September 17, 2009, 5:13 PM — When Raghu Bala, CTO of Automotive.com, joined the company in 2006 to help it build out its business intelligence infrastructure, he put up on his office wall a picture of a cup stuffed with dollar bills and change, with a caption underneath that said, "Panhandling for BI budget."
"I actually got some money out of that," Bala mused during his talk Wednesday at Computerworld's Business Intelligence Perspectives conference. He and other IT leaders shared their tips on how to deploy BI cheaply, profitably and effectively.
A former Yahoo Inc. engineer, Bala was surprised more than once at how much of a "shoestring budget" the company, which operates 200 magazines and Web sites such as Motor Trend magazine, ran on.
When Bala requested to buy a pricey storage area network (SAN) for the data warehouse's storage, management balked.
"One of the guys said, 'I can go to Fry's and buy a bunch of 10 TB drives and string them together. Why do you need to spend $200,000 for a SAN?'" Bala recalled. "I had to explain to him all about snapshot backups, Raid 5, Raid 10."
When that didn't convince his manager, an exasperated Bala sarcastically suggested that he "just go and get a bunch of thumb drives and cascade them together, since they'd be free after all."
Eventually, Bala agreed to have his team build its own SAN out of a cluster of Linux servers. That involved $12,000 in hardware and software costs, and two to three months of time for its lead developer. The result is impressive: a custom operational data store with an "extremely fast" in-memory cache that Bala says can store millions of events per hour.
Bala also saved money other ways by using the free Excel-based BI tools that Microsoft Corp. bundles with its SQL Server database, sharing database administrators with the engineering team, and building cubes for reports only after watching what reports the power users at Automotive.com built for themselves.
This saved IT the labor of having to create and e-mail dozens or hundreds of canned reports, and instead trained the users to "generate their own reports and generally think for themselves," he said.
Know your profitable and unprofitable customers
For companies with long-term customer relationships, one of the best things BI can do is help figure out which are the profitable ones and which are the money-losers.
According to Basil W. Blume, chief analytics officer of Colorado Capital Bank, the average ratio of profitable:unprofitable customers in the retail banking industry is 70:30.
"It's just not profitable when all they have is a $1,000 checking account," he said. "But the other 30% of customers make it up."
If not directed by management to solve this problem, Blume suggests that BI managers step in and draft a plan of how to figure this out.