Finance reports that your AWS Savings Plans utilization is 100% but coverage is around 40%. What do those two numbers measure, and what does that combination tell you to do next?

## Two numbers, two different mistakes Every commitment programme can fail in exactly two directions, and there is one metric for each. **Utilization** is measured against the commitment. You promised, say, $10 per hour. In a given hour, eligible usage consumed $9.50 of that at discounted rates. Utilization for that hour is 95%. The missing 5% is not deferred and not refunded — it is spent. Persistently low utilization is the expensive failure: you are paying for capacity you do not run, on top of whatever you do run. **Coverage** is measured against your usage. Of all the compute usage eligible for a commitment discount in a period, what fraction was actually billed at commitment rates? If you ran the equivalent of $25 per hour of eligible compute and $10 of it was covered, coverage is 40%. Low coverage is the cheap failure: you are paying On-Demand rates you could have discounted, but you have not lost anything you already committed. A useful mnemonic: **utilization is about the money you promised, coverage is about the money you spend.** Over-commitment shows up in utilization. Under-commitment shows up in coverage. A single number cannot tell you which problem you have, which is exactly why both exist. ## Reading the four quadrants - **High utilization, high coverage.** The target state. Almost all eligible compute is discounted, and no committed dollar is wasted. Keep an eye on it as the fleet changes. - **High utilization, low coverage** — the scenario in the question. The commitments you hold are correctly sized but too small. This is a *good* place to be, because increasing coverage is a straightforward purchase decision with a known baseline underneath it. - **Low utilization, any coverage.** The dangerous quadrant. Something you committed to no longer exists at the volume you promised: a workload was decommissioned, an environment was shut down overnight, a migration moved usage to a family or service your plan does not match. Investigate before buying anything else. - **Low utilization and low coverage together** usually means the commitment is not just too small but pointed at the wrong thing — for example, an EC2 Instance Savings Plan scoped to a family the fleet has migrated away from, while the new family bills On-Demand. ## Why 100% coverage is the wrong target It is tempting to treat coverage as a score to maximise. It is not. Chasing 100% coverage means committing to usage that includes your spikes, your short-lived environments and anything you might delete — and every one of those turns into a utilization problem the moment it goes away. Because Savings Plans cannot be cancelled or resold, that mistake persists for the whole term. Most teams therefore set a deliberate coverage target below 100% — commonly somewhere in the 70–85% range for a stable estate, lower when the fleet is in flux — chosen so that utilization stays essentially pegged. The precise number matters less than the reasoning: commit to the part of the load you would still be running if half your growth plans fell through. ## Turning the diagnosis into an action For the 100%/40% case: 1. Take the eligible usage that is *not* covered, hour by hour, over a representative window of several weeks. 2. Find its trough — the level that uncovered usage exceeds in essentially every hour, including weekends and quiet periods. 3. Subtract anything you already plan to remove, rightsize, move to Spot, or re-platform inside the commitment term. 4. Commit to a conservative fraction of what remains, choosing plan type by how stable the underlying fleet is. 5. Re-measure after the purchase settles and repeat. Incremental purchases are much safer than one large one, because each is sized against observed reality rather than a forecast. ## The nuance that separates senior answers Both numbers are reported over a time window, and the window hides things. A month at 98% utilization can contain a week at 60% if a large environment was down — averaged away, but real money. Similarly, coverage computed across a whole organisation can look healthy while one team's fleet is entirely uncovered. Look at the shortest granularity available and at per-account or per-tag breakdowns before concluding that a commitment portfolio is healthy. And when utilization slips, treat it as an early warning that the fleet has changed shape, not merely as a billing metric — it is often the first place an unannounced migration shows up.