Ch. 7
Building Data-Driven Decision Support Systems

In Spring 1995, MasterCard International Inc., based in Purchase, N.Y., set out to build from scratch a data warehouse that could deliver consumer purchasing data to the desktops of the credit card company's member banks. Up to that time, the banks had to formally request such information and wait, sometimes for days, while MasterCard staff prepared reports. Not a bad system but definitely not the cutting edge in an age when the rallying cry "marketing to millions, one at a time" was achieving mantra status.

Over the next five months, an internal development team brought in vendors, gathered a terabyte of data from the company's transaction-processing systems, integrated an end user front end, and tested, debugged and delivered the service in beta to its first group of customers (see "Mining the Possibilities" CIO, June 1, 1996). Today that application is an integral part of the company's IT infrastructure, closing in on 2 terabytes (an Oracle 7 database running in an AT&T System 3600 parallel processing environment), and at least a dozen decision-support applications have since been spun off--not only for member banks but also for MasterCard's internal operations. The challenge now facing business and IT leaders inside MasterCard is to keep up with the demand for new ways to use the data.

"We're always getting asked, 'Can we get this out of the warehouse? Can we get that?'" says Sam Alkhalaf, senior vice president of technology and strategic architecture. "Demand has gone nowhere but up."

One of the applications spawned by the data warehouse MasterTagger allows member banks to combine their customer data with the MasterCard data warehouse and run queries on the combined database. When a bank puts together a direct-marketing program, MasterTagger offers a more detailed profile of purchasing patterns than the bank could formulate on its own. Another example of the surge in database demand is Authorization Advisor, which provides MasterCard and its member banks with performance data on authorizations by crunching the numbers about accounts, amounts, peer group averages and average transaction times.

A number of technical aspects of the data warehouse have been improved: There's been an upgrade to a newer version of Oracle (7.3), plus enhancements to connectivity, backup processes and batch processing. Yet, says Anne Grim, formerly MasterCard's senior vice president of global information services, some of the most important developments have come "not so much in the technology as in the understanding of how to use it on the business side. It's been satisfying to see our member banks realize an incremental lift in the efficiency of their marketing."

The data warehouse has been good for internal relationships at MasterCard as well. Having constructed the data warehouse primarily for an external audience, the company finds itself managing a technology evolution that's the inverse of the typical scenario: Internal operations has caught wind of the resource that customers are using and wants a piece of the action. Alkhalaf notes that the user-query tools that MasterCard originally installed, Information Advantage Inc.'s DecisionSuite business analysis software, had to be tailored and augmented with home-grown tools to keep up with the shifting demands and diverse user groups. "You realize over time that not all user groups want one way of accessing the data," he says. "It turns out that it's pretty unrealistic to think that one tool fits all."

For example, the member banks needed to drill down deep into the data, automatically segment customer groups, and create control groups to test marketing programs on and track results over time. Meanwhile, some internal users were looking for a simple way to do basic MIS reporting. "Some of the tools we had in place were just too complex and unnecessary for what people were using them for," says Alkhalaf. "We've increased training on those for the users who need to drill real deep and developed templates for the more basic and common questions lots of other people need to run."

Alkhalaf's group has worked to improve the performance of the data conversion, extraction and loading processes. The company had some processes where moving data from operational systems to the warehouse took nearly two weeks. That's been cut to two days, says Alkhalaf, primarily by perfecting scrubbing routines, the processes by which redundant and superfluous data is identified and removed. Alkhalaf has also learned that there's a direct relationship between data conversion and extraction costs and the degree to which an organization standardizes data within the operational data source systems.

Accordingly, his group has established business rule tables to automatically cleanse data at the transaction level. The kind of fine-tuning that MasterCard has done the past two years is "not as bold and exciting as what we did on the front end," says Alkhalaf, "but it has helped us get to the core of the data and really make it useful."

Key lesson learned? "You really have to build in flexibility," says Alkhalaf. "Unless you believe you can guess all the technology pieces you're going to need--and you will not be able to--you've got to be flexible enough to be able to continually re-examine the pieces of the technology and implement new ones without affecting the entire infrastructure. That, we've found, is the magic of a good, strong architecture."