Cloud's costliest mistake isn't failure but neglect. Someone provisioned m5.4xlarge "generously" three years ago; the service actually needs m5.large but no one revisits. CPU averaging 4% runs 24/365 at full price. On-premises this waste stayed hidden—server was sunk cost. Cloud bills hourly, so over-provisioning hits invoices instantly. Right Sizing is narrowing the gap: provisioned capacity vs actual usage.
The problem: humans can't eye-measure. CPU alone misses memory; averages miss peaks; one instance misses two hundred