The reason why the AdS/CFT correspondence is interesting is not that AdS space is supposed to describe our universe, which, as you have correctly pointed out, would lead to conflicts with experiments. In the context of the correspondence, a four-dimensional (conformal) field theory is mapped to a string theory living in an $AdS_5\times S^5$ space, although there exist generalizations in which the AdS part is of higher or lower dimension than five.
This duality in principle allows one to carry calculations from one side to the other, making it possible to choose the framework in which the solution to the problem at hand can be found conveniently.
One key observation in this context is that the duality can map a strongly coupled theory to a weakly coupled one, circumventing the failure of perturbation series. This is especially interesting with respect to QCD, where a a conventional perturbative low energy description is not possible. Even though an exact holographic dual of QCD is yet to be found, there are theories (for example the Sakai-Sugimoto model) that capture important features of QCD surprisingly well.
One may now ask what is so special about AdS space that allows for such a duality? One way to approach this is to point out the rich symmetry content of this kind of spacetime. The isometry group of Anti-de Sitter space is given by $SO(4,2)$, which is precisely the conformal group in four dimensions.
Regarding de Sitter space: the nature of this spacetime makes it difficult to formulate a correspondence analogous to its positively curved counterpart. See this article for more information.