Building a machine learning system and maintaining it in production are two very different things. Some folks over at Google wrote a paper that shares their thoughts around all the items you might want to test or check for your production ML system.

Relevant links:
https://research.google.com/pubs/pub45742.html