At first I thought this technique sounded like it would work very well for determining if, for example, a European user was using an American IP, but not work well for bordering countries, simply because I assumed the error margin would be fairly large. However, according to the source paper, the error margin is only a few kilometres. Impressive.