BankAmerica Keeps a Competitive Edge in Stock in Its 'Warehouse'

BankAmerica Corp. believes the best way to exploit its vast cache of information for a competitive edge is to dig deeper into its data warehouse.

The bank company, according to Charles Griffith, system director and vice president of data warehouse, has explored several customer segmentation schemes. But "customer mining is more sophisticated than a segmentation scheme of a few segments," he said.

At Bank of America, the flagship subsidiary, 1,200 of the 97,500 employees have access to the data warehouse. They make 3,000 queries a day.

This elite group is made up of users with analytical backgrounds who support management decision-making and can figure out, among other things, how to develop and sell retail products effectively.

Ph.D. types, for example, do time series analyses of customer product ownership based on which product was acquired first. This helps determine the next most likely product to sell a customer.

Although such sophisticated users develop valuable insights, some industry observers believe it important for banks to give employees who deal directly with customers the ability to enter the data warehouse and manipulate information.

To that end, Bank of America is starting to broaden its end-user group.

Mr. Griffith and his team are doing a pilot test in which district sales staff, armed with sophisticated client/server access tools, can perform functions such as mining for certain product conditions that indicate whether a customer would want additional products.

Customers with a couple of bounced checks, for example, could be targeted for overdraft protection and direct deposit of paychecks.

Mr. Griffith pointed out that the successful rollout of the program requires intensive up-front training of the sales staff in the new technology. In addition, the bank must ensure that an information "free- for-all" doesn't ensue, so sales employees will be restricted to information about their respective customers.

On the wholesale banking side, the bank is not as far along with data mining capabilities.

Though wholesale data can be viewed on a summary level, drill-down capability is not yet up to speed. The wholesale business runs a few hundred systems, which makes information integration difficult.

Like many banks, Bank of America is also capturing and segmenting customers' transaction behavior, including automated teller machine, branch, and telephone banking usage. "We're broadening our traditional customer profile, which was limited to a financial perspective," said Mr. Griffith.

The captured transaction data will be used to help decide which distribution channels to market to customers. In addition, said Mr. Griffith, as the bank continues to capture transaction data - and allocate transaction costs - "profitability at the customer level will be refined."

Capturing customer transaction information, banks are discovering, can be daunting.

"Although most banks have set up customer information files, they rarely collect transaction information," said Larry Russell, managing vice president, First Manhattan Consulting Group.

Assembling that information - plus reliable transaction-related cost allocations - into one neat pile is a massive job, he added.

The rewards for this hard work are great, though. The information banks collect reveals their customers' preferences. And these can identify to what products and services customer groups will most likely respond, according to Mr. Russell.

Equally important, the information can be used to boost profitability. "Consumer financial products are one of the few classes of consumer products whose profitability is affected by the way in which they are used," Mr. Russell explained.

Some checking account customers, for example, may come into a branch once a week to reconcile their accounts. They're using that service in a way that average pricing never anticipated; thus, other customers subsidize their behavior. "Those customers being subsidized have had a terrific run," said Mr. Russell.

Gathering transaction information on existing customers and analyzing it with traditional statistical methods offer enough benefits that banks don't have to leap to the next level of sophistication, according to Mr. Russell and others.

That next level is the application of artificial intelligence, such as neural networks, which can be used to predict behavior.

"If you want to do a very focused analysis of a particular set of data, or group of customers, you have to move to neural nets," asserted Joseph Drake, a principal at CSC Consulting, Waltham, Mass. Neural networks, he added, can detect patterns and other characteristics that humans can't.

BankAmerica, which is exploring modeling software to predict customer behavior, would not say whether that software includes neural networks.

In addition to integrating information, banks must establish exactly what sort of behaviors it needs to know about and is capable of documenting. "You can't collect everything and shouldn't collect everything," Mr. Drake said.

Banks also have to be sure they don't raise the ire of the growing number of consumer privacy advocates as they gather and tweak more and more information on their customers.

At the same time, though, banks view proprietary information about their customers as a competitive weapon against traditional nonbanks and, possibly, on-line service providers.

BankAmerica, said Mr. Griffith, imposes stringent security controls on its customer data. And as First Manhattan's Mr. Russell noted, "Banks certainly won't sell or misuse the data. That would be crazy."

The amount of information BankAmerica is guarding is enormous. Its data warehouse - with 800 gigabytes of memory - currently consists of more than 30 banking systems, mostly retail.

Although virtually all the banking systems feeding the warehouse are mainframe-based, the bank recently added data from a client/server loan application system. The move reduces the need for ad hoc reporting and reduces the amount of historical storage on the loan application system.

Account data are pulled together - in two to three days - to construct a relationship view of customers across the systems. Views are available on the individual customer, account, and household levels. The data make up the bank's internally developed marketing customer information file - an integral part of the warehouse.

In addition to its 40 million households, the warehouse also has information on millions of noncustomer prospects, as well as purchased data from various suppliers for marketing efforts. Because of its huge size, the bank runs the warehouse on massively parallel processors, from AT&T Global Information Solutions.

The bank drives all of its marketing campaigns off the warehouse. For example, based on various types of information, it searches for potential first-time homebuyers from customer and noncustomer groups.

It then sends the selected group invitations to bank-sponsored seminars on the ins and outs of buying a home.

In addition, it recently repriced its credit card portfolio based on customer relationships. Interest rate reductions were given to customers who had checking accounts with the bank. Extensive analyses were done on the warehouse, then data were sent back to the credit card system.

BankAmerica began building its data warehouse in 1986. As a pioneer, it had to cope with difficulties that even today plague such projects.

For example, it had to standardize the huge amounts of data from its banking systems that feed the warehouse. Those systems include ones that were purchased, purchased then brought in-house and customized, or developed entirely in-house.

"You can't underestimate the amount of dirty work involved up-front to get all this disparate data together," said Mr. Griffith.

In addition to breaking ground with its use of parallel processing, the bank grappled with client/server technology before many of its peers. In 1991, it pioneered client/server methods of gaining access to the warehouse. Back then, said Mr. Griffith, "there were only a few vendors with very shaky PC access to the warehouse."

Today, the bank supports a variety of standard tools, and about 300 users have PC/LAN access.

So what's next? Mr. Griffith won't disclose many details, but it's a sure bet that the bank will continue to work on its mining technique.

Ms. O'Heney is a freelance writer based in New York.

For reprint and licensing requests for this article, click here.
MORE FROM AMERICAN BANKER