Intelligent Enterprise

Philip Bierhoff, Systems Manager at Procter & Gamble, spoke at last week's FASTforward conference about strategies to increase user adoption as business intelligence goes mainstream [Editor's note: it's a topic very much at the center of Cindi Howson's recent feature on "Pervasive BI"]. P&G's Symphony project creates "decision cockpits": dashboards based on specific roles and corporate divisions, and including information ranging from traditional BI reports to documents to news.

The underlying data landscape has moved from their first iteration of a common data warehouse in the mid-'90s with regional servers plus ETL, storage and aggregation, where BI was driven by stored aggregations; to the current atomic data warehouse with a central server plus ETL and storage, where BI is driven by query rewrite — effectively, aggregation on the fly. They also have SAP generating data into SAP/BW; altogether, they have about 65 TB in the data warehouse and 50 TB in SAP/BW.

Originally, P&G used a traditional monolithic BI approach using a BI tool that acted as a portal as well as providing dashboards, graphs and reports as well as its own security layer. They had some challenges with scalability, and the boundary behavior when they pushed the size of the envelope. They felt restricted by the functionality of the tool and by the lack of flexibility. Based on these restrictions, they developed a new vision of BI based on service-oriented architecture (SOA) and industry standards, working directly from the corporate data warehouse instead of reporting data marts. A thin layer of BI is one of multiple layers of technology that can be changed out as required. The resulting solution uses an SOA-based (I assume JSR168-compliant) portal with various portlets, fed by FAST Radar for BI repots and graphs, and SharePoint for document storage. Under all of this is a service bus connecting to a data services platform, which in turn connects directly to the data warehouses and legacy applications. A WS-Security stack sits alongside to provide standard security and authentication.

Bierhoff says P&G selected FAST Radar because it pairs simplicity with power for fast response times, fits into a portal architecture, downloads to common formats such as Excel and PDF, requires no data model maintenance and can use Web services as a data source in additional to traditional database access. The company has found that FAST Radar needs very little training for creating and using reports, says Bierhoff, although they're still using it only for predefined reports with drill-downs rather than allowing ad hoc report creation. The time to create complex reports has dropped from 30 to 60 seconds with the old BI tool to five to ten seconds using FAST Radar.

The resulting architecture looks like a textbook example of how to do SOA correctly. They're getting a good level of reuse of services, especially the data access services, and they're positioned for future extensions without completely overhauling the architecture. They're able to provide customized dashboards for their 40,000-person user base in a reasonable time frame, in a large part because they've designed the underlying services and components for reuse.