How Do You Set Up Retraining for Your Pipelines?

I call it the shoemaker son goes barefoot, right? Because we all of us were data people. And I think 90% of the time I get seasonal like frequency based retraining. Now obviously it's not robust in any way, right? because seasonality occurs and ongoing world is like that is shifting no matter what you'll do. But I would consider maybe two or three strategies to observe your model,. Interesting is distribution shifts, you need to make sense of the difference between training data set and production data flows into your model. Second thing is integrity based or missing values or most frequent value that I don't know for some reason, exceed is it threshold and so on.