---
title: "How a Leading US Mortgage Lender Made Its AWS Data AI-Ready"
url: "https://atlan.com/know/mortgage-lender-ai-ready-data-atlan-aws/"
description: "How a leading US mortgage lender uses Atlan on AWS to generate and standardize documentation, raising coverage from 11% to 48% across 9.8 million assets."
author: "Mark Manning"
author_role: "Director of Product Marketing"
published: "2026-10-13"
updated: "2026-10-13T00:00:00.000Z"
---

> Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/

## At a glance

| | |
|---|---|
| **Company** | A leading US mortgage lender. Over **$100 billion** in closed loan originations a year and **nearly 3 million** loans serviced. |
| **Challenge** | Seven million data assets, growing by three to five hundred new tables a week, documented by hand, with **117 separate naming conventions** in use for a single customer field. |
| **Solution** | Atlan as the metadata control plane over an AWS-anchored estate, with AI Steward generating and standardizing documentation at estate scale. |
| **Result** | Documentation coverage rose from **11% to 48%** in twelve months while the estate grew more than 40%. Governed business terms grew from **1 to nearly 18,000**. |
| **Outcome** | **Trusted answers on regulated borrower data**, at a volume no team could document by hand. |

---

## A hundred billion dollars a year, and all of it is data

This lender closes more than $100 billion in loans a year and services nearly 3 million loans. Every one of them generates data that has to be right: borrower information, risk inputs, servicing events, and the figures behind regulatory reporting. Ninety-seven percent of servicing clients stay, and that retention rests on decisions made from data at a volume no team can inspect by hand.

The business was consolidating onto a new cloud data platform, with core customer, risk and servicing workloads standardizing on AWS as legacy systems were decommissioned. Governance had to move with them, at the same pace.

## 300 new tables a week, and 117 names for the same thing

Three to five hundred new tables arrive every week, on top of an estate that had already passed seven million assets. Documentation was applied one asset at a time, and the same business concept was defined over and over. The governance team counted 117 separate naming conventions for a customer's first name alone.

Analysts couldn't tell from an asset whether they had found the right one, so discovery routed back to the data team. AI and agent use cases couldn't be grounded reliably either, because an agent reading only technical metadata resolves meaning from schema names instead of the organization's own definitions.

What happens without governed context
The agent still answers. It answers wrong, and nothing in the output shows which of the 117 definitions it used. The query runs, the answer looks plausible, and nobody catches it. In lending, that is a more dangerous failure than an outage. A better model doesn't fix it, because the ambiguity is in the organization, not the reasoning.
Writing the definitions by hand isn't an option either. The estate grows by about 56,000 assets a week. A steward writing one description every two minutes, eight hours a day, covers about 1,200. That leaves the lender choosing between ungrounded agents and no agents at all.

## Define it once, let the platform apply it everywhere

Atlan connects to the lender's AWS services directly. Amazon Athena holds the largest governed footprint, at roughly 300,000 assets, with Amazon Redshift and Amazon DynamoDB alongside it. Metadata, lineage and glossary terms come together in Atlan without the underlying data moving.

AI Steward workflows run across that estate, generating and standardizing column and table descriptions. Stewards review and certify what the workflows produce instead of writing each description themselves.

The governance team set one rule: define a concept once and let the platform apply it everywhere. One definition of a customer's first name now propagates to every asset that carries it, in place of 117 conventions maintained by hand. That governed layer reaches analysts through search, and AI agents through Atlan's managed MCP server, so both work from the same certified context.

## Quadrupling documentation coverage in twelve months

In October 2025, about 780,000 of 6.9 million assets carried descriptions. Twelve months later, roughly 4.7 million of 9.8 million did. The estate grew more than 40% over the same period, so coverage rose against a growing denominator.

| | |
|---|---|
| **11% -> 48%** | Documentation coverage across the governed estate, October 2025 to October 2026 |
| **1 -> ~18,000** | Governed business terms over the same twelve months |

Described assets jumped from roughly 2.5 million in August 2026 to 4.7 million in September, as AI Steward enrichment landed across the estate. Coverage on the AWS services runs ahead of the estate-wide figure, at about 55% across Athena, Redshift and DynamoDB, with Athena alone near 57%.

The internal goal, set in June 2026, was 50% coverage by the third quarter. The program reached 48%.

An analyst or an agent working on borrower data can now find the right asset and read the lender's own definition of what's in it, on an estate taking in hundreds of new tables a week. A larger stewardship team wouldn't have delivered that.

## What the foundation now supports

Query log mining was approved in October 2026, after a security review. And as the core warehouse migrates, the metadata layer moves with it, so the next platform inherits the context instead of starting over.

## Where it leaves them

The lender set out to define a concept once and let the platform apply it everywhere, instead of documenting its estate asset by asset. Twelve months on, it has. What remains is keeping pace with new tables as they arrive; at this intake rate, the backlog was never reachable. And because the context lives in the metadata layer rather than the warehouse, it moves when the warehouse does.

    Ready to make your AWS data estate AI-ready?

    See Atlan on AWS ->