1. Pick the right people. Workers trained in Six Sigma techniques act like internal SWAT teams, coming together to tackle a given process, then breaking apart and reforming, often in a different configuration, to undertake the next challenge. Depending on what level of Six Sigma training employees have undergone, they're known as green belts, brown belts or black belts, with black belts most often deployed to lead a team or tackle projects on their own.

It's important to tap motivated employees for training, which produces a high-performance team and at the same time signals that Six Sigma is part of an upward career path. "If you grab the C- and D-level players because you can spare them, you're sending the wrong message," says Textron's Bohlen. "We started with the best and brightest, and showed that [Six Sigma training] accelerated their careers. Now I have a waiting list of people wanting to get black belt training."

Be careful about training people but not giving them a project right away to work on, warns Chase Financial Services' Costa. "We have found it very beneficial for people to take a project with them into training or to start something right after," he says.

2. Don't substitute Six Sigma for thinking. For IT staffers particularly, the urge is strong to first use tools they know (that is, technology) before dipping into untried methodologies. Resist that urge, counsels Fidelity Wide Processing's Sutton. "If you just throw technology at a business problem, all you wind up with is a bad process with new technology. If you wring out the process first, then you can really use technology to move it to a higher level."

As a case in point, Fidelity Wide Processing used Six Sigma tenets to reengineer its inbound, paper-based, customer data processing operation. Only after gaining 32 percent improvement by streamlining the process did Sutton's department start to consider new technologies. The company is currently expanding its use of advanced character recognition technology to speed processing of customer data.