Mail order giant uses business intelligence to make every mailing count

Retailer uses BI to minimize catalog waste

Catalog retailer Fingerhut (www. fingerhut.com), which is being acquired by retailing giant Federated Department Stores (www.federated-fds.com), mails an average of a million catalogs daily. At the heart of that mailing system is Fingerhut’s massive data warehouse containing 7 million active customers. The company maintains detailed information about each customer, including birthdays, anniversaries, payment preferences, product purchases and product interests.

To manage the vast amount of data stored in its warehouse, Fingerhut originally relied on a Promotional Scoring System (PSS) to score the value of customers and to create catalog distribution lists based on those scores. To modernize the system, Fingerhut has developed a Mail Stream Optimization (MSO) program, consisting of an IBM SP2 (a distributed memory parallel system with four processors) and a business intelligence application. The MSO program is a huge project intended to reduce what the company calls "advertising"Ñits regular catalog mailings.

According to Randy Erdahl, group manager responsible for analysis at Fingerhut, the system is used to make mail/no mail decisions for each catalog. "We built a system to evaluate what the customer was seeing," he said. "The system recommends where to cut back."

Many of the software tools used in MSO were developed internally using C++ code that calls in SAS (www.sas .com) and IBM’s (www.ibm.com) Optimization Subroutine Library. The MSO application runs on an RS6000 SP2 with four nodes. According to Erdahl, there is no real database within the MSO application. Large files containing millions of customers, each with a unique set of attributes, are processed in MSO.

"The MSO input data is extracted from several databases today; someday soon it will access the corporate warehouse, which is a DB2 database," Erdahl said. "Because of the application’s requirement for lots of computing and the disposable nature of the data, it may never have databases inside the scoring process."

The only way to process the entire warehouse of data and make weekly reporting a reality was to run the business application in parallel, using the power of all four processors at once. However, the business intelligence application was designed to run sequentially, on one processor at a time, even on multiprocessor systems.

To solve the problem, Fingerhut uses Torrent Systems’ (www.torrent.com) Orchestrate to process data across multiple processors and then stream the data through multiple, parallel instances of each application step.

"The Torrent tools speed the processing of C++ shell code and the SAS code, but not the imbedded IBM OSL code," Erdahl explained.

"Of the more than 100 mailings per year, it was unusual for anyone to get all 100, but they were overlapping," he continued. "We felt we’d gone too far. We always thought this was a problem but had no way to computationally solve it. We now handle business intelligence problems that were too big to even think of solving before."

As for future uses, Erdahl said, "We foresee other ways that we could reap benefits from this investment. We need to make decisions today as to what catalogs to be mailing to our customers. We can’t mail 100 catalogs to all 7 million customers."