Saving time and money! And give to inferential statistics a reason to exist ^_^
I'll give you an example: I live in italy. Suppose that a certain company wants to make a study on how italians approach the world of work after completing school: the population will be all the italians (more or less).
Italy has 61 million inhabitants.
So this company would have to make 61 million forms or 61 million calls (ecc..) to collect all the data required. Hope you get my point, that would be impossible!
So it's common sense to extract a sample from the population (it has to be a RANDOM sample!) and then make the study on it.
Once the informations of interest have been extract from the data of the sample, inferential statistics can provide a "final judgement" of what could be the behaviour of the population, with great gain in time and money (and a little loss in accuracy, but it's fair enough ;) ).