Abound Credit Union opened years of member history to its analysts in 3 months

The workflow before#

Abound Credit Union serves members across Kentucky, and its leadership wanted two things: a soft-pull credit scoring model and a single view of each member. Neither was possible with the data as it stood. Member records lived in core systems that were hard to extract from, and older history sat on optical storage that analysts couldn't query. In practice the team could reach about 6 months of history, pulled partly by hand. Nobody had written down the governance rules for member data, so even the data they could reach wasn't something a regulated lender could build a credit model on with confidence.

What we built#

We built Abound a single home for member history. Automated pipelines pull records out of the core and archive systems into a central data lake, governance rules define who can see what and how accuracy is checked, and a modeled layer in Snowflake gives analysts years of history to query instead of months. The point was to give Abound a base it could trust before anyone trained a credit model on it.

Pipelines that retire optical storage

The pipelines extract from Abound's source systems on a schedule and land the data in Azure Data Lake. Older records that once sat on optical media now live in the lake with everything else. We made the lake the system of record for history so there's one place to look, not a current system plus an archive request.

Member data analysts can query

In Snowflake we modeled member, account and transaction data into tables an analyst can read without knowing how the core system stores it. An analyst asking how a member's balances moved over five years writes one query instead of filing a request. We modeled around the member rather than the source system because the member is what the credit and 360 work both start from.

Governance set before the models

We scoped the rules for member data with Abound's team: security, access and accuracy. Each rule says what it protects and who owns it. We did this first on purpose, because a credit union has to show examiners how a credit decision's inputs were controlled, and that's much harder to add after the fact.

How we delivered it

A lean team of three ran the work: a senior data engineer, a data architect and an analyst. We worked over 3 months alongside Abound's IT and analytics staff, starting with extraction from the hardest systems, then modeling, then governance sign-off. Abound's analysts tested the modeled tables against questions they already knew the answers to before the old process was retired.

Before and after#

What's next#

With history and governance in place, Abound can build the soft-pull credit scoring model it set out to get, trained on years of member behavior instead of a few months. The same modeled data supports a member 360 view, so member service staff see one picture of each relationship.