Commitment Coverage
The proportion of steady-state usage covered by discounted commitments, balanced against the risk of committing to capacity that is no longer needed.
Providers offer substantial discounts — commonly 30–70% — in exchange for a one- or three-year commitment. The decision is a forecast, and coverage is how it is managed.
Cover the baseline, not the peak. The workload's floor over the past year is what can be committed safely. Anything above it belongs on on-demand or spot pricing, where flexibility is worth the premium.
Target coverage of roughly 70–85% of steady-state usage for most organisations. Higher coverage increases savings and increases the risk of paying for capacity after an architecture change removes the need for it.
The commitment types differ in ways that matter: instance-specific reservations give the deepest discount and the least flexibility; flexible compute commitments apply across instance families and sometimes across services, trading a few percentage points for the ability to change architecture without stranding the commitment.
Two operational requirements: monitor utilisation of existing commitments, since an unused commitment is pure waste and is invisible without a report; and stagger expiry dates rather than letting a large tranche mature at once, which turns renewal into a rushed decision.
The counter-consideration worth stating: a three-year commitment on a workload you intend to re-architect is a bet against your own roadmap.