This post makes the good point that companies would have far better return if they look for ways to collect better input data rather than poring all the effort into getting minuscule improvements from the generic recommendation algorithms.