I love this post — thank you, Taylor. Very good stuff.

In my experience, we cannot avoid the data science side. As you stated, the “the world of data science and AI” creates new problems that add another layer of complexity. Feature coding, model assumptions, and constraints that were added because of data pipeline decisions in deploying the model lead to significant context loss when users try to decide when it is appropriate to use a feature for a particular business question and when it is not.

I wish we had a way to represent complex context to guide users in appropriate use of measures that are the result of models with many embedded assumptions and constraints. I do not know the answer to this. Loss of context is hard to live with.