Data-warehouse projects have a reputation for being complex, costly, and almost certain to fail. Well, they are complex and costly. And the rate of data-warehouse project failures, while down from analysts' estimates of 60% to 90% in the 1990s, still hovers around 40%, according to a recent report from the Cutter Consortium. But more companies are successfully building and operating data warehouses that boost operational efficiency, lower operating costs, and get them closer to their customers.

This year's winners of the Data Warehousing Institute Awards prove that data-warehouse projects can be a critical piece of a company's IT efforts. Here's a look at four of the winners; for more, go to "The Data-Warehouse Advantage, Part II".

Continental Airlines Inc. is realizing numerous competitive advantages with the central data warehouse for which it won this year's award for best enterprise data warehouse.

The original objective of the data warehouse, built in 1998, was to accurately forecast passenger bookings. Before the system went into production, data was scattered across numerous operational systems, making it nearly impossible to pull together information for decision making. "We didn't have the breadth and depth of information that we needed," says data-warehousing director Alicia Acebo.

	Continental's customers want faster access to information, data-warehousing director Acebo says. Photo by Matthew P. Mahon

Today the data warehouse, based on hardware and software from NCR Corp.'s Teradata division, collects data from 41 sources, including flight schedules, seat inventory, revenue and ticketing data, customer profiles and information about One Pass frequent fliers, and employee and crew payrolls. Thirteen hundred employees in 35 departments have access to the system, many using Brio Software Inc.'s query and reporting software, for a growing range of analytical chores.

The data warehouse is primarily used for revenue management, customer-relationship management, fraud detection, and crew payroll-management applications. But the list is always expanding: Acebo's team is adding aircraft parts and maintenance data to the system to help with inventory and purchasing decisions. Continental says that in 2002 it realized millions of dollars in savings with the system and increased revenue by several million dollars.

As user demands for more up-to-date analysis have increased, Acebo's team has adapted the data warehouse to operate on a near-real-time basis. The mainframe and Cobol tools that originally handled data transformation and loading chores have been replaced by a faster network of Windows servers and custom-built C++ tools. Now users analyze flight operations and reservations data that's only seconds old. "Business users at Continental are very savvy. They always want more [information], and they want it faster," Acebo says.

The data warehouse's real-time architecture and automated data-transformation capabilities are two of the best practices that brought Continental the award. The system's design simplifies combining data from different operational areas, making it easy to get a single view of a customer, for example. Standard data definitions developed by a steering committee and used throughout Continental make such cross-functional analysis easier. And revenue and profitability are factored into any decision, no matter how minor.

At Lands' End Inc., the clothing, luggage, and home-products retailer, the number of sales lost because products were unavailable was on the rise in 2001. While service and other facets of the business are important, inventory availability "is really the one that drives satisfaction for our customers," CIO Frank Giannantonio says. Reducing lost sales not only boosts revenue, it improves customer loyalty and retention.

To remedy the problem, Lands' End last year built an inventory-management workbench, a business-intelligence and alerting system that helps inventory planners analyze and monitor the company's available stock. The system, which reduced Lands' End's lost sales by a third during the last holiday season, was the winner of the Data Warehousing Institute's award for best advanced-analytics system.

The inventory-management workbench--built on Business Objects SA's Application Foundation analytic engine and deployed using that vendor's WebIntelligence infrastructure--constantly monitors order-fill rates, back orders, and lost sales. Inventory managers use the system for online reporting and analysis, but it also automatically alerts them when popular items such as white turtleneck shirts need to be resupplied. "That's a very proactive use of business intelligence," Giannantonio says.

The inventory-analysis tool runs on a data mart based on an Oracle database running on a Unix server. That, in turn, taps into the 2-terabyte data warehouse Lands' End built in the mid-1990s using IBM's DB2 database running on a mainframe. The data warehouse holds seven years of customer-order, shipping, and financial information, drawing data from the company's sales, order-fulfillment, inventory, and direct-marketing systems. It also has a link to a massive customer-data warehouse operated by Sears, Roebuck & Co., which acquired Lands' End last year.

The data warehouse's effectiveness lies in its ability to combine information from different operations, such as inventory data from warehouse management and inventory turn-rate data from customer sales, for detailed analysis. It even contains regional weather information to help analysts understand why, say, raincoat sales spiked in May in the Northeast. Altogether, about 600 Lands' End employees directly access the data warehouse through ad hoc queries using Business Objects' software.

With the success of the inventory workbench, Lands' End also has built business-intelligence workbenches for business-to-business sales and to handle analytical chores such as labor scheduling for its warehouse operations. A business-intelligence workbench for sales and service is being assembled, and one for merchandising is planned. The former will provide analysis of metrics such as service levels and call-abandonment rates. Today, some 150 to 200 workers use the workbench data marts.