How a Century-Old Mutual Insurer Governs Faster Than It Grows

At a glance

Company	A major mutual life insurer in the United States, serving customers through roughly 12,000 agents and advisors.
Challenge	Onboarding a single dataset took weeks and sometimes up to two months, because no one could say with confidence which columns held personally identifiable or protected health information.
Solution	Atlan as the governed metadata layer across a medallion architecture on AWS, connected over AWS PrivateLink, with column-level classification driving access policy.
Result	Lineage coverage rose from 45% to 52% while the cataloged estate grew more than 90%. Governed business terms grew from roughly 4,700 to 8,200.
Outcome	Sensitive data is identified automatically, so new datasets no longer wait on someone to read every column.

---

A century of policyholder data

The company has written life insurance policies for well over a century. As a mutual insurer, it answers to policyholders rather than shareholders, and it has been collecting data about them since before the telephone was in common use.

For most of that time, one central team ran governance. As the organization changed shape, the team moved to a federated model: governance runs across value streams, and accountability sits with the business. The model fit the organization better. The existing tooling didn’t. The team had built plenty of governance content, and few people used it.

Two months to onboard a dataset, because no one knew where the sensitive data was

Spreadsheets went back and forth between the business and the governance team as they cataloged each dataset to find where protected health information and personally identifiable information sat. Every new dataset waited weeks for that work, and some waited two months.

Audit added a second requirement. The company had to prove that personally identifiable information stays protected when data moves between cloud environments.

An easy button for users

The platform had to fit the federated model. It had to align with the data engineering stack, because data engineering was leading the wider platform review. And after years of governance content nobody opened, it had to be easy to use. The CDAO set the bar in one line: find an easy button for users. Make data easy to find, and easy to use.

Metadata crosses the boundary. The data does not.

The architecture is a medallion flow. Raw data lands in Amazon S3, is cataloged and processed in AWS Glue, and is served from Amazon Redshift. Atlan connects across all of it over AWS PrivateLink. Metadata, lineage, query logs and policy cross the trust boundary on an outbound connection that Atlan initiates. The data itself stays where it is.

The estate also runs AWS Batch for pipeline orchestration, along with Amazon Aurora and Amazon MSK, and governance is most complete on these services. AWS Batch carries lineage on roughly 97% of its 116,000 assets, and Aurora on effectively all of its 10,000.

Classification works at the column level, and the tags drive access policy instead of sitting in documentation. A column tagged sensitive is governed wherever it appears, including in columns created after the policy was written. On an estate growing this fast, a policy covering only what existed on the day it was written falls further behind every week. Lineage supplies the traceability the audit requirement depends on: where each piece of data came from, and every transformation applied as it moves between environments.

The AWS Glue implementation went live in about 30 days.

Governing faster than the estate grows

In April 2026, about 4.5 million of 10 million assets carried lineage. By October, roughly 9.9 million of 19.2 million did. The estate grew more than 90% in six months, and coverage still rose. The number of assets carrying lineage more than doubled.

45% → 52%	Lineage coverage while the estate grew more than 90%, April to October 2026
~4,700 → ~8,200	Governed business terms over the same six months

Automated classification and lineage kept pace with that growth. Manual curation didn’t. On an estate growing this fast, any governance step that needs a person to write something for each asset falls behind. Build those steps so people review what automation produces, from day one.

Until the sensitive fields in a dataset are identified, nothing in it can be released to anyone. Coverage is how quickly that happens across 19 million assets. The CDAO’s verdict on the old cycle: if it takes two months, there is no value there.

What the foundation now carries

With the governed layer in place, the team has taken on work it couldn’t have scoped before:

• a production agentic use case served through Atlan’s managed MCP server
• documentation generated automatically from governed metadata
• an expansion into the actuarial function, where pricing lineage and model traceability carry regulatory weight

Where it leaves them

Governance, as the governance lead puts it, helps people find the right data, trust it, and use it at the right time for the right purpose, so the business reaches insight faster and does it responsibly. As the company becomes more data-enabled, he argues, governance stops being a back-office function. The data strategy, and the AI plans built on it, depend on it. Getting it wrong costs far more, in business and reputation, than getting it right from the start.