Coastal Federal Credit Union ($2.2B) began as the credit union for IBM in Raleigh, NC. It still serves the Research Triangle area today and is the second largest credit union in the state. Below Kris Kovacs, senior vice president of operations, shares his thoughts on the Big Data phenomenon and the role complex information analysis has played in Coastal’s long-term success.
Can you define Big Data?
Kris Kovacs: Not really. It’s a little bit like trying to define “the cloud.” In fact, the lack of a concrete definition can make some people uncomfortable. But in a way, Big Data is what people have been doing all along — looking at and analyzing data. I don’t know the tipping point where a credit union moves from generally looking at data and is suddenly in Big Data.
How did this strategy evolve at Coastal?
KK: We run our core system in house and it includes an open database. We made copies of this database available for analysis and reporting but despite those preparations we kept hearing the same thing from our business units: “We don’t have the data.” So despite making every piece of data in the core available, we had more work to do to reach our data goals. We began to think maybe the problem wasn’t the data at all.
How did you proceed?
KK: First we needed a partner. We at Coastal are very fortunate to have as one of our SEGs a global leader in business analytics, SAS, whose campus is just outside Raleigh. When we have an intriguing or challenging data question we go to them for ideas — and they come to us when they need some thoughts on the financial services industry. So we knew we had this powerful resource available to us.
What we needed at Coastal was trained talent. So we went both to the outside to acquire it and to the inside to discover and develop it.
About two years ago we created an ad hoc users group to be trained on SAS tools. We called it the Business Analytics Group and it drew from all aspects of the credit union — lending, operations, retail, marketing, and even internal auditing — with the idea that this group would develop a data program for Coastal. They would identify who needed training and what sort of training as well as identify new sources of data to interconnect in order to do new types of analysis. They would be self-sufficient, elect their own leader, and report quarterly to me on what they had done. We did not give these people explicit deliverables. We wanted them to explore the topic on their own.
What has all of this brought you?
KK: Today we have an SAS business intelligence tool that connects databases from all over the credit union — from lending, retail, core, online banking data bases, etc. — so that we can analyze across all channels and all products. Now when questions come up — and we are getting better at this — we look at the data. The old way was, “I’ve been in credit unions for 20 years and this is how it’s done.” But we’ve learned you have to balance that by looking at what the data is truly telling you.
So we look at the data and use it in models. Today we model new loans, our loan portfolio, predicted loss rates, and member behavior, and then use the results to make budget decisions. The analysis helps us determine what loans to make, how to price products, how to position our checking accounts, and so forth. It has allowed us to be more nimble in our decision-making. One thing it does is eliminate the “What about this? And what about that?” questioning that can slow decision-making. Today even the questions have become more data focused. We can be more confident in the decisions we are making.
Can’t data also overwhelm and slow decisions?
KK: It can unless you achieve the balance of talent and training. If you put the right data in the wrong hands you can be swimming in that data forever. You’ve got to get people to the point where they understand what’s relevant and what’s not, and that takes time. You can bring in people from the outside to help but when they are inside your data they won’t understand the historical record of decisions embedded in it. They might see anomalies and simply not understand them. It takes someone who has grown internally with the product who can translate some of what is behind the data.
How do you get the talent?
KK: Ultimately we made two efforts, one to grow talent from within and the second to hire some “data-centric” managers to encourage data analysis. Growing talent from within was the ultimate goal of establishing the ad hoc Business Analytics Group. The team met regularly to discuss their projects and demonstrate how they were using the data. This led other group members to start asking questions about their data projects, and they were able to build the skills and experience necessary.
We also went outside the credit union to hire talent. We brought in some very talented data-centric managers who helped us in two ways: one, by increasing our capabilities, and two, by serving as great mentors to people developing those skills internally.
How translatable are these skills? Can someone move readily from credit union to credit union and be as effective?
KK: We had someone come in here from the banking world. He adapted quickly because he knew the data structures he was looking for. We had to help translate what we had with respect to what he was looking for; but it worked out pretty well. Based on this experience, I guess you can say that once you understand the data structures they are pretty translatable across institutions because you understand the inputs and outputs of the process.
Isn’t data all about the past and policy all about the future? How is the one connected to the other?
KK: You can use the data of the past to predict the future. Credit unions are just adapting to the environment and tools that are now available. For example, credit unions look at a credit score and — along with other criteria — decide to make or not make a loan. A credit score is a predictive tool. If you look at a body of loans you have made at different credit scores you should see some trends in those scores, which allows you to perform some predictive analysis. Credit agency models will tell you that loans to borrowers with, say, a 670 credit rating will go bad X percentage of the time. Believe them and build your analysis around their stated predictive value. When you do that, you start predicting out into the future a bit better.
Do you have an example?
KK: Here is something we did. We took all our delinquent accounts and all the data points about them — the history of payments, when they failed, household income, credit scores, and so on. Then using tools from our partners we asked which data points two years previously were the most predictive of how a loan would end up. Using this sort of regression analysis we are able to apply what we learned to data and applications today.
Once you’ve predicted an outcome now you can test strategies to change the future. This is a key point and it is not as difficult as it might sound. You can create a champion-challenger type of test. The champion is the normal way of doing things and the challenger is a process you are considering as a new tactic or strategy. You create two test groups. First is a control group, one that has your standard behaviors and is statistically neutral, say, accounts that end in digits 00 to 09. Then you take another group, the accounts that end in digits 10 to 99, and you treat this second group under a process you are testing — it might be with a model that is more aggressive or less aggressive than the one used for the control group. Next you monitor the performance of the accounts for six to twelve months. Then you can look at the performance of the accounts and see if the challenger group has performed better or worse for the credit union than the champion group.
Who looks at the data the most in your organization?
KK: The business people. If they want some information they go to the data. If they believe they need to see something more or something new they go to the IT people for help. Generally, we rely on our VP corps, which consists of about 16 people. If they want to make a business case, they do the data analysis in support of it. They would present the case and the supporting data to the senior management team, which reviews and makes decisions on large requests.
How expensive is this kind of analysis to get into?
KK: The tools, the training, and sometimes the talent are not inexpensive. But the alternative is making loans you shouldn’t make, paying interest and dividends you shouldn’t have to pay, placing branches in the wrong areas, and missing members you could be serving. These are very costly mistakes.
What do you feel is critical to success with Big Data?
KK: You have to have directors and senior managers who are supportive and understand there are revelations this data can provide. We are lucky here at Coastal to have a management team that has asked for more data, better data, and improved reporting — thereby raising the bar for everyone here.