DBMS

Targeting Customers

By Peter Brooks
DBMS, December 1996

MCI Leverages Data Warehouse Technology to Strengthen its Marketing Campaigns.


MCI Telecommunications Corp. (Washington, D.C.), a pioneer in the telecommunications industry as well as in the use of large-scale databases, has developed a data warehouse called warehouseMCI to support MCI's transition from a product-focused company to a customer-focused company. Based on relational database and online analytical processing (OLAP) technology, MCI's efforts serve as an example of the architecture required to support large data warehouses. WarehouseMCI is innovative because of:

In the Beginning . . .

Approximately three years ago, MCI decided that it needed a data warehouse of customer information, which would serve as a key component in achieving the company's target marketing goals. Improved marketing information and capabilities were needed to market MCI against the traditional long-distance companies, compete with local telephone companies, and partner with other companies. MCI's legacy systems held much fragmented data. Customer information was spread across customer, product, account, and prospect files. Data quality, management, and hygiene all required attention. Performing queries against customer data required that marketing analysts ask IT staff to write COBOL programs, a process that would take weeks or months.

Cross-selling was a problem: Customers with both an MCI credit card and a non-card plan could receive two solicitations, rather than a single offer, based on their appearance in multiple files. The existing database didn't have customer attributes that could be used by marketing. Scoring models and scoring (the assignment of a number to a customer to indicate how strong the customer ranks on certain scales) had to be outsourced.

The development of the warehouseMCI architecture was sponsored by the Database Marketing Operations division within MCI's Mass Markets organization, after a gap analysis was performed to identify the required capabilities. The team followed a staged "build it and they will come" strategy to generate user interest. Larry Greenfield, senior manager in Database Marketing Operations at MCI, said that his goal was to give users decision-making-based data and then validate the system with focus groups rather than have the users start from scratch. MCI was assisted by Joe Reese, president of the Littleton, Colo.-based Cornerstone Concepts, who was the lead data mart designer and architect.

WarehouseMCI was built to contain all customer marketing information. All types of customers are included -- residential, commercial, long-distance, cellular, and paging services. The host warehouse is the single source of marketing information. Functional data marts contain selected subsets of host warehouse data. One data mart is the Database Strategic Demographic Analysis Decision Support System (DSDA DSS), which uses MicroStrategy Inc.'s (Vienna, Va.) relational DSS Agent OLAP product to perform database marketing analysis.

The new architecture provides significant improvement over the prior system. By consolidating marketing information in one place, fragmentation is eliminated. Database marketing techniques such as creating scoring models and scoring the database can now be done in a matter of hours. Ad hoc queries take seconds or minutes for end users at a terminal rather than weeks of COBOL programming. The user organization justified the warehouse, funded it, and controlled its capital. As time goes on, IT will take over more of the day-to-day operations.

Background

The telecommunications industry continues to undergo significant change. One facet of this change is that successful telecommunications companies must become adept marketers. Who hasn't received a solicitation to change telephone service from one telephone company to another? I used to get such offers fairly often but have not received one in a quite a while. I have, however, received offers to acquire extra calling cards and join additional calling plans from my local telephone company, my long-distance company, and other long-distance companies.

Rather than simply mailing out offers to a large number of people, companies now use target marketing analysis and techniques to increase their marketing effectiveness. With a potential customer base covering virtually the entire country, mail and telemarketing costs are very high. Instead of sending out offers to such a large base, it is much more effective to identify a service of value to a selected market segment and then tailor an offer to that segment. Once the target market and offer benefit are identified, a marketing campaign must be carefully designed to achieve the optimum customer acceptance rate at the best cost.

Target marketing can most effectively be accomplished with a marketing data warehouse that allows easy interrogation of the data. With this approach, customers can be analyzed and segmented by common values, needs, and characteristics -- not simply by the products they buy. "Database marketing," "one-to-one marketing," and "relationship marketing" are other terms used to describe the use of data warehouses in this manner.

Data Warehouse Architecture

MCI developed warehouseMCI to contain spending pattern, product usage, demographic, and lifestyle information of over 100 million current and potential customers. The database exists on an IBM SP2 running MVS. There are currently 104 nodes in this SMP platform. Informix 8.0 is used to store and manage the three-terabyte relational database. Data, indices, aggregations, and other database overhead are approximately one terabyte out of the total of three terabytes. The database is mirrored, so six terabytes of disk storage are actually used. The host warehouse is the source of record for customer marketing data, including the data used for decision-support systems. It holds up to 3500 characteristics associated with each customer. This data warehouse is different from a billing database -- no individual customers are identified in the data warehouse. Host warehouse data, stored in a fairly normalized format as opposed to a star schema design, is not directly accessible to end users. The host data warehouse is updated from operational systems as the data becomes available, on a daily to quarterly basis.

MCI's data warehouse contains at least one characteristic of data mining databases: "scores," which are numerical ratings of customers along certain dimensions. Say that potential customers with a high propensity to travel might seem like a good market for cellular service, and that such customers seem likely to accept a direct mail offer. MCI staff members, often with a strong mathematical background, create a scoring model and then run regression and other statistical analyses against demographic, psychographic, lifestyle, and spending-pattern information in the host warehouse to identify characteristics of customers who travel. The propensity of a customer to travel could then be indicated by a numeric score, which is used to analyze this new market. This process of creating a model and then scoring -- a traditional database marketing technique -- used to take MCI six weeks. Now scoring can be done in hours. This speed and cost savings give MCI a competitive advantage.

MCI attempted to use neural network technology to perform scoring but has generally found that human involvement is hard to beat. In addition, one of the common problems with using neural networks to perform scoring is that such networks do not indicate how a particular scoring model was derived.

The data mart layer was included in the architecture to avoid forcing end users to access the host data warehouse and to enable users to extract information easily. Each of the three currently available data marts consists of distinct business views, one of which is the Strategic Demographic Analysis database. Twelve more data marts are planned for the near future. Each mart strikes a balance between containing all rows of customer information that are in the host warehouse but very few attributes per customer and fewer customer rows but many attributes per customer.

A typical mart contains a business view of between 100 and 200 elements out of 3500 and a random selection of customer information rows. The data marts do not contain any customer-specific information such as names and addresses, but they can contain over 80 million rows of aggregate and summary information. Each data mart consists of a Sun 2000E SMP platform with 85 to 100GB of disk space, including indices and other overhead. Informix 7.1 is used as the database manager, and a star schema is used that contains one fact table and 140 dimensions.

MicroStrategy's DSS Server 3.05 is used both to let users submit asynchronous requests and to manage the data mart decision-support query environment. DSS Agent (the client component) and DSS Server let users submit batch queries that will run immediately on the server or at a specified time, thus freeing up the users' workstations. MCI's IT staff can access DSS Server statistics tables to monitor the SQL submitted by users. Long-running queries can be analyzed and either the query can be modified or the data restructured to shorten query time.

Data marts are updated from the host warehouse on a biweekly basis, and the entire update process ranges from 30 to 70 hours per data mart. Of this total, host warehouse processing can take six to 14 hours. Actual data transferring from the host data warehouse to a data mart can take two hours or more. The bulk of the update processing time is spent creating indices and aggregations in the data marts.

One of the project challenges has been in building the data mart database to handle 140 dimensions rather than the typical 20 or 30. Each dimension, such as a customer's age, can consist of a discrete value and also placement of the value into a band (in this case, 21 to 30). The number of dimensions, in addition to the volume of the data, originally led to performance problems. These problems were solved by a unique data compression and hashing algorithm approach that results in the fact table holding all of the 140 dimensions in what appears to be approximately 30 dimensions.

The users had to be educated in the capabilities of data warehousing. To design the data mart databases, users were asked to come up with a list of the most common questions they needed answered. Users initially replied that they wanted to be able to ask any question and get an answer back within 30 seconds. (This timing, obviously, was not feasible.) Users were educated in the data mart capabilities and in some of the tradeoffs between data availability and performance. Even now, end users would prefer to access the atomic level rather than summary tables. Of course, the summaries are needed to provide good response time.

End users access the DSDA DSS using MicroStrategy's DSS Agent 3.05 on Windows NT, Windows 95, or Windows 3.1 client PCs. There are approximately 15 users per data mart. The average query time is approximately 15 seconds.

Although most queries are performed against the data marts, sophisticated questions or analysis must sometimes be run against a larger selection of customer rows or a larger number of attributes per customer than exist in a data mart. However, the normalized data in the host warehouse is not as easily understandable or accessible to users. Giving end users the unrestricted ability to run queries against the host warehouse may require complicated logic and could result in performance problems. Consequently, end users submit ad hoc query requests that need to access data in the host warehouse onto a Web page. The requests are routed from the Web page to an ad hoc query team staffed by the IT organization. This group uses SQL and C++ to run the requests, carefully structuring the requests to ensure accuracy and provide the best possible performance. A 24-hour turnaround is the group's goal for these types of host database requests.

Application

A primary application for warehouseMCI is dynamic market segmentation using the data marts. This segmentation lets MCI create products customized to small, focused market segments rather than offering only products with mass-market appeal. Marketing analysts perform an iterative process of using their business acumen to identify potential market segments and then test their hypotheses against the data. In turn, the data can provide indications of potentially profitable market segments. Once a segment is defined, a marketing campaign will be tested and rolled out to the actual or potential customers in the segment. After the campaign is complete, it can be analyzed on a return-on-investment basis. Actual results are compared to anticipated results.

WarehouseMCI uses MicroStrategy's DSS Agent decision support software to perform the marketing analysis against the data marts. MCI analysts log in to DSS Agent and have access to up to 140 dimensional views of a data mart. Dynamic market segmentation is performed by selecting customer rows, which is accomplished by selecting from among the views. For example, all owners of certain car models may be selected from a dimension that includes car ownership and purchases. This selection may have five million members and be too scattered for the target market. The segment will then be refined by further selecting by geography, demographic characteristics such as age, or lifestyle information such as antique ownership. Contact history must also be taken into consideration so that the same people are not always solicited. DSS Agent makes it very easy to see the results of the selections and change the segments dynamically.

Users take advantage of the MicroStrategy's DSS Agent 3.05 and DSS Server 3.05 capabilities to submit queries to the server or schedule queries to run at a later time. Most queries complete in under two minutes. MCI is investigating migration to the DSS Agent 4.0 product line, which will provide a predictive analysis of query response.

The host data warehouse can be used to perform churn analysis -- or analysis of the reasons people join and leave long distance companies. For example, customers who leave MCI long distance services can be contacted to discern their reasons for leaving: Were they offered a monetary incentive by another company? Were they dissatisfied with service? Did they move? The reasons could be logged into the data warehouse and then analyzed, counter offers developed, and the results of the counter offers reviewed. It may turn out that there is no financial justification for trying to recapture frequent switchers who do not make many telephone calls but who change companies every time they are offered a $20 or $50 check from a competing company. On the other hand, perhaps frequent switchers who ring up a large phone bill each month can be persuaded to stay with MCI with a new type of plan.

Future Directions

WarehouseMCI's growth is evidence of its success. As part of a trend to use warehouseMCI rather than the legacy systems to generate lists and produce actionable information, a new contact management application is being built using the host data warehouse. MCI is in the early stages of testing IBM data mining tools to see if this new technology will work better than existing techniques, and 12 additional data marts are planned for the near future. MCI even envisions that warehouseMCI can be integrated with its front-end operational systems.

The "blocking and tackling" of large data warehouses continues to be a challenge. As warehouseMCI grows, issues such as data quality, use of the host warehouse for multiple applications, time needed to transfer data from the host warehouse to several data marts, query response time, and management of end-user expectations will not go away -- in some cases, they will become more significant over time.

MCI considers its data warehouse and DSDA DSS a competitive advantage because they integrate several sources of customer data, offer the ability to create and apply score modeling quickly, and let end users perform dynamic segment marketing. Other companies should look to this implementation of a large data warehouse, targeted data marts, and innovative DSS as an example of how leading businesses will use database technology in the years ahead.


Peter Brooks is a management consultant with the Advanced Technology Group of Coopers & Lybrand Consulting, based in Boston. He specializes in helping organizations achieve strategic business value by applying solutions that include business intelligence systems, data warehousing, and Internet technologies. You can email Peter at 74477.3043@compuserve.com.


MicroStrategy Inc., 8000 Towers Crescent Dr., Vienna, VA 22182; 800-927-1868, 703-848-8600, or fax 703-848-8610; http://www.strategy.com.

Subscribe to DBMS and Internet Systems -- It's free for qualified readers in the United States
December 1996 Table of Contents | Other Contents | Article Index | Search | Site Index | Home

DBMS and Internet Systems (http://www.dbmsmag.com)
Copyright © 1996 Miller Freeman, Inc. ALL RIGHTS RESERVED
Redistribution without permission is prohibited.
Please send questions or comments to dbms@mfi.com
Updated Monday, November 18, 1996