Michael Cutler, cofounder of TUMRA, gave a nice talk to the University of Oxford Computer Science Department. The following quote from his talk sums up his idea.

Given a choice between a “best guess” now, and a “marginally better” answer later, I’d take the best guess every time.

Many times, academic people focus a lot of attention on improving the accuracy of an algorithm, when the resulting solution is too slow for industrial purposes.

reference: TUMRA Blog