The first description I ever heard (from my econ professor)of a practical use for econometrics was of a student who used one variable to predict electricity demand. His finding was so robust that Commonwealth Edison hired him before he even graduated. It is likely that the model is more sophisticated now.
I don't know the case, but I guarantee you that he didn't just estimate a single variable regression. To establish a robust relationship, you have to throw in all other plausible variables to ensure that the one being tested is what's driving things and not something else with which it is correlated.