- Coverage = percentage of your eligible resources covered by a commitment.
- Utilization = percentage of that commitment that is actually being used (i.e., matched by running usage).
- Neither metric alone tells you whether you saved money. You can have 100% coverage and lose money. You can have 100% utilization and leave money on the table.
- Savings live in the ratio between the two, adjusted for how your workload actually runs. The real metric is Effective Saving Rate (ESR) — covered in article #05.
The two metrics, defined
Before the trap, the definitions.
Coverage is how much of your eligible usage is sitting under a commitment. If you run 5 EC2 instances and you bought RIs for 4 of them, your coverage is 80%.
Utilization is how much of the commitment you actually burn. RIs are paid hourly, whether or not an instance matches. If the covered instance runs 10 hours out of every 24, your utilization is 10 / 24 = 41.7%.
5 running instances RI commitment (covering 4 instances)
┌───┐┌───┐┌───┐┌───┐┌───┐ ┌───┐┌───┐┌───┐┌───┐
│ A ││ B ││ C ││ D ││ E │ │ A ││ B ││ C ││ D │
└───┘└───┘└───┘└───┘└───┘ └───┘└───┘└───┘└───┘
Coverage = 4 / 5 = 80%
Per-instance over 24 hours:
┌─────────── RI hour-slots (24) ───────────┐
│ ████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░ │ Instance ran 10h
└──────────────────────────────────────────┘
Utilization = 10 / 24 = 41.7%So far, so good. Now the interesting question: if both metrics are high, does that mean you saved the most money? Not necessarily.
Does high coverage mean high savings?
It looks obvious. More coverage = more resources getting a discount = more savings. So just cover everything, right?
No. Here is why.
On-Demand: $1/hour per instance. RI price: $0.5/hour per instance.
Workload: 4 instances running 24/7 for a month (720 hours)
┌─────────────────────────────────────────────────────────────────┐
│ Scenario 1: 75% coverage (3 RIs for 4 instances) │
│ │
│ 3 × $0.5 × 720 = $1,080 (RI-covered) │
│ 1 × $1.0 × 720 = $ 720 (On-Demand) │
│ Total = $1,800 │
│ │
├─────────────────────────────────────────────────────────────────┤
│ Scenario 2: 100% coverage, but bought 6 RIs (over-committed) │
│ │
│ 4 × $0.5 × 720 = $1,440 (RI fully used) │
│ 2 × $0.5 × 720 = $ 720 (RI paid, no match, WASTED) │
│ Total = $2,160 │
└─────────────────────────────────────────────────────────────────┘Scenario 2 has 100% of instances covered but costs $360 more, because you paid for commitments that had nothing to match against. Coverage is a count, not a result. Overcommitting pushes the count up. It does not push savings up.
Does high utilization mean high savings?
Flip it. If overcommitting is the problem, just undercommit until every commitment is used 100% of the time. Surely that is the answer.
Also no.
Same pricing. Same 4-instance workload running 24/7 for 720 hours.
┌─────────────────────────────────────────────────────────────────┐
│ Scenario 3: 25% coverage (1 RI), 100% utilization │
│ │
│ 1 × $0.5 × 720 = $ 360 (RI fully used) │
│ 3 × $1.0 × 720 = $2,160 (On-Demand) │
│ Total = $2,520 │
│ │
│ RI is 100% utilised — but only $360 of spend saw a discount. │
│ You left $1,440 of discountable spend at On-Demand rates. │
└─────────────────────────────────────────────────────────────────┘Scenario 3 has perfect utilization and is the most expensive option. Utilization is a score on the commitments you bought, not on the decision to buy. You can score 100 out of 100 on something that was too small to matter.
So what about high coverage AND high utilization?
Now it gets tricky. From the two cases above, the sweet spot is clearly somewhere in the middle — enough coverage to discount most of the workload, but not so much that commitments sit idle.
But here is the catch: utilization is only 100% if the workload runs 24/7. If your compute goes up and down during the day — which is the whole point of cloud — then utilization is a moving target.
24/7 workload Variable workload
(data center-like) (typical cloud)
Demand over a week:
┌──────────────┐ ████████████████ ██▄▄██▄▄██▄▄▄▄▄▄
│ Mon → Sun │ ████████████████ ██▄▄██▄▄██▄▄▄▄▄▄
└──────────────┘ ████████████████ ██▄▄██▄▄██▄▄▄▄▄▄
Ideal strategy: Max coverage. Coverage must match the
Utilization = 100%. floor, not the peak.For a workload that actually runs all the time, the answer is easy — buy commitments up to the ceiling. But most cloud workloads are not like that. They have off-hours, weekend drops, seasonal demand, growth. Coverage and utilization become two variables that wobble together, and the “right” commitment level changes with them.
Now try to graph the optimal mix across Standard RIs, Convertible RIs, and Savings Plans over a 1-year or 3-year horizon, with multiple workloads, for dozens of instance families. That is the real problem. It is not something you solve with two metrics on a dashboard.
The verdict — what actually matters
Coverage and utilization are inputs. They are useful diagnostics: low coverage means you are leaving easy savings on the table, low utilization means you bought too much. But neither of them tells you how much money you actually saved versus the alternative.
The metric that does tell you that is Effective Saving Rate (ESR): the ratio of your actual spend under commitments to what you would have spent at pure On-Demand rates. ESR collapses coverage and utilization into the number that matters — dollars saved per dollar of eligible spend. Article #05 walks through the formula.
In the meantime, two rules that hold up regardless of which metric you track:
- Do not buy commitments because the discount rate looks big. Buy them against spend you are confident will be there.
- Do not run resources 24/7 to “justify” a commitment. That reverses the logic. The commitment exists to discount real usage. Inventing usage to match a commitment is how you lose money while looking efficient.
TIP: Stuck on finding the right balance? Send me a note — happy to look at your specific case.
Summary
- Coverage and utilization describe the shape of your commitments. They do not measure savings.
- High coverage + low utilization = overcommitted. High utilization + low coverage = undercommitted.
- The right balance depends entirely on how elastic your workload is. 24/7 is easy. Variable is hard.
- ESR is the metric that ties everything together — use it as the success criterion, not coverage or utilization in isolation.
Thanks for reading. If this helped, share it. Questions or topic suggestions — send them through.