real time before payment. A plan can run each product fully managed by Machinify's experts, or pair it with its own analysts and reviewers. Nothing published says which determinations an agent completes alone, what threshold routes a finding to a reviewer, or whether a prepayment edit can reduce a provider's payment with no person reviewing it.

That boundary matters most to the party outside the contract: a provider receives an adjusted payment or a recovery demand, not the reasoning.

Nothing in the architecture is identified, though it is described in general terms. The home page names four layers: a unified data layer joining claims, medical records, contracts and policies; foundation models trained on billions of healthcare data points to learn coding and payment; task specific agents that execute work independently; and feedback loops that learn from human reviewers and business outcomes. The products page adds that the systems are self learning and improve with every claim processed.

No model is named, the training data is not characterized beyond its scale, and no versioning or release practice is published, a gap that weighs more when the company says the models change with every claim. The home page also calls the platform transparent, which can only be tested against the basis given for a finding.

No model provider is named. The home page describes foundation models trained on healthcare data, task specific agents and a unified data layer, without saying whether the models are built in house or adapted from a third party model, or where they are hosted. The privacy notice lists Machinify subsidiaries and service providers as recipients of the personal information it covers but names none, and client data and protected health information are outside its scope. There is no subprocessor list.

The company combines several businesses, and the compliance page says their integration is still under way, so a plan contracting today may have its data handled by more than one subsidiary.

The home page reports 85+ customers including 18 of the top 20 health plans, 270 million member lives across commercial, Medicare and Medicaid, and more than $6 billion in annual cost avoidance and recoveries realized. These are the company's own figures from its own customers, with no method, baseline, period or definition of cost avoidance behind them, and scale of use is not evidence of accuracy. Unnamed customer quotes credit the product with raising recoveries by over 30 percent and eliminating false positives, again with nothing behind them. The products page points to a KLAS First Look report (October 2025) on customer ratings, which measures customer experience rather than the accuracy of findings.

No precision, false positive rate, overturn rate on appeal or audited savings study is published.

The compliance page (updated 29 September 2025) states that Machinify is a business associate to health plans and other covered entities and will not use or disclose protected health information except as the business associate agreement permits or the law requires; the privacy notice (7 October 2025) excludes client data and protected health information from its own scope and places it under the client agreement. That is a general assurance. Nothing published states how long claims data and medical records are kept, whether they are de identified, or whether they train the models.

The training question is live. The home page says the foundation models are trained on billions of healthcare data points and the products page says the systems improve with every claim processed, without saying whose claims, whether one plan's data trains models serving another, or how that sits with agreements the combined businesses signed with their own customers. How an erroneous finding is caught before it reaches a provider is not described either.

Business associate status is stated and supported. The compliance page (updated 29 September 2025) says Machinify serves as a business associate to health plans and other covered entities and will not use or disclose protected health information other than as the business associate agreement permits or the law requires. The privacy notice (7 October 2025) places protected health information processed as a business associate under the client agreement, and says the company acts as business associate for the health plan when it contacts people about subrogation or recovery.

No agreement is published, and neither document says which entity in the group signs it.

The compliance page (updated 29 September 2025) states that Machinify, Inc. and its subsidiaries are HITRUST r2 and/or SOC 2 Type 2 compliant, and that the company is working toward unified compliance standards as it integrates its subsidiaries. These are recognized attestations of how the company operates security, presented as its own.

The scope question is in the sentence itself: 'and/or' does not say which entity or product holds which, and no report, certificate date or trust portal is published.

No device pathway applies and none is claimed. Payment integrity is administrative and financial rather than clinical.

The regulation that does apply is extensive. Claims determinations sit under Centers for Medicare and Medicaid Services payment rules, state insurance regulation, and prompt payment statutes that constrain how long a plan may take and on what basis it may adjust. Government program work adds federal contracting and audit obligations. Recovery after payment carries its own state law constraints. None of that is device law and all of it bears directly on what this product may do.

Machinify publishes statements about how its models are overseen, with no evaluation behind them. The home page calls the platform's AI safe and transparent and says its feedback loops improve continuously from humans in the loop and business outcomes; the products page describes autonomous actions combined with human expert oversight, and systems that are self learning and improve with every claim processed. The company publishes no evaluation method, monitored metric, error or overturn rate, change control or bias review.

The self learning claim is what makes that gap matter. A model retrained on recovery outcomes can change between any two audits, and nothing says who approves a change, what is watched afterward, or whether findings are checked for skew across provider types, specialties, claim types or programs (commercial, Medicare and Medicaid). The party affected is usually a provider who sees an adjustment, not the reasoning. Several states now require a qualified person to make or review AI driven coverage and payment determinations.

Nothing on the home page, products overview, compliance page or privacy notice says what happens when a finding is wrong: no overturn rate on provider appeal, no false positive rate, no remedy or fee reversal when a recovery is overturned, and no warranty attached to a finding. The privacy notice says that when the company contacts people about subrogation or recovery it acts as business associate for the health plan, and directs coverage and reimbursement questions to the plan. The provider or member affected by an adjustment therefore has no published route against Machinify, and whatever recourse the plan has sits in its contract.

An unnamed customer quoted on the home and products pages credits the product with eliminating false positives; no figure, method or period stands behind it.

The systems that matter here are the plan's claims, eligibility and contract systems and the medical records pulled for chart review, not a provider's record system. The products page says the platform integrates with the plan's core systems and works across claims, charts, contracts and eligibility data, but names no claims platform, standard or interface, so there is nothing to verify.

The integration question is also internal: the businesses combined into Machinify each brought their own platforms, and the compliance page says integration of the subsidiaries is still under way.

The privacy notice says the services are controlled and operated from the United States; nothing names a hosting provider or region. The deployment choices the home page describes (insourced, hybrid or fully managed) are operating models for who does the review work, not hosting options.

The compliance page says the company is still integrating its subsidiaries toward unified compliance standards, which leaves open whether client data from the combined businesses sits on one platform or several. For the personal information it covers, the privacy notice allows storage in any country where the company or its service providers have facilities; client data sits outside that notice.

No price is published, but how the product is sold is discoverable. Five modular products (Coordinate, Subrogate, Audit, Pay, Pharmacy) are sold to health plans, and each can be launched fully managed by Machinify's team, in a hybrid model, or insourced with the plan's own analysts and reviewers, which is the main driver of what a plan is buying.

The unit of charge is not stated. Payment integrity is often priced as a share of what is recovered or avoided; whether Machinify prices that way, per product, or on a fixed fee is not published, and it matters on a product that affects whether claims are paid.

Wide across the payer side and absent from the provider side. Capability spans coordination of benefits, subrogation, chart and itemized bill audit, prepayment claim editing and repricing, and pharmacy payment integrity: five products rather than one applied broadly.

The buyer is a health plan, with the company reporting 85+ customers including 18 of the top 20 health plans and 270 million member lives across commercial, Medicare and Medicaid. Coverage is specialty agnostic because claims arrive from everywhere. United States only, which is inherent since payment integrity is a function of a specific reimbursement system.