What Circular Letter 7 Actually Requires

NYDFS issued Circular Letter No. 7 in 2024 to clarify how existing insurance law applies when insurers use artificial intelligence systems and external consumer data and information sources, often shortened to ECDIS, in underwriting and pricing. The circular does not create a new AI statute. It tells insurers that the fair-underwriting and anti-discrimination standards they already operate under apply explicitly to algorithmic decisions and the third-party data feeding them.

Three expectations run through the guidance: governance, testing, and explainability. Insurers need documented oversight from senior management over any model or data source used in underwriting. They need evidence that a model or an external data source does not create a disparate impact on a protected class, even through an unintended proxy variable. And they need the ability to tell a regulator, or a policyholder who was declined or repriced, exactly which data contributed to that outcome.

That last point sets the bar. It is not enough to say an external data source was part of the model somewhere. Insurers must identify the source of information behind a specific decision. Bringing ECDIS into an underwriting pipeline also raises ordinary data security questions about who can access it and how it is stored, on top of the fairness testing the circular focuses on. For a carrier running dozens of ECDIS feeds through a modern underwriting stack, all of this is a considerably higher bar than most model documentation was built to meet.

Why External Data Creates a Traceability Problem

Underwriting and pricing models rarely rely on a single data source. A risk score might blend motor vehicle records, property characteristics, claims history, and newer non-traditional data, run through feature engineering, then scored by a model retrained on a quarterly cycle. Each step transforms the data. Each step also raises a data quality question: is the model still using what the source vendor intended, or a poorly mapped proxy for it?

Most insurers do not run a data lineage tool that tracks that chain automatically. Model documentation gets built once, at deployment, and rarely updates as vendors change fields, models retrain, or new ECDIS sources get added. By the time a regulator asks for evidence, the lineage on file no longer matches the pipeline running in production.

This is not a paperwork problem. Circular Letter 7 assumes insurers can answer why a model reached a decision, and where the inputs came from, on a case-by-case basis. Without automated data lineage connecting source, transformation, model version, and outcome, that question turns into a weeks-long forensic exercise instead of a lookup.

Where Insurer Data Practices Fall Short

A few patterns show up repeatedly in how insurers manage AI and ECDIS governance today, and each one works against what the circular expects.

Alex Solutions built its lineage and governance capabilities around closing exactly this gap.

An Auditable Chain From Source to Decision

Closing this gap does not require replacing existing underwriting models or vendor relationships. It requires connecting them. An auditable chain runs from the original external data source, through every transformation and feature engineering step, into the model version that produced a decision, and back out to the individual case a regulator is asking about.

That chain has to update itself. A data lineage tool that only reflects a pipeline as it looked at deployment is not meaningfully different from a manual document. Metadata orchestration across every source, model, and pipeline change is what keeps the evidence current instead of stale. Analysts including Gartner have pointed to lineage as a foundational requirement for trustworthy AI governance, not an optional add-on.

Governance also has to sit at the point of decision, not only in a policy binder. A model or data source that falls out of compliance should trigger a review before it reaches a real underwriting decision, not after an exam uncovers it.

How Alex Solutions Turns Lineage Into Compliance Evidence

Alex’s Enterprise Data Operations Platform gives insurers the connective layer between raw ECDIS feeds, model pipelines, and underwriting outcomes. Its Knowledge Graph maps every data source, transformation, and model dependency across the enterprise, so a single decision can be traced back through the full chain to the external source that informed it.

Governance and Change Risk Agents evaluate every new data source, model update, or pipeline change against an insurer’s own policy rules, flagging anything that could introduce unfair discrimination or an untracked ECDIS feed before it reaches production. Playbooks turn that evaluation into a repeatable step instead of a one-time review.

This is agentic data operations applied to a regulatory problem. Instead of insurers reconstructing evidence after the fact, Alex Solutions generates it continuously, as a byproduct of how data already moves through the business.

A New York Underwriting Scenario

Consider a New York-licensed personal auto insurer blending a credit-based insurance score, motor vehicle records, and a third-party property-risk index into its pricing model. A policyholder disputes a rate increase, and compliance needs to show, within days, exactly which data point moved the price and where it came from.

With lineage connecting source to feature to model version to outcome, that answer is a lookup, not an investigation. The compliance team can show the regulator which ECDIS feed was involved, when it last updated, and that the model passed its most recent disparate-impact test. Nothing about the underwriting model has to change. The evidence already exists.

Frequently Asked Questions