Even in Asimov's world, psychohistory only works on groups that don't practice psychohistory themselves. Harry Seldon only kept things from going off the rails by making the science die out, and by starting a Second Foundation of telepaths.
Once someone starts making predictions from data aggregation more effective, the race will be on to duplicate or improve on it, and then nobody's prediction algorithms will work.
Almost sounds like someone should write a dystopian Foundation book, where the mathematicians race to predict each others' predictive abilities (and of course, stop them!)