Milliman

Several products are models through and through, inside a firm that is not an AI company. Nodal is a pure model product: strip the machine learning and the natural language processing off it and nothing remains, since its entire function is extracting signal from adjuster notes and claim text to predict litigation, severity and excessive cost. The Curv predictive models, the fleet score, the attorney billing algorithms and the fraud detection platform are the same shape. The firm also licenses actuarial computation platforms that carry no inference at all, and its largest business by far is consulting.

The automated path is where oversight matters most, and it is the one point left undescribed. Output is scores, alerts and high, medium and low bands presented to a claims department. The firm's own actuaries, data scientists and claims professionals work alongside implementation and continue after it. People are staffed next to the model rather than placed after it as an adjudication layer.

The product also offers to automate low risk claims so skilled staff can concentrate on complex ones. Nothing published states the threshold for that automation, what review the automated path receives, or what happens when a claim is banded wrongly.

The firm publishes research built on its own models, which is itself a form of transparency. That work includes comparisons of machine learning against traditional mortality models, and an analysis using the claims product that found a condition underreported in workers compensation. One product is described as an explainable platform, with explainability named as a property rather than implied.

Models are tailored to each client's own data and claims can be rescored, so versioning and refit are working features. No model documentation, performance figures, validation report or drift policy is published. Input quality is a live risk. The statutory obligation to follow reasonable procedures for maximum possible accuracy has been the subject of both a federal enforcement action and class litigation alleging mixed files containing other people's records.

A quantified outcome is published: average cost savings of five to fifteen percent for companies deploying the claims product. Named client work exists, including a public risk pool case study. The award for the claims product was chosen by an independent panel of senior industry experts rather than bought. The limit is that the quantified figure is aggregate and self reported, and the named references and the numbers never join.

Independent corroboration exists for the consumer report product: federal court filings name the carriers receiving its reports, which establishes the installed base from a source with no marketing interest.

A shared reference database sits behind every client model, and the data in it is highly sensitive. The claims product states that it tailors models to a client's own claim and text data while leveraging its own database as support. Nothing published says what enters that database, whether contribution is optional, or whether one carrier's claim narratives improve models sold to another. The underlying material is not trades or public filings but individuals' prescription histories, medical claims, adjuster narratives, medical notes and police reports.

A statute supplies the privacy posture, not a policy page. Under FCRA the consumer report business collects under an authorization the applicant signs and may release only for a permissible purpose. It is required to give an individual free access to their own file within a statutory window. It appears on the Consumer Financial Protection Bureau's official list of consumer reporting companies, with the access route published.

No data processing detail, subprocessor list or residency statement is published. The data is highly sensitive, and the accuracy obligations attached to it have been contested in enforcement and in court.

No certification is published in the vendor's own material. A firm handling protected health data at this scale for life underwriting almost certainly has a SOC report and operates as a business associate under HIPAA. The consumer report business runs on a separate domain from the parent firm, which may carry security detail not covered in this profile.

The regulated status attaches to the product itself, not to an adjacent business. The consumer report business operates as a consumer reporting agency under FCRA, supervised by the Federal Trade Commission and the Consumer Financial Protection Bureau, and it appears on the bureau's official published list of such companies. The status comes from operating in a regulated category with supervisory oversight. That is closer to program enrollment than to a license granted after examination. The firm has been the subject of an enforcement action under that statute.

The exposure is direct and nothing is disclosed: no fairness testing, no protected class analysis, no governance disclosure. One model predicts whether a claimant is likely to litigate. That is a prediction about whether a person will assert their rights, and it correlates with income, geography, language and access to representation. A second product line scores insurability from prescription and medical claim history. That tracks health status and disability by construction and includes mental health prescribing, since the firm's own published research uses that data to study depression.

An individual affected by the consumer report product has recourse that is statutory, individually exercisable and published by the vendor. They can obtain their own report free of charge, dispute any item with the vendor directly, and trigger an internal reinvestigation. Where the report is revised, the corrected copy goes both to them and to the entity that originally requested it. Two limits matter.

First, it reaches the underlying data and not the model: there is no route to contest a predicted litigation likelihood or a risk score, only the facts beneath it. Second, the mechanism has a performance record, and it is poor. A federal enforcement action was brought under the statute. Class litigation has alleged mixed files containing other people's prescription records, non disclosure of data sources, and a reinvestigation process that pushes the burden onto the consumer.

The vendor's own published process partly matches that criticism by asking disputants to obtain records from providers themselves. A statute created both the right and the complaint record, which is why this recourse can be assessed at all. Where nobody has standing to sue, an absence of litigation is not evidence of good behavior.

Cutting edge machine learning techniques, powerful algorithms and proprietary language processing, with no model, provider, architecture or version named anywhere. The interesting asymmetry is that the data supply chain is disclosed at category level, since pharmacies, pharmacy benefit managers and medical providers are named as sources, while the specific sources for an individual record were the subject of a non disclosure claim in litigation, and the model supply chain is not disclosed at any level at all.

The consumer report products are pulled inside carrier new business workflows at the point of underwriting, which is about as deep into a core process as a data product gets. Nodal is a fully managed service requiring no client installation, scoring and rescoring claims against a client's own systems of record. The scenario generator is cloud native with an integration ready interface, and the healthcare line ships a data warehousing platform. No named policy administration or claims system connector is published, so integration is described by function rather than by named counterparty.

Described as fully managed software as a service with no client installation, running on hyperscale cloud infrastructure. Nothing published on region, residency, tenancy or hosting arrangements, which is a notable silence for products holding individual prescription and medical claim histories across multiple jurisdictions.

No price, tier, unit or pricing basis is published for any product in the portfolio. Everything routes to a contact form, which is standard for a consultancy led seller.

Coverage is broad across buyers, lines and functions. Buyers include life carriers, health carriers, property and casualty insurers and reinsurers, along with self insured employers, third party administrators and health plans carrying government program risk. Lines run across life, health, long term care, disability and annuities, plus workers compensation, auto liability, general liability, property and flood. Products address underwriting, claims, reserving, pricing, capital and payment integrity, and the firm operates globally rather than in one market.