A large data analytics project is like painting a bridge, says Deva Annamalai, vice president and product and marketing technology strategist at Zions Bank in Salt Lake City.
"You start from one end and when you get to the other end, you've got to start at the beginning again," he says.
Annamalai heads a group Zions created about six months ago called Marketing Insights. Over the next several months, the team's goal is to identify business use cases and "data recipes" that will help the marketing department come up with more advanced customer segmentation capabilities, provide next-best offer products based on predictive analytics, and optimize marketing campaign execution.
One thing that makes this analytics team and its mission different from others is the source of much of its data the fraud analytics team headed by information security manager Michael Fowkes.
"Our fraud analytics group started a Big Data setup two years ago," Annamalai says. The group, which includes two data scientists, started collecting information primarily to protect customers from fraud. It reached a point where it had the bandwidth to offer the data analytics to other departments, starting with marketing.
"They were able to say, in the past you probably didn't have access to some of our bank information the way the systems are siloed, but now we're collecting all this information and we can get this data to you, if you ask the right questions," Annamalai says. For instance, the marketing group used to have limited access to branch transaction data. "That quickly opened up new ground for us, because from a marketing perspective getting the right data is always a challenge."
In one example, the marketing department might want to launch a commercial business card. To do so, it needs to find out which business customers already have a card. "If they do have a card, we want to know if they're making big payments to American Express or some other card provider," he says. This type of query, which might sound simple, is actually hard, Annamalai says, because the marketing department doesn't normally have access to the different sources of account and transaction data needed.
"We would scramble between multiple teams to try to come up with the data for a campaign," Annamalai says.
What's in it for the fraud analytics team to share their enterprise data warehouse?
It comes down to business justification for the project, Annamalai says. "They've reached a point where in order to expand to more systems within the bank, they need to be able to prove a business value beyond fraud modeling. Fraud modeling is a use case, but the next biggest use case is marketing insights, mining information about customer acquisition plans and attrition plans."
The fraud analytics group uses a Hadoop cluster-based data warehouse that houses about five petabytes worth of information. It receives feeds from 140 data sources, including core banking, online banking and loan servicing data. Some are real time; others are on a nightly, weekly or monthly basis. Some information is available in a few minutes, more complicated queries might take a few hours.
The group uses business tools to monitor activity across channels; for instance, noting if a customer makes a branch transaction at the same time as a mobile banking transaction.
The marketing insights group recently tried to build a business case for mobile remote deposit capture, using the fraud database. "We had to ask the question, how many mobile banking customers still walk into a branch to deposit a check?" Annamalai says. "In the old model, it would have been impossible to ask the question. With the amount of data these guys are tracking, they can tell which mobile banking customers have been active in the last 30 days, what checks have been deposited in their accounts, and the value of those checks."
Another marketing use for the system is cross-sales. The bank could query the data warehouse to find all customers who paid $1,000 or more on a loan not written on Zions. "That's something we could use to promote home equity products to them," Annamalai says. This could be set up as an automated, repeating query.
Annamalai is currently creating a "lifestyle score model" that would show how digital a customer is.
"The idea is, I want to be able to attract more customers who have a propensity to use the digital products I have," he says. "Are they using my online banking, mobile banking, bill pay? Are they signed up for e-statements? Are they signed up for purchase alerts?"
Once he completes and back-tests the model to see how well it works, he'll have a picture of customers that have a high propensity to use digital products that can be applied to the broader customer base.
Another use case is customer attrition; creating a model that would help determine why people leave the bank. "To do that, we need to look at all the customers who've left us and backtrace the behaviors that led up to that." It might turn out that someone who stops using bill payment is likely to leave within six months. Having that foresight would help the bank do something to encourage them to stay.
In the future, Annamalai hopes to use the Big Data project to help with customer acquisition (perhaps by applying existing models to outside lists). And at some point, it may be applied toward customer service. "If a customer calls the call center to resolve an issue and the tone of their voice is angry, and the agent didn't quite help them solve the problem," that's good information to mine, Annamalai says. By measuring the emotional interaction the customer has with the bank, combining that with the attrition model, and layering that with information about how valuable the customer is, the bank can build a plan for keeping that customer or not, as the case may be. This information could all be on a call center agent's screen as a customer calls in.
