100% utilised can still mean badly placed.
Coverage and utilisation are the two numbers most often quoted about a commitment portfolio, and the two most often confused for each other. They measure opposite failure modes, and reading either one alone reliably produces the wrong conclusion.
Two metrics, two opposite failures, routinely reported as one
Utilisation measures the use of purchased commitment. Coverage measures the eligible consumption receiving commitment benefits. Review both alongside future demand: low coverage can be a deliberate choice to preserve flexibility.
Perfect utilisation, poor coverage
Everything you bought is being used, and most of your eligible spend is still paying on-demand rates. The portfolio looks healthy on the metric being watched while the majority of the opportunity sits untouched.
Strong coverage, weak utilisation
Most eligible spend is nominally covered, but part of what was purchased is not being consumed — so you are paying for commitment that produces no benefit while simultaneously appearing well covered.
Both look fine, in aggregate
Aggregate figures average across instrument types, regions and families. A commitment locked to a family you are migrating away from can be stranded inside a healthy-looking portfolio average, invisible until it expires.
The measurement basis changes the answer
AWS defines these metrics for each instrument. Keep those definitions visible and distinguish any additional internal measures so unlike percentages are not compared.
- 01
Savings Plans coverage uses On-Demand-equivalent cost. RI coverage may use instance hours or normalised units; label the basis before comparing instruments.
- 02
Eligibility follows the instrument rules. Check the included services and configuration scope rather than treating all AWS spend as eligible.
- 03
Benefit application follows AWS ordering and sharing rules, including account and instrument scope; it is not an unrestricted allocation to the highest discount.
- 04
Averaging over a month hides intra-month and intra-day shape: a commitment fully utilised on weekday peaks and idle at night can average acceptably while wasting materially.
- 05
Multiple overlapping commitments make attribution order-dependent, so per-instrument utilisation figures depend on the order benefit was applied.
- 06
A target figure borrowed from a benchmark is meaningless without the risk appetite behind it — the right coverage level for a volatile estate is not the right level for a stable one.
How we measure it properly
We measure both metrics on a stated basis, at the granularity where stranded commitment actually hides, rather than at the portfolio average where it does not.
- 01Coverage measured in spend and in hours, both reported, with the basis stated
- 02Utilisation per individual commitment, not portfolio-averaged
- 03Hourly utilisation profile to expose intra-day and weekday/weekend shape
- 04Coverage by instrument type: Compute Savings Plans, instance-locked plans, Standard and Convertible RIs
- 05Coverage by region, instance family and size, to find locked positions drifting from actual usage
- 06The eligible base definition itself, stated explicitly and tested for over-breadth
- 07Benefit application order across overlapping commitments
- 08Trend across the full portfolio history, separating structural change from seasonal movement
- 09Stranded commitment: purchased capacity whose matching workload has moved or ended
- 10Realised effective discount rate versus the rate implied at purchase
What we determine
Which failure mode you actually have
Under-commitment, over-commitment, or misplacement — three different problems with three different remedies, frequently reported under one heading and treated with the wrong one.
Where commitment is stranded
Specific commitments whose matching consumption has moved region, changed family, or gone away, identified individually rather than averaged into a portfolio figure that conceals them.
Your realised discount rate
What the portfolio actually achieved against list, compared to what the purchase decisions assumed. The gap between those two is the honest measure of commitment performance.
The right coverage target for this estate
A defensible target derived from your own consumption volatility and business risk appetite, rather than a benchmark percentage borrowed from an estate that behaves nothing like yours.
Measured properly, these two numbers tell you whether your next move is to buy more commitment, to restructure what you hold, or to stop buying and fix the consumption underneath it first.
- Whether the estate is under-committed and how much headroom remains before risk rises
- Whether eligible Standard EC2 RIs can be resold, or Convertible RIs exchanged; Savings Plans are not resale instruments
- Whether a low coverage figure reflects real opportunity or an over-broad eligible-base definition
- Whether utilisation gaps are structural or a shape problem that scheduling can fix
- What coverage target to hold going forward, justified by your own volatility
- Whether commitment performance justifies the strategy that produced it, or a change of approach
What we hand over
Dual-metric position
Coverage and utilisation reported side by side on a stated basis, at instrument granularity, with the failure mode named rather than left to inference.
Stranded-commitment register
Every commitment whose matching consumption has moved or ended, with remaining term, residual value and the recovery route available for each.
Target and rationale
A coverage target derived from your consumption volatility, with the risk trade-off written down so the number can be defended rather than merely quoted.
Measure both
Stage 00 is a 30-minute qualifying call at no cost. If the timing or the estate does not justify an engagement, we say so on that call.