Measuring the effects of the Web

One of my clients is looking at the Web as a means to deflect calls away from the call center (aka online self-service). Nothing new. They are, however, very keen to find an air-tight way to prove that the investments in the E-technology are recouped by calls avoided. Conversely, if the call volume goes up, they want to show whether it is related to the Web or not.
Any idea how to translate web measurements into calls avoided???