In-Depth

TDWI Conference Kernels: Monday Keynote on Extending BI Architectures with Enterprise Integration Technology

When the Group Chief Architect for ING talks, TDWI attendees listen.

• By Eric Kavanagh
• 02/22/2006

A record turnout at TDWI’s winter conferences in Las Vegas heard some unconventional wisdom from a stalwart information architect on Monday. Raymond Karrenbauer, Group Chief Architect for ING Group, offered both high-level insights and granular takeaways on the increasingly important discipline of enterprise integration, in particular enterprise information integration (EII).

“Companies today don’t really compete with each other,” he said, “not in the size that you have with an ING. Supply chains compete with each other.” He continued: “Information is absolutely critical to the success, efficiency, and response of an organization. Companies compete on analytics, which means technology. That’s why EII is so crucial, no matter what scale. So, are you positioned?”

What is EII, exactly? Karrenbauer provided a general definition, then outlined several key components that comprise a viable EII solution. In essence, EII is the integration of data from multiple systems into a unified, consistent and accurate representation geared toward the viewing and manipulation of the data. Effective EII solutions, he says, contain the following elements:

• standard data model;
• information hub
• shared data
• reusable infrastructure
• common processes
• centralized security
• information quality management
• lower cost in the mid- and long-term

On the flip side, the ING architect noted that there are key elements which are decidedly not EII in nature, such as:

• customized data models
• point-to-point, no hub
• redundant data
• incremental infrastructure
• specialized process
• distributed security
• limited quality control
• lower cost in the short-term

Karrenbauer explained that a common approach is paramount to enabling integration across a far-flung organization. However, he cautioned against choosing one model and forcing everyone to stick to it. Different methods work in different scenarios, and it’s important to allow some flexibility to accommodate various needs. Still, whichever methods are chosen, they should be polished, understood and followed to ensure a high level of data quality.

What constitutes a high level? In the realm of data quality, the metric is error rates: how much bad data exists within an organization’s repositories. Karrenbauer says ING shoots for—and attains—very good marks in that category. “You want to hit about two percent error rates,” he said. “That’s what we’ve experienced. We’ve gone from 18 percent to two percent.”

How did they do that? Karrenbauer says a strict regimen of monitoring data is key, including an understanding of how and where data is captured. It’s also important that data quality, in and of itself, be evangelized throughout the organization as a primary goal. “The value of data quality cannot be underestimated,” he cautioned.

Another key to ING’s success, says Karrenbauer, is their approach to integration. A global organization, ING doesn’t actually share data from one region to another. Rather, they share the infrastructure, methodologies and processes. That, he says, allows for overall consistency across the organization, without placing undue limits on any given region or department. “The first rule of integration is that the next business unit always wants a solution that is slightly different,” he explained.

On the subject of BI architectures, Karrenbauer denounced the one-size-fits-all model. “There are probably a thousand different BI architectures to choose from,” he said, “but there’s only one that’s right for your organization.”

And in an impressive bow to total disclosure, the TDWI keynoter provided some time-tested kernels of research on the subject of optimizing an EII-style information hub: “We’ve come to conclusion, in a hub, the most optimized number of data stores is two; the optimal amount of warehouses within a hub is one; and the optimal amount of data marts within a hub is somewhere between 12 and 15.”

Be sure to jot that down—it will be on the test.