Prediction markets have a lot of applications that go beyond making a direct prediction on the outcome of an event. One of the more sophisticated usecases is a decision making process in a decentralized market-driven way.
Lets say you are building an decentralized exchange that takes fees. Now you might want to built in a mechanism to adjust fees. One way would be to allow shareholders to do a voting. A more elegant way could be to let the shareholder only vote on what metrics should be maximized and than let the free market figure out how. In the exchange example the metric would like be simply the revenue generated - a voting on this might not even be necessary (although an alternative decision could be to optimize the first years purely on trading volume and later on fee revenue).
The genreal idea is simple: first a market is created for n different decision options the DAO has. In this example the alternatives could be: 0.1%, 0.5% fee. The next step would be to create 2 different markets that forecast the revenue of the exchange or more broadly spoken: that forecast the metric the DAO wants to optimize for.
Now you can compare the forecast for the different choices. After a defined trading period the choice with the highest forecast will be chosen.
In this example the revenue forecast is higher for the 0.5% option - so this could automatically be picked. Note that if the actual revenue will be lower the market maker (the DAO) would profit from the prediction (up to the full amount in principal) On the other hand if it is higher - part or everything of the additional revenue will go to the speculators as well.
More on this topic: https://epicenterbitcoin.com/podcast/098/