The Hidden Cost of Over-Modeled Analytics
Clean schemas are useful. But when modeling matters more than the user's job, analytics becomes expensive theater.
There is a point in many data teams where modeling stops serving the business and starts serving the model.
You can feel it. Every new question requires a new dimension. Every answer waits on a refactor. The semantic layer is beautiful, and the business is starving.
When aggregation breaks trust
Over-modeling usually hides behind good intentions: consistency, reusability, a single source of truth. But each layer of abstraction is also a layer between the user and the raw fact. When users cannot reconcile a number to something they recognize, they stop trusting it — no matter how elegant the lineage.
Star schemas are not a religion. They are a tradeoff you should be able to defend out loud.
Modeling for meaning vs. modeling for performance
These are different goals and they pull in different directions. Conflating them produces models that are fast and meaningless, or meaningful and unusable. Name which one you are optimizing for, per use case, and accept that the answer changes.
A cheaper path
Model the things that are stable and shared. Leave room for the messy, specific, fast-moving questions to be answered closer to the source. Not everything needs to graduate into the warehouse's permanent vocabulary.