Rightsizing
Matching provisioned resources to observed demand, rather than to the guess made when they were created.
Instances are sized before the workload exists, then never revisited. The consistent finding across cloud estates is average utilisation well under 20%, which is money spent on nothing.
The routine that captures most of it: review utilisation over a representative window (including peaks and month-end), resize to peak plus headroom rather than to average, schedule non-production environments to stop outside working hours — often 60–70% of their cost — and delete the orphans: unattached volumes, old snapshots, idle load balancers, unused IP addresses.
The instinct that keeps oversizing in place is fear of an incident, and it is answerable with data: right-size against the peak, keep a documented headroom margin, and make resizing routine so correcting a mistake is cheap.