@pschwede You are actually getting into my profession even closer with this question. Most of what I do is predictive analysis of some form when hired by a client.

So as a general rule the more stats you have relevant to the question the better you can predict the question.

So in this case presuming ALL you know is the past sales data, that in isolation is a rather limited amount of information. While you can make some prediction of the future off of thatyou'd find you would have a high error rate. For example you might notice that every year around christmas you see a surge of about 10x your usual sales. You could therefore predict the same would be true next year. While that is somewhat likely to be true without knowing all the other factors involved it would be hard to discount the possibility that this year might be different.

However if we have more information then we can predict the future better. If for example we also had data about what the weather was over that same period then we can use this extra information to make our predications more accurate. We might, for example, see that around christmas time our success is partly dependant on if it is snowy out or not, when it snows around christmas the higher sales are more likely than on christmases when it doesn't snow. So now we can make even more accurate predications and the error rate is even lower.

Best yet in statistics we know what our error rate is based ont he sample size and deviation. So if we make a prediction we know how confident we are in making it.

Thats what is nice about statistics, we know how much we dont know and when we dont know, and if we need to collect more data before we can know.