I run into these all the time in Finland. I have never actuated their buttons. I often have feelings about the quality of the establishment or the experience but they can never be expressed adequately by a 1-5 scale. I suspect the data gathered has a significant bias.
The thing that really annoys me is when the question doesn't match the way the data's used. My ISP asks for a rating out of 10 for "how likely would you be to recommend <ISP> to family and friends" after every support call, but the phone operators explicitly tell you that they're rated on the score that you give them. Those are two different questions - I'd like to rate the phone operator highly, the back-end technical administrators as average (they once assigned the same IP address to myself and another customer - it's disappointing that this is even a manual process!) and the ISP as a whole as 'could be improved' due to their lack of upstream bandwidth at peak times.
It seems to mostly be about knowing "hey, there's a problem of some sort right around here," under the theory that once you know that, finding the solution is relatively easy.
It works that way in debugging software, at any rate.
