First Commerce Corp. of Louisiana, one of the early adopters of data base marketing, is upgrading its technology to create more refined direct marketing campaigns.
The $8.4 billion-asset banking company recently installed software that can store in a relational data base and retrieve millions of files of customer data, including individual banking transactions, from disparate systems across the bank.
The system feeds details about customer behavior to an existing neural network that generates complex models of potential customers.
The new software, called Analytix, from Customer Insight Co. of Englewood, Colo., runs under Windows and is being used with a neural network from HMC Corp. that has been in place since early 1993.
First Commerce was the test site for Customer Insight's link between its data base software and the neural network from HMC, said Terry Larrew, president of Customer Insight.
First Commerce, which began its data base marketing program five years ago, declined to say how many customers typically respond to its campaigns.
But it did say that the neural network has helped increase response to the bank's marketing campaigns by four to eight times. The improved response is one of the factors - along with faster loan turnaround times - that has led to higher loan growth.
Data base marketing "has helped us maximize investment elsewhere in the company," said Jenny Cromer, senior vice president, client information at First Commerce.
"It used to be that you'd send a mailer out to 100 prospects and you'd be lucky if you got a 5% response, and really lucky if you got 10%," said Carol Post, vice president of marketing for Customer Insight.
The average response, using traditional techniques, would be around 3%, she said.
First Commerce began using software to profile customers from Customer Insight in 1988. In 1991, said Ms. Cromer, who worked in marketing for a brokerage firm before coming to First Commerce, she was approached by the bank's president and chief executive, Ian Arnof.
"He wanted us to build a data base that would be an off-balance-sheet asset to First Commerce," Ms. Cromer said. "Not many CEOs were thinking that way in 1991."
Since installing the neural network in early 1993, First Commerce has relied more heavily on a technique called modeling, as opposed to collecting profiles of customers, to generate direct mail campaigns.
The neural net added new layers of subtlety to the approach.
Profiling would yield, for example, a picture of a customer who is likely to take out an installment loan. This prospective customer could belong to one of several age groups, and one of several income levels. But it was essentially a "very shallow" picture, Ms. Cromer said.
The neural network "ferrets out a pattern that the brain couldn't find," she said. It comes up with a list of likely customers whose common attributes are so complex that it can at times appear they have nothing in common.
"You might be able to say that a typical installment loan customer has several common attributes - but then the top 100 people on your list might not have any of those attributes," Ms. Cromer said.
Last year, the bank conducted an experiment comparing the modeling technique with the profiling technique. The bank sent out 25,000 letters to customers whose names were culled by the neural network. These letters did not include prior credit approval.
The bank also sent out 100,000 letters to customers whose names were selected by profiling the customer base. For this group, the bank went to the added expense of including preapproved credit.
The modeling technique proved to have better results. While the larger, preapproved mailing generated 50 more auto loans than the smaller mailing, the smaller mailing - which was also the less expensive, since it did not include preapproved credit - generated 100 more loans of all types.
