MCI Builds Customer Relationships with MPP

It's the essential paradox of mass marketing today: How do large companies make each of their millions of customers feel as if the company knows and treats them as individuals? How do they determine what products and services to offer to which customers?

MCI Corp., the Washington, DC-based telecommunications firm, found the answers to those and other questions in a large-scale sales and marketing data warehouse system. Called warehouseMCI, the system combines very large database (VLDB) and massively parallel processing (MPP) technologies to produce one of the largest commercial data warehouse systems in existence today, storing more than six terabytes (TB) of information on more than 20 million MCI customers.

More importantly, warehouseMCI is also the basis for virtually all of MCI's marketing and sales activities. "We are using the warehouse to build relationships with our customers and to understand what services people want," says Ryan Souza, senior manager for warehouseMCI at the company's information technology facilities in Colorado Springs, CO. "The trend today is to move from mass marketing to relationship marketing. We needed a cost-effective way to provide individual service to a mass market."

That challenge is further complicated by the firm's rapid growth and expansion of services over the past few years. With nearly $5 billion in sales last year, MCI was one of the first to provide an alternative long-distance telephone service to AT&T. As the company has grown, so have the kinds of telecommunications services it provides. Today, the company also provides cellular phone, electronic mail, Internet access and more. "How do I understand 20 million customers and what they want? How do I know what they need?" Souza asks.

MCI stumbled on its data warehouse system almost by accident. In September 1993, the Colorado Springs sales and marketing organization undertook a complete reengineering of the IT systems that support the group's business activities. The telecommunications market was changing rapidly, with deregulation, more open competition and many new services, including cellular and paging. MCI knew it had to change with the times. "We had decided back then that we needed to change from a product-driven marketing strategy to a customer-driven marketing strategy," says Souza.

After some study of the sales and marketing group's needs, the IS group determined three core competencies or components for a new computer system. The first was a client management component that would enable customer service and sales representatives to recognize customers when they called in; for instance, once the customer provided a phone number or some other form of identification, the MCI representative's computer screen would load up with details on all previous contacts with that customer. The second system needed was a decision support component, which would help provide strategies for marketing to customers. The third component, a contact management system, would store information about the customer and provide a detailed record of every contact MCI has had with its customers.

Souza says these three systems are, in fact, circular in nature; the client contact information feeds the decision support system, which in turn provides information to the contact management system. The contact management system likewise provides information to the client management system and so forth. Ideally for MCI, the three components, when combined, would help to provide, in Souza's words, "what customers want, when they want it, through the channel they want to use."

MCI didn't recognize right away that it needed a data warehouse, but it did want a system that would act as a central storage point for the information that could feed all three systems. "We knew what we needed," he says, "but we didn't know what it was."

While some companies have to draw up separate business plans to justify a data warehouse, Souza says the earlier management decision to reengineer the existing sales and marketing systems provided the rationale and business case for building the warehouse. Today, information about every contact between an MCI customer and a company representative gets entered into the central data warehouse as soon as it happens.

Back to Basics

In starting out on the project, MCI first chose a computer platform. Souza says MCI realized it would need an expandable, scalable computer platform for its data warehouse, as it anticipated rapid growth both in data storage needs and the number of users supported. At the time (mid-1994) he felt that symmetric multiprocessing (SMP) systems were limited to eight to 12 processors. "We didn't feel that SMP systems were inherently scalable enough," says Souza. "We wanted to be able to increase the size of the machine and incur incremental costs in as linear a fashion as possible."

Although clustered SMP systems today offer greater scalability than they did a year or two ago, Souza says the company turned its attention to the only available platforms that would meet its needs at the time, MPP systems. Of those around, it found the IBM RS/6000 SP to be the most viable; it offered one of the largest expansion capabilities at the time--a maximum of 512 IBM RISC-based CPUs.

MCI evaluated three databases--Oracle Parallel Server, Informix-XPS and IBM's DB2 Parallel Edition--that could run on the RS/6000 SP. The firm selected the Informix-XPS product. In particular, Souza says, MCI felt that its shared-nothing architecture would support the scalability of the system better than the shared-disk approach some other data warehouse solutions take.

The warehouseMCI system went into production in late 1994 with a 16-node IBM RS/6000 SP and an early beta copy of Informix-XPS. Souza says the system has grown faster than anticipated, having been upgraded first to a 52-node system and then to the current 104-node configuration in mid-1995. Each node has 25MB of dedicated main memory. The system has 800MB of "live" data that is fully mirrored. The remaining 4.4TB is used for staging data for various analyses.

The MCI system, by the way, is the largest known commercial IBM RS/6000 SP in operation today. Only Cornell University in Ithaca, NY, is known to have a larger system--it has the maximum the system's architecture supports, 512 processors--but it is used for academic and scientific research, not commercial data warehousing.

Offloading from the Warehouse

MCI doesn't permit its marketing analysts to work directly against the warehouseMCI system. It is an enterprise data warehouse, intended as a central repository of information that the whole corporation can access for data. "We have no users directly hitting the [central] warehouse," says Souza.

Instead, to support the activities of its various business units, MCI has established four production data marts. Each contains targeted segments of the data warehouse's information base, not the entire amount, which are tailored to the needs of a particular marketing or business group. Each data mart runs on a Sun Microsystems Sparccenter2000 or Sparcstation-20, configured as a server and located in each business unit. "The business units pull the data they need from the corporate warehouse," says Souza. "They can then do the real business operations on the mart side."

Because each business unit operates its own data mart, MCI's IS group doesn't have to make promises about update schedules or around-the-clock operation guarantees. This eases the management burden of such a large data warehouse. And the central IS organization supports local data marts only in business units that don't have in-line IS groups of their own.

This not only preserves the integrity of the enterprise warehouse, it permits each business unit to choose whatever tools it deems best for its own decision support and analysis activities. "We don't want to have a central IT organization that users have to go through to get their jobs done," says Souza. "We allow them [the business units] to purchase and use any tool they need to do their jobs."

One of the first tangible changes to come from the warehouseMCI project is the marketing of MCI One, a package of various telecommunications services the company tailors to meet the specific needs of individual customers. For instance, a customer can combine long-distance, cellular and e-mail services for less cost than purchasing each of the three services separately.

Thus far, MCI has concentrated warehouseMCI on getting to know its residential service customers. As the project expands, MCI plans to extend access to the data warehouse to its commercial sales and marketing organization, so they too can get to know customers in a personal way.