Roger,

What are the two variables?

What you don't mention is some way to correlate questions initiated on the robot with questions initiated via phone. For example, does every question start with the robot? Is there a log for the robot? Any unique identifier such as incident number attached?

If nothing else, do you have some measure of call volume before and after the robot was started? If not, turn it off for a few days ... for maintenance if you need an excuse and look at the difference in call volume. Probably want to first look at call volume and have a good handle on it's time dependence, e.g. big peak in calls just after lunch local time on tuesdays, very few calls on Sunday. Then compare with/without robot for comparable times.