That is the perfect high-level overview of my previous job.

I'm semi-retired now...

(edit: I forgot to note that the "let's not" part was always overridden by "You're wrong, this will fix it. Do it!" by management. Then we would eventually find and fix the actual problem (because it didn't go away), but the cluster size -- and the cost -- would remain because "No, it was too slow with so few replicas".)