Blog 1 established what your estate can safely commit: the Minimum Commitment Threshold, and the ladder above it. This post adds the second metric — the Effective Savings Rate, the number the whole series is denominated in — shows that it decomposes into two independent factors, coverage and discount, and derives the three levers that move them. By the end you will be able to tell whether your savings gap is structural, an execution gap, or a pricing gap — and what one point of it is worth in dollars.
The Appeal of ESR
Effective Savings Rate (ESR) is the percentage of your commitment-eligible on-demand equivalent AWS spend that you did not pay because commitments reduced the bill.
Start with the counterfactual: what would this eligible usage have cost at on-demand rates? Then compare it to what you actually paid after Savings Plans, Reserved Instances, and other commitment discounts applied. The gap is your effective savings. Express that gap as a percentage of on-demand equivalent spend, and you have ESR.
Defined precisely:
ESR = 1 - (actual_spend / on_demand_equivalent_spend)
Take Alpenglow Outfitters, our fictional outdoor-gear e-commerce, AWS-native since 2018, ~450 people. Its commitment-eligible on-demand equivalent spend is ~$8.0M/year (precisely $7.884M). Alpenglow covers 35% of that eligible ODE with commitments, and at a ~30% blended discount that works out to an ESR of 10.5%: after Savings Plans and Reserved Instances it actually pays $7.056M, so $828K is avoided relative to the on-demand counterfactual. Tomás, the analyst who owns Alpenglow's cost dashboards, can report both numbers today.
The ESR gives to FinOps practitioners something coverage and utilization never could: one outcome metric for commitment performance that mapped directly to money saved.
That is why ESR feels like the right metric.
It is an outcome, not an input. It is expressed in dollars, not in abstract utilization percentages. It answers the question leadership cares about: of the eligible spend that could have been paid at on-demand rates, how much did we avoid paying because we used commitments well?
That makes ESR a better north star than the metrics AWS puts in front of you by default. Utilization tells you whether the commitments you already bought were filled. Coverage tells you how much usage received a commitment discount. Both are useful signals, but neither is the economic result. ESR is closer to the thing you actually want.
That is clean. It is defensible. It is easy to put in front of a CTO or CFO. It is also the number this whole series is denominated in: every recommendation in the final post is priced in ESR points.
What ESR is not is a diagnosis. On its own it cannot tell you what to do next, because any ESR reading is the product of two independent factors — and it is the factors, not the total, that have owners, fixes, and risks. So the move is not to replace ESR with something better. The move is to decompose it, and to know which lever moves which factor.
One ESR Point Is Real Money
Before getting into the diagnostic problem, it is worth stating why ESR deserves attention at all.
ESR is not calculated against the entire AWS bill. It is calculated against the on-demand equivalent spend for the commitment-eligible segments you are evaluating. Compute has its ODE. Database has its ODE. SageMaker has its ODE. Each segment is measured on its own eligible usage, then the segment-level ODE values can be rolled up into one aggregate denominator for the overall commitment portfolio.
Once that denominator is defined, one ESR point is one percentage point of eligible ODE. The dollar value is simple:
annual eligible ODE spend x 1%
| Annual eligible ODE spend | Value of 1 ESR point |
|---|---|
| $1M | $10K/year |
| $5M | $50K/year |
| $7.884M | $78,840/year |
| $10M | $100K/year |
| $50M | $500K/year |
For Alpenglow, with $7.884M/year of eligible ODE, one ESR point is worth $78,840/year. Moving ESR from 10.5% to 11.5% is roughly $79K/year; closing the gap to the 13.5% ceiling Alpenglow's minimum level supports is three points, roughly ~$237K/year — the kind of single number Walter, the CFO, wants on one page. This is why commitment strategy is worth treating as an engineering problem rather than a quarterly procurement ritual.
One arithmetic warning, because it is the most common error in this space: never multiply an ESR point by the discount rate. ESR is already defined against on-demand equivalent spend, so the discount is inside the number. One point is one percent of eligible ODE, full stop.
ESR Decomposes
ESR is not a metric to replace. It is a number to decompose:
ESR = Covered ODE % x blended discount rate
The first factor lives on the commitment ladder from Blog 1: it is bounded by the Minimum Commitment Rate unless you deliberately commit above the minimum level. The second factor is invisible to that ladder entirely. It is set by which commitment products you hold and how you pay for them, and it can differ by more than 2× between two companies standing on the same rung.
Take two customers with identical estates: same Minimum Commitment Threshold, same 45% Minimum Commitment Rate, both covering 35% of eligible ODE. On the ladder they are the same company. On every coverage and utilization dashboard AWS gives you, they are the same company.
- Customer 1 holds 1-year Compute Savings Plans, No Upfront — a 26.5% discount. ESR = 35% × 26.5% = 9.3%, about $731K/year against a $7.884M eligible ODE base.
- Customer 2 holds 3-year EC2 Instance Savings Plans, Full Upfront — a 60.5% discount. ESR = 35% × 60.5% = 21.2%, about $1.669M/year.
Same ladder position, 2.3× the savings — a ~$938K/year difference that no coverage metric can see.
To be precise about what the second customer paid for that gap: three years of lock-in instead of one, a commitment scoped to an instance family instead of all compute, and cash out the door up front. None of that is free, and the right answer depends on your cost of capital rather than on the headline rate — Blogs 7 and 8 do that arbitration properly. The narrow point here is that the discount factor is a real lever of real magnitude, and ESR is the only number on your dashboard that registers it.
Same 35% coverage, two discount mixes: ESR 9.3% vs 21.2%
Identical commitment rate, different discount rate, 2.3× the savings.
The Coverage Factor Is Bounded
Blog 1 built the commitment ladder. The Minimum Commitment Threshold (MCT) is the hourly eligible ODE level covered in every hour; divided by the Ideal Commitment Redline it becomes the Minimum Commitment Rate. Above the threshold you are in the Risk Zone, up to the Commitment Redline — the point where the expected value of the marginal committed dollar turns negative.
That is the bound on the first factor, so it is also the bound on ESR. Covered ODE % — the share of eligible ODE actually receiving commitment discounts — is what you compare against the Minimum Commitment Rate, because both are measured in ODE dollars. Covering 43% of ODE against a 45% rate is strong execution; covering 43% against a 70% rate is a large missed opportunity. Same coverage, opposite readings.
ESR lives on a different axis, so it gets a different comparison:
ESR Ceiling at the minimum threshold = Minimum Commitment Rate x blended discount
This is the risk-free ceiling: the most your estate, as it exists today, can save without taking any breakage risk. Alpenglow's 45% Minimum Commitment Rate at a ~30% blended discount puts it at 13.5%. Current ESR is 10.5%. The gap is three points, about ~$237K/year of savings available at zero risk.
It is not the maximum achievable. The ceiling has a second variant at the redline — (Commitment Redline ÷ Ideal Commitment Redline) x blended discount — the most you can save while every committed dollar still carries positive expected value. For Alpenglow that is 71.8% × 30% = 21.5%. The eight points between the two ceilings are the Risk Zone, expressed in ESR points: reachable, positive-EV, but not free of breakage. In this post, "ESR Ceiling" means the minimum-threshold variant unless stated otherwise.
The rule is one line. Compare Covered ODE % to the Minimum Commitment Rate. Compare ESR to the ESR Ceiling. Never compare ESR to the Minimum Commitment Rate — that mixes a savings rate with a coverage level and produces a number that means nothing.
And note that the ceiling itself moves. Reshape the estate and the Minimum Commitment Rate rises, taking the ceiling with it. Improve the discount mix and the ceiling rises without a single extra covered hour. 13.5% is Alpenglow's risk-free ceiling today, not a property of the universe.
Same ESR, Opposite Problems
Take two companies with the same AWS bill and the same ESR.
Company A covers 35% of its eligible ODE, against a Minimum Commitment Rate of 37% — an ESR of 10.5% at a ~30% blended discount.
That means the estate can structurally support commitment on about 37% of its ODE, and the team is already covering 35 points of it. The gap is only two points. Execution is strong. Buying more commitments of the same kind may still be rational, but only as an explicit move into the Risk Zone with modeled breakage risk.
Company A's next low-risk savings probably come from raising the Minimum Commitment Threshold: pool commitments across business units, fill off-peak valleys, move flexible batch workloads into trough hours, or use continuous micro-commitments to track a moving minimum level more closely.
Company B also covers 35% of its eligible ODE — the same 10.5% ESR — but its Minimum Commitment Rate is 65%.
That is a different business entirely. The estate has enough stable, committable spend to support much more conservative savings, but the company is only covering a little over half of the available minimum level. The issue is not the compute shape. The issue is execution: wrong commitment mix, under-buying, poor timing, excessive caution, broken sharing configuration, or unmanaged breakage risk.
Company B should not start by redesigning workloads. It should fix acquisition.
Company A and Company B as side-by-side commitment ladders: same 35% coverage, execution gaps of 2 vs 30 points
Two points of execution gap versus thirty — the same 35% coverage, read against two different ladders.
Neither the coverage nor the ESR changed between the two companies. The interpretation changed completely once we compared coverage to the Minimum Commitment Rate.
Both companies also hold the same ~30% blended discount: this comparison deliberately freezes the second factor so the first one can be read cleanly. Hold the discount constant and coverage is the whole story. Let it vary — as it does in every real portfolio — and you need both factors before you know what you are looking at.
Two Metrics, Three Levers
Two metrics bound the problem. Three levers move it.
Metric 1: Minimum Commitment Rate. It answers: what percentage of our on-demand equivalent spend is stable enough, after segmentation into exclusive commitment segments, to support commitment in every hour? This is the bound on the coverage factor, and the distance from it to the Ideal Commitment Redline is your Orchestration Headroom.
Metric 2: ESR. It answers: how much of our ODE spend did we actually avoid paying because of commitments? This is the outcome, and it is the product of both factors: Covered ODE % x blended discount.
Lever 1 — Reshape. You raise the Minimum Commitment Rate and the Commitment Redline by changing the shape of consumption: pooling heterogeneous business units, filling off-peak valleys, moving flexible workloads into trough hours, smoothing seasonal variance. This raises the ceiling rather than closing the gap to it. It is not a purchasing problem; it is a systems and governance problem, and it is where Blogs 4, 5 and 9 live.
Lever 2 — Buy. You raise Covered ODE % toward the minimum level by buying the right commitment products, at the right time, in the right order, with the right sharing rules, in small enough increments to track a moving minimum level, while avoiding breakage. This closes the gap to the ceiling. It is a buying and operations problem — Blogs 3 and 6.
Lever 3 — Structure. You raise the blended discount by structuring the commitment terms deliberately: 1-year versus 3-year, No Upfront versus Partial versus Full, Compute Savings Plans versus the narrower EC2 Instance Savings Plans, and any negotiated agreement stacked on top. This lever changes zero coverage. Not one additional hour receives a discount; the discount on the hours you already cover gets larger. That is exactly why no coverage metric and no utilization metric will ever surface it, and why the two-metric framework needs a third lever to be complete. This is Blogs 7 and 8 — and it is the lever that gets dismissed as a finance detour right up until someone computes the $938K/year gap between the two identical-coverage customers of the decomposition example — same rung, 26.5% versus 60.5% discount.
Three levers, three owners, three timescales. Reshape belongs to engineering and governance and plays out over weeks to months. Buy belongs to FinOps and procurement and plays out in days to weeks. Structure is a finance decision, audited against your cost of capital rather than against any coverage dashboard. The next section gives you the ratio that tells you which of the first two to pull.
Minimum Commitment Capture Rate
The "Same ESR, Opposite Problems" comparison worked because it divided coverage by the Minimum Commitment Rate. That division deserves a name, because it is the triage ratio for the whole framework:
Minimum Commitment Capture Rate = Covered ODE % / Minimum Commitment Rate
It measures how much of the Minimum Commitment Threshold you are actually covering — the denominator coverage was always missing. Because both terms share the same ODE base, it reduces to pure ladder arithmetic, no discount rate and no annualisation:
Minimum Commitment Capture Rate = Covered ODE $/hr / MCT $/hr
Capture is Blog 1's commitment ladder collapsed into a single number — where you sit on the line, as a ratio instead of a picture — which is what makes it comparable across estates of different sizes.
The two companies from earlier now separate on sight: Company A captures 0.95, Company B 0.54. Alpenglow sits between them — $315/hr covered against a $405/hr minimum level, capture 0.78 — squarely in the mixed band where both coverage levers are live. Ten points of the minimum level are uncaptured, a Buy gap; and the 45% Minimum Commitment Rate is itself suppressed because each business unit buys in isolation instead of pooling at the payer, a Reshape gap.
Capture says nothing about the third lever, by construction: it is a ratio of two coverage levels, so a portfolio structured badly and one structured well score identically. Alpenglow's is structured badly — $1.2M locked in 3-year Full Upfront plans at an implied rate well under its 12% cost of capital, a ~$66K/year capital-cost mistake that Blog 7 prices out. Note the direction: that position slightly raises ESR, because Full Upfront carries the larger discount. The leak never shows up in ESR; it shows up in what the cash cost to get it. So Alpenglow has real work on all three fronts, which is why it is the running example for the rest of this series.
Covered ODE % vs Minimum Commitment Rate for Companies A, B, and Alpenglow
Same 35% coverage, three different verdicts: capture 0.95, 0.78, and 0.54 against Minimum Commitment Rates of 37%, 45%, and 65%. ESR compares to the ceiling at the minimum threshold (Minimum Commitment Rate × discount), never to the rate itself.
That single comparison tells you which lever to inspect first — and it is exactly the recommendation Cloud ex Machina computes from your hourly billing data, before you buy anything.
| Minimum Commitment Capture Rate | Likely diagnosis | First place to look |
| ≥1.0 | Commitments are above the Minimum Commitment Threshold | Model breakage risk and verify the upside is intentional |
| ~0.85-1.0 | Execution is strong; savings are probably near the Minimum Commitment Threshold | Raise the Minimum Commitment Threshold before pushing risk harder |
| ~0.5-0.85 | Mixed case; both coverage levers may matter | Inspect Buy gaps and structural constraints |
| <0.5 | Large execution gap | Fix acquisition strategy first — Buy before anything else |
A Minimum Commitment Capture Rate above 1.0 is not automatically bad. It means the company is committing beyond the Minimum Commitment Threshold, into the Risk Zone. That can pay dividends if the extra commitment is still positive expected value, or if the organization can actively raise the minimum level through workload shifts, pooling, growth, or micro-commitments. But it is risk-taking, not free efficiency.
These are triage bands, not laws of physics. The point is not to automate judgment away. The point is to stop guessing which class of problem you are looking at.
Read as a two-by-two, capture names the lever directly:
| Low Minimum Commitment Rate | High Minimum Commitment Rate | |
| High coverage | Minimum-level-limited or risk-taking. Check whether coverage is near the Minimum Commitment Threshold or above it, then decide whether to raise the minimum level or accept modeled risk. | Strong position. Keep acquisition disciplined, monitor drift, and look for marginal structural gains. |
| Low coverage | Hard case. The Minimum Commitment Rate is low and execution is not capturing all of it. Close low-risk acquisition gaps, then work on structural changes. | Execution problem. The estate can support more conservative savings; inspect under-buying, product mix, timing, sharing, and breakage controls. |
Quadrant diagnostic: coverage vs minimum commitment rate
The 2×2 in one picture — the diagonal is capture = 1.0; where you sit relative to it names the lever.
Most internal debates about commitments are really arguments about which quadrant the company is in. The two-metric framework makes the disagreement inspectable: establish the Minimum Commitment Rate, measure capture against it, then choose between Reshape and Buy. The quadrant cannot arbitrate Structure — that one is audited separately, against your cost of capital.
Why AWS Tooling Hides This
AWS's default metrics — utilization (were the commitments you bought filled?) and coverage (how much usage got a commitment discount?) — are both execution-side signals. They tell you how your current commitment position behaved; they do not tell you whether the estate could safely support a larger or smaller one, or whether a position above the minimum level is an intentional risk or an accident.
They are also blind to the discount factor entirely. A portfolio of 1-year No Upfront Compute Savings Plans and a portfolio of 3-year plans covering the same hours read identically on both dashboards, at 2.3× apart in realised savings. Utilization and coverage measure hours; only ESR measures money. The full edge-case taxonomy for where these metrics mislead — including cases where both look healthy but buying more commitment is still profitable — is the subject of the next post.
When ESR Moves
The decomposition earns its keep on the month ESR changes. A single number went down; three different things could have caused it, with three different owners.
Attribute the move before reacting to it:
- Coverage change. Commitments expired, or usage grew faster than you bought. Buy — or Reshape, if the minimum level moved under you.
- Discount-mix change. Coverage held, but the portfolio's blended rate shifted: a 3-year tranche expired and renewed at 1-year rates, or a higher-discount plan rolled off. Structure.
- Breakage. Commitments are billing against hours your usage no longer fills. Neither buying nor reshaping helps until the breakage is stopped.
Alpenglow drifting from 10.5% to 9.8% could be two points of lost coverage or a mix shift, and the two demand opposite responses. The decomposition tells you which before anyone signs a purchase order. This is the one genuine weakness of ESR as a headline metric — it moves for reasons it does not explain — and the factors are the fix.
What To Do With Two Metrics and Three Levers
The workflow is straightforward.
First, calculate the Minimum Commitment Threshold from hourly ODE spend. Segment the spend into exclusive commitment segments, use a full seasonal lookback, take the level covered in every hour of each segment, and express the result as a Minimum Commitment Rate.
Second, calculate your Covered ODE % — the share of eligible ODE receiving commitment discounts — your blended discount across the live portfolio, and the ESR the two produce, using the same ODE definition throughout. One denominator, or none of the comparisons mean anything.
Third, divide Covered ODE % by the Minimum Commitment Rate to get the Minimum Commitment Capture Rate, and multiply the Minimum Commitment Rate by the blended discount to get your ESR Ceiling at the minimum threshold.
Then choose the lever:
| Your reading | Lever | Action |
| Capture below 1.0 | Buy | Close the gap to the Minimum Commitment Threshold through better acquisition |
| Capture near 1.0 | Reshape | Raise the Minimum Commitment Threshold by changing the shape of consumption before pushing risk harder |
| Capture above 1.0 | Risk decision | Treat the excess as modeled Risk Zone exposure, bounded by the ceiling at the redline; keep it only if the expected value is positive or you have a credible plan to raise the minimum level |
| Capture healthy, ESR well below the ceiling | Structure | The gap is in the discount factor, not the coverage factor: audit term, payment option and product scope against your cost of capital |
That is the framework: two metrics, three levers, and one Risk Zone above the Minimum Commitment Threshold.
ESR tells you what you saved. The Minimum Commitment Threshold tells you what was structurally safe to save. The Minimum Commitment Capture Rate tells you whether the next dollar should come from buying better, reshaping the estate, or deliberately accepting risk above the minimum level — and the blended discount tells you whether the dollars you already committed were financed on the right terms.
The next problem is that the metrics AWS puts in your dashboard can point you in the wrong direction even when they look healthy. That is where utilization and coverage start to break down.
