Working for big tech, our Director wanted to add "ML" to one of our workflows no matter what. He identified a "problematic step" and a team of scientist spent 1 year coming up with a ML model to replace this step. I was asked to deploy this solution in production.

When we tested the model (which somehow hadn't happened before) we found that the workflow was 5% quicker, but the results were more than 10% worse, and was more expensive to run. Everyone hated the idea, but the director kept pushing to do it, "or I'll find someone else to do it". I ran away and never looked back.