I used to be excited about prediction markets for a while, but I've since accepted that, except for some niche applications, they won't be terribly useful and possibly even dangerous.

I was ultimately convinced by NNT's arguments, that a) prediction is the wrong thing to focus on (robustness or "anti-fragility" to wrong predictions is) and b) humans aren't good at predicting [the outcomes of complex systems] one way or another. PMs may be better than expert panels, but reality doesn't award extra points for being marginally less wrong.

Prediction markets are also called "information aggregation markets" and in that capacity they may be useful. Assuming all relevant information is out there, but doesn't properly propagate upwards a hierarchy for "behavioural" reasons (e.g. nobody wanting to be the face of bad news), then an anonymous (!) market could be the way for information in the low ranks to move where it is needed. In this case, there is hardly any "prediction" involved, and I assume it could work for that reason.