I buy into the argument made by a new white paper by Blue Ridge Numerics that doing a CFD can save tens of thousands of dollars. Even if proof of concept units work well a thermal or structural failure can occur during more rigorous alpha or beta testing potentially wasting the cost of those units or the cost of replacing components or subassemblies.
CFD analysis can answer many questions. Some of the common ones are wind loading, thermal management of electronics enclosure and thermal safety of devices.
I would suggest that this arguments apply to FEA analysis for strength as well.
Rigorous expensive analysis is not required for each and every project though. A skilled engineer knows when rule of thumb or paper analysis has reached its limits and a more through analysis is required.