2 Lessons From A $3 Billion Data Powerhouse

“BCU ($3.2B, Vernon Hills, IL) has an insatiable appetite for data” says credit union vice president and chief data officer John Sahagian.

BCU has had analysts embedded within different departments since the early aughts. By 2005, the $3 billion credit union needed to make the sheer volume of data it generated accessible to everyone who wanted it. So, it allocated funding for the research and development of a data warehouse. Once the analysts recognized the power of the warehouse, things quickly took off.

“The analysts started applying use cases to it” Sahagian says. “Then they started asking for more: Can you add this field? Append this table? They really wanted to expand their use of it. Within a few years, it became an important reporting and analytics tool for the entire organization.”

Building its data warehouse case by case has its benefits namely, understanding the organization s needs and growing to satisfy them. But being on the analytical forefront has also presented BCU with challenges, and for the past few years the Illinois cooperative has been working to alleviate pain points arising from structure inefficiencies and core governance.

Remap And Refactor

BCU has built a data warehouse based on the queries of curious, analytical minds. The result is a patchwork quilt of data sets and query fields that contains information valuable enough that someone, at some point, went looking for it.

“I’m a big advocate for building it this way” Sahagian says. “That’s versus building the Titanic before deciding where to sail.”

Still, the piecemeal process has resulted in inefficiencies surrounding duplicative, inconsistent, and hard-to-find data. Finding the right data can require some digging, so organizing the warehouse is an ongoing endeavor at this big, data-driven organization.

“We need the right balance of standards to empower the business to run more efficiently and effectively without creating the kinds of bottlenecks and obstacles that will take away from the value of the data warehouse” Sahagian adds.

BCU is still investigating what that balance is, but for now, it has identified a few places to make adjustments.

First, the origin of new data introduced into the warehouse must be clearly defined so analysts know the validity of what they are citing. Second, new data can’t already exist in the warehouse. So, BCU must refactor its standards as well as remap the data already in its warehouse. That s a tall task with potentially big rewards.

“We don’t want to have to ask if the data is reliable or if this is the proper place to query it,” Sahagian says. “If it’s well-mapped and well-defined, then we’ll know it s high quality and consistent. We won’t need to spend too much time validating our analysis.”

Centralize Requests

The patchwork nature of BCU’s data warehouse isn’t the result of case-by-case queries alone. The credit union’s decision to adopt a decentralized functional analyst team played a role, too. With no high-level group ownership of the data, the warehouse swelled into an inefficient store shaped by deep dives with narrow departmental focuses.

That’s why BCU reorganized internal talent to deploy a member intelligence team in 2016. Sahagian oversees the three-person team that includes two full-time analysts and an experienced data technology professional who acts as the group s manager. One of the analysts is a member analyst who works on a range of analytical and utilization projects, including standardizing how BCU visualizes projects through Microsoft s Power BI program. The other is a data scientist who focuses on developing models for advanced analytical projects.

The member intelligence team is laying the groundwork for more formalized data governance standards while at the same time completing enterprise project requests funneled through a central portal.

Requests vary but have included analysis into exposure for BCU’s legal team and calculations into insurance limits for its deposit team, among others. “The biggest application” says Sahagian, “has been with BCU s marketing department. The credit union has historically segmented campaigns by rule-based targeting, but over the past year, the member intelligence team has helped develop propensity modeling with the aid of machine learning.”

“We’ve seen a terrific lift in our response rates with model-driven targeting,” Sahagian says.

Functional department-level analysts still remain vital to BCU’s operations, but Sahagian anticipates continued evolution of their hybrid model for managing data and analytic projects in the future. As the member intelligent team fulfills more requests and the data ecosystem continues to grow, Sahagian recognizes there might be concern about scale. He hopes funneling more requests through a centralized team that applies consistent use of data will increase efficiencies and alleviate concerns.

He also expects the way analytical insights are gathered to change. Today, human curiosity and SQL queries allow analysts to report on data; tomorrow, Sahagian expects modern tools and techniques to surface insights in real time.

“We want to leverage machine learning to surface insights we might not have otherwise noticed,” he says.

Sahagian and the BCU analytics team understand that what they’ve set out to do is a tall task in a fast-moving field. The chief data officer believes in iterative growth starting small and learning from success and failure. That s exactly what the credit union has done over the past 12 years. It built a data warehouse one use case at a time, learned from mistakes, and remained flexible as times changed.

“You’ve got to have specific use cases in mind when you build something like this” Sahagian says. “If you try to boil the ocean, you are going to spend an insane amount of time and money and likely not reach your goals.”