C

Covalynt

The models drive the flagship product within a wider data engineering suite. ClaimScore's scoring runs on what the vendor calls a proprietary expert system AI whose criteria are weighted by neural network machine learning trained on cohorts of known valid and invalid claims, and every claim's score and rejection turns on it. The rest of the suite is described as data science and identity resolution rather than model work: ClassResolution's record reconciliation for certification, DeepValidation's contact enrichment for notice, and bespoke data engineering. It would function without the scoring models.

The scoring method and how accuracy is tested are described, with accuracy figures shown only inside an illustrative example. The scoring system page explains that each claim starts at 1,000 points, loses points on more than 65 weighted criteria, is tagged with deduction codes naming each failed criterion, and is rejected below 700. Accuracy is measured through regular control studies, with data split into learning and testing sets.

The ROI page shows 99.5 percent identification of valid claims and 98 percent fraud identification accuracy, but inside a worked example case rather than as a labeled result with a sample, period and method. The Accuracy Explained section was not reached and may say more. For this product the question is the reliability of its own eligibility scoring.

Automated eligibility decisions are described with a threshold and reasons, but no human review position is published. The scoring system marks any claim that falls below 700 points as rejected. The product page describes automated claims processing producing eligibility decisions, delivered in real time to claims systems through the API. Deduction codes give an administrator the reasons for each score, which is a real review point.

Nothing published says whether a person reviews rejected claims, whether rejection is final or a recommendation to the administrator, or whether a claimant flagged in error can cure. These decisions determine who in a class is paid.

Named deployments are published without measured outcomes from them. The case experience page lists about twenty matters with docket numbers and courts, including Lopez v. Apple, Katz-Lacabe v. Oracle, Brooks v. Thomson Reuters, In re Broiler Chicken Antitrust Litigation and Kessler v. Quaker Oats. It names some thirty federal, state and Canadian courts where the vendor has served, and describes its work as court appointed neutral and consultant.

The home and product pages add further matters, including mass torts. None of these is paired with a result, and the ROI page's savings figures come from an illustrative example case, so no outcome is tied to a named matter with a method.

Confidentiality rests on an audited control set, while the only published policy covering the Services permits selling or sharing personal information. The security page says the SOC 2 Type II examination covers the Confidentiality trust services criterion and describes role based access controls. The privacy policy states that it applies to the vendor's class action settlement services. It says that identifiers such as name, postal address and email, along with commercial and online activity information and inferences, will be sold or shared with advertising and analytics partners, and it permits deidentified use to improve the Services.

No customer agreement setting confidentiality terms for class member data is published, and privilege and work product are not addressed.

Nothing published on the product's role relative to legal judgment. The product produces claim eligibility decisions presented as able to withstand court scrutiny, but no statement addresses who is responsible for those determinations, whether they constitute advice to counsel or the court, or what counsel and administrators must verify. The published terms govern the website only.

A testing practice is described, without a governance framework or any disclosure on uneven impact. The scoring system page says accuracy is measured through regular control studies, with criteria weighted by machine learning on known valid and invalid claims and separate learning and testing sets. Nothing names an accountable owner or governance process. Nothing addresses whether fraud criteria such as shared contact attributes or name variations flag some groups of legitimate claimants more than others, which is the bias question a court or class counsel would ask.

Security practices are listed while the data terms are generic and permissive. The security page describes encryption in transit and at rest, role based access, WAF and DDoS protection, backups, disaster recovery, penetration testing, vendor risk management and an incident response framework. The privacy policy keeps personal information as long as reasonably necessary, names no subprocessors beyond categories, sets no incident notification timeline, and permits sale or sharing of personal information with advertising and analytics partners. No retention or deletion commitment for customer claim files is published.

Nothing published on liability for the product. The Terms of Use govern the website and refer product use to a separate license agreement and ordering documents, which are not published; no warranty, indemnity, cap or insurance position for the scoring services is published.

The integration paths are described with what they move, and the documentation could not be read. ClaimScore connects in real time through an API key to an administrator's digital claim form and sends results directly to claims or client management systems. A retro review path takes bulk claim files by secure transfer and returns result files with match and validation fields. The developer page describes API keys per case that keep each case's claim data separate, webhooks returning a score, determination and deduction codes for each claim, usage alerts, validation status updates, a mock server, and sandbox, staging and production environments.

The API reference itself sits behind a login at docs.claimscore.ai, so it could not be read and may say more. No specific claims administration or case management system is named.

Cloud delivery is described without a named host, tenancy or region. The security page refers to trusted cloud partners for redundancy, and the privacy policy says the company and its sites are based in the United States for U.S. users. No hosting provider, data location or tenancy model for claim data is stated.

A named attestation is published with scope and a request route, without auditor or period. The security page states SOC 2 Type II examination by an independent AICPA accredited firm covering the Security, Availability and Confidentiality trust services criteria, with current reports available on request under a mutual NDA. It also lists penetration testing and vulnerability assessment. The auditor and report period are not named.

The model architecture is described as built in house, without versions or change notification. The vendor says it owns its own data and builds its own models, and describes ClaimScore as a proprietary expert system of more than 65 criteria with weights learned by neural network machine learning on cohorts of known valid and invalid claims. No third party model provider is indicated, and no model version or update cadence is published. Nothing commits to telling customers when scoring criteria or weights change during a case.

Cost is described only relative to value, without a rate or unit. The ROI page says the cost of ClaimScore averages under five percent of the amount saved from fraudulent claims regardless of case type, and walks through an example case, but no price, unit or pricing model is published and buying runs through a demo request.

The segment and the litigation stages covered are described with substance, without stated limits. The vendor serves settlement administrators, class counsel, defense counsel and courts across class actions, mass torts and mass arbitrations. It covers class certification data, notice, claims processing and payment, and lists consumer, privacy, antitrust, securities and pharmaceutical matters. It does not state case sizes, claim types or jurisdictions it handles less well.