Customers access PolkInsight through secure Web connections, using Oracle's analysis tool Oracle Discoverer Plus, which creates and displays reports online. Polk chose Oracle Discoverer as its analysis interface because it can easily organize auto data into logical groups of new- or used-vehicle sales, time periods, sales totals, and other data points. "It's a very intuitive, easy-to-learn interface that lets you select the elements you're looking for, hit the Run button, get your data back, and use drag-and-drop to build reports," Topper explains.

Oracle Database 10g Enhances BI Oracle Database 10g introduces several new enhancements that are important for business intelligence and data warehousing. Particularly important for data warehouse administrators is Automatic Storage Management, which brings disk mirroring and volume management inside the database, rather than having it run as discrete activities through operating systems or third-party tools. With this capability embedded within the database, administrators can set up and maintain storage for data warehouses scaling into the terabytes and be assured that their storage will be configured for the high performance necessary in demanding data warehouse environments, says George Lumpkin, director of product management for data warehousing. Oracle Database 10g also delivers better performance. It is faster than previous versions at table scans, the fundamental database operation used in virtually every query to the data warehouse. Low-level optimizations written into the database core have been shown to boost speed in this area by as much as 30 percent, depending on the configuration, says Lumpkin. For better ETL processing, the database platform offers Asynchronous Change Data Capture to quickly read data from the log files of the operational system and pull only data that has been revised into the data warehouse. "You don't want to pull your entire customer table every night from your operational system. What you really want to know is the specific customer records where someone has updated the address, phone number, or other customer attributes," Lumpkin explains. "This feature provides a very efficient mechanism to identify the specific records that have been updated since the last time the system pulled new records." Now available for sites that are implementing their own ETL processing, this feature will also be supported by Oracle Warehouse Builder. Oracle Database 10g Release 2 further enhances the analytic capabilities of the database, with extensions to its OLAP and data mining options, as well as the introduction of a family of statistical functions.

Mapping takes the reports to the next level. End users can zoom in or pan around the maps to track market penetration and find the best spot for a new dealership. "Before they had this capability, customers would have needed two or three days to create reports and generate this kind of map," Topper says.

This means that BI is becoming more ubiquitous for Polk's customers. "Now we have everyone from executives to local field force representatives doing all the analysis on their desks, to find all the information they need in whatever format they need it. It's really amazing to see how many different groups the data is touching, in ways we never expected," Topper continues. "Even engineers are looking at the competition and saying, 'Why is this vehicle beating us in this area?' It could be because the competition has more horsepower and because that performance factor is attracting numerous male buyers between the ages of 20 and 30. So the engineers say, 'We know this is a group of gearheads and horsepower maniacs. Maybe our engines are a little too small and we need to increase performance a bit.'"

Extending BI Across the Entire Company

Although Netherlands-based Randstad Group was a BI veteran, the company realized two years ago that virtually everyone in the company, from top executives on down, needed to use BI. The company provides temporary and contract staffing for clients throughout Europe and North America and realized that it could boost its European operations by giving managers in each country the ability to analyze their local markets without having to send report requests to headquarters.

"We needed reports to be available to more people and generate faster," says Paul Arkenbout, manager of business intelligence. "We decided we needed a business intelligence platform based on the Oracle relational database." Therefore, Randstad switched from a paper-based solution with an aging OLAP implementation to Oracle9i, including Oracle Database, Oracle Application Server, Oracle Warehouse Builder, Oracle Discoverer, Oracle Reports, and Oracle Portal. Today 3,000 staff members rely on summaries created with Oracle Reports and about 30 people, mainly in the Amsterdam headquarters, use the sophisticated yet simple Oracle Discoverer tools. Oracle Discoverer offers analysts the flexibility to build reports themselves.

Randstad's reports are operational summaries or business analyses. The latter help identify client growth trends, new employees, and contract hours. Since the switch to Oracle, Randstad managers are able to access Oracle Portal, choose data relevant to their operations, and generate on-the-fly reports.

This flexibility will soon become available to the thousands of Oracle Reports users. Randstad plans a wide-scale rollout of the revised Oracle Discoverer in 2005, which will shift report building from a core group of report designers in Amsterdam to business users. The Oracle Reports users today rely heavily on Arkenbout's eight-person technical staff to generate reports, which are requested three or four times a month and take approximately three days to complete.

This is an especially big challenge for a technical staff supporting global coworkers who speak a wide variety of native languages. "We're looking at a model in which the power users in each country build reports for the local users and support them in their own language," Arkenbout explains. "We will have more-detailed data available to more people through online access to the information. So we'll move away from reviews of historical data in the data warehouse environment and start working with the increased 'what if' possibilities afforded by business intelligence."

Balancing the BI Workload Using Grid Computing

As business intelligence systems become an integral part of a company's operations, keeping these systems available is critical. For example, three years ago, oil and gas giant Halliburton's Energy Services Group (ESG) built a data warehouse, based on Oracle9i, solely to support its tax accountants. Over time, about 3,000 people in finance, HR, marketing, manufacturing, and business development came to rely on the information held in the warehouse, which now holds about 2 terabytes of data. The warehouse has become so critical to the business, ESG recently boosted its performance underpinnings with a grid computing network. Today, the system provides the "uniform consolidated view of data" used for making financial and strategic decisions, says Gary Gill, senior manager of IT for ESG.

Next Steps READ more about Oracle Business Intelligence products more about Oracle Discoverer LEARN about Oracle BI from Oracle University

Getting at this data wasn't always easy. Almost a decade ago, Halliburton implemented an ERP system to capture operational data and to standardize business processes. But over time, the company began focusing not only on capturing data but on analyzing it to make forward-looking business decisions. This shift and the need for a corporate tax data warehouse to better substantiate and support tax credits and deductions gave impetus to Halliburton's interest in BI. It created its data warehouse and dubbed it TAXWIRE (Tax Worldwide Information Reporting Environment). Prior to the warehouse, tax accountants had to forage through two ERP systems, spreadsheets, and an old accounting system when tax time neared, Gill says. "The warehouse's added efficiency enabled our tax accountants to spend time analyzing tax data, not collecting and organizing it."

As the data warehouse became integral to Halliburton's operations, high performance and uptime became essential. To meet its current and anticipated computing needs, ESG began building an enterprise computing grid network, running Linux on 32-bit Intel Pentium servers for the application server grid and Oracle9i Real Application Clusters (RAC) on 64-bit Intel Itanium-based servers for the database grid. "I wouldn't say we were guinea pigs within Halliburton when we went to [Oracle] RAC, but there is a lot of visibility for this project in terms of potential future deployments of RAC," Gill notes.

Launched last October, the grid architecture is showing initial success. It helped ESG reduce software licensing costs by leveraging the cost benefits of Linux. The operating system also permitted ESG to replace proprietary RISC servers with more economical Intel-based hardware. Grid helps ESG balance workload spikes by reallocating computing resources as potential bottlenecks arise. For example, Gill can allocate additional processing power to an ETL (extraction, transformation, and loading) operation before it slows down the query processing activities. Similarly, as permanent demand for the data warehouse grows, grid scalability means ESG only needs to plug in additional less-expensive computers to add processing power. Finally, clusters offer redundancy for high availability, says Gill. If one server node fails, the remaining nodes keep the applications running.

"We became the data provider to many applications within Halliburton, so performance and uptime are critical to us. With a worldwide user base, we're restricted on scheduled downtime durations. And unscheduled downtime would have a real impact on the business." Gill explains. "That's really one of the driving factors for going with clusters, and so far, it's paying off." In all, it's an intelligent way to keep an intelligent system running.

Digging Deeper for Data Mining Gold Three years ago, the Oracle database became a unique business intelligence platform, when an OLAP engine and data mining algorithms became part of the core relational database engine. Now Oracle 10g's Data Mining option supports a wide range of data mining algorithms, for classification and prediction; clustering algorithms, for revealing hidden market segments; association rules, for "market basket analysis"; and attribute importance, for identifying key factors associated with a business problem. Today, Oracle Data Mining provides basic tools for finding patterns hidden within database information. In the upcoming Oracle 10g Release 2, three new capabilities build on the database's analytical capabilities, giving users new data mining algorithms and modeling, building, and scoring functions. The first addition is decision trees, a widely used type of data mining algorithm. Decision trees can piece together trends and characteristics in data to find clues valuable for strategic business decisions. By mining information collected from store affinity cards, analysts for a retail chain might use decision trees to find customers likely to respond favorably to the affinity program. Additionally, decision trees can provide detailed explanations or profiles associated with each customer group. For example, a decision tree may find that likely responders are females who have visited the chain's stores within the last month and will probably surpass a certain spending threshold. The resulting customer profile can be the basis for a new marketing campaign. "The algorithms we have today can do similar things in terms of making predictions, but they are a little 'black-boxish,'" explains Charlie Berger, Oracle's senior director of product management for data mining and life sciences. "They're not as transparent as decision trees in terms of presenting the rationale for their predictions. One of the attractions of decision trees is that they provide fairly detailed rationales for each prediction." The second new capability centers on SQL predictions and the ability to score and rank predictions. For example, analysts for a telecom company might rank customers according to their likelihood to "churn" or switch to a competitor. "The scoring is lightning-fast; the database performs it as a standard database operation," Berger says. Comparing the process with systems that rely on a database and standalone mining and scoring applications, he adds that the Oracle 10g approach means "you don't have to move the data out to a separate analytical server and you don't need a data analyst to build a model and do the scoring." The third new feature is an enhancement to the Support Vector Machine algorithm that improves the handling of so-called one-class models, which look at clusters of data and identify data points that are different from the rest.

Bridging the OLAP and Relational Worlds OLAP data and relational data used to be like oil and water: They never mixed. When the Oracle database incorporated an OLAP engine into its standard platform, in 2002, the processing distinctions between the two data types began to blur, at least in the behind-the-scenes architecture. Now, with the latest version of Oracle Discoverer, the marriage of these two data groups has taken center stage. "People want to have one environment, one interface—and train their users only once" notes Christina J. Kolotouros, senior manager of business intelligence solutions product management at Oracle. Traditionally, companies interested in analyzing multidimensional data had to install point-solution OLAP management and reporting tools. "The problem is that those tools couldn't work with the relational data. So if they had data in their ERP or CRM systems that was based on a relational database, they couldn't access it," says Steve Illingworth, senior director of Oracle business intelligence solutions. "They had to take the data out of the ERP or CRM systems and move it into the OLAP systems." Similarly, relational reporting tools were blind to multidimensional data. Bridging this gap, Oracle Discoverer provides a single environment for analyzing both data types and even combining data from both areas into a single analysis. To make this happen, Oracle Discoverer was reengineered to recognize and understand OLAP APIs embedded within the Oracle database engine, besides standard relational database APIs and SQL calls. "We had to recognize and build to go against the OLAP APIs that are now part of the database. Oracle Discoverer had to know and recognize some new functionality of multidimensional analysis, such as ranking and forecasting," Illingworth explains. A single interface for data consumers isn't the only advantage of this approach. Because all the information processing happens within the core of the Oracle database, the performance optimizations built into the platform, the processing power of the high-end server hardware, and the additional analytics in the database come into play. "We see higher performance, because the database returns results to desktop machines rather than sending the data across the network for processing by the front-end hardware," says Illingworth. While innovations in Oracle Discoverer bring new end user capabilities to multidimensional data processing, unseen but significant performance boosters are enhancing Oracle Database 10g's OLAP engine. Historically, relational data types have been more scalable than their multidimensional counterparts. In Oracle 10g Release 2, Oracle introduced new capabilities for multidimensional data types that let them scale for large data sets. Some of these capabilities, including partitioning parallelism, are familiar to relational DBAs. One is unique to multidimensional data types. "We introduced a completely new storage algorithm for aggregate-level data," says William Endress, director of product management for OLAP, "that is extremely efficient for cube building in preparing data for queries." Oracle benchmark tests have shown that Oracle Database 10g requires only about an hour to build multidimensional data sets that used to take 24 to 36 hours, Endress says. This lets customers process multidimensional data more quickly and perform more-detailed analyses, he adds.