Ask how a platform was sized and the honest answer is often that it resembles the previous one with more of everything. It is a defensible instinct — over-provisioning rarely causes an outage — but it produces estates that are expensive in aggregate and still wrong in specific places.
Measure before you specify
Utilisation, peak concurrency, IOPS profile and growth trend are usually available from the existing environment and rarely consulted. A fortnight of collection ahead of a design produces better decisions than any amount of vendor sizing guidance.
Headroom is a design decision
Failover capacity, growth allowance and burst headroom should each be stated explicitly and justified, rather than bundled into a general margin. Stated separately, they can be discussed; bundled, they are simply cost nobody can defend.
Right-sizing is not a one-off
Workloads change. A platform sized correctly at design time will not stay correct, which is an argument for periodic capacity review rather than for buying three years of margin up front.
In short
Collect real utilisation, state each margin separately, and revisit capacity as a routine.