[abhgh]

Thats a sweeping statement which, in many contexts, is in error. A given ML algorithm is not effective with any kind of data or problem. Typically you need to understand the domain to pick the right algorithm and/or features and to tweak it right.

Of course, a case could be made for a product where prediction is just one not-so-significant feature, and here you can use something off-the-shelf without worrying too much about performance. But where predictions are the driver, you would need to put in some serious work to get good results. (not talking about Grid here - this is a general statement)