I replaced most of our custom models (build on scikit-learn) with this. Pros: - Super simple API, you can be up and running in a few hours - I don't need to

I replaced most of our custom models (build on scikit-learn) with this. Pros: - Super simple API, you can be up and running in a few hours - I don't need to maintain machines/code to do all the training and querying (huge win for us) - Relatively fast - Very limited ways to influence the models it builds Cons: - Like other Google APIs, it 500's from time to time (~0.5%) for no good reason - It's a big black box, doesn't expose much information about the models it builds other than a confusion matrix. - Very limited ways to influence the models it builds If ML is critical to your business and you know what you are doing, you can get better results with a custom stack.