Stranded Commitment Waste Rate (SCWR)

GMI-IT-015IT OperationsPilot-Ready

Definition

Stranded Commitment Waste Rate (SCWR) is the proposed metric for this: the percentage of committed cloud spend that is paying for capacity no longer used by any active workload.

SCWR makes the stranded commitment problem specific and trackable.

The Measurement Gap

Your RI utilization report says 87%. That sounds healthy. For most organizations, 87% is the number FinOps teams point to in quarterly business reviews as evidence that the commitment purchasing program is working. The dashboard is green. The AI recommender's confidence scores were high. Finance signed off on the purchase plan.

What the 87% number doesn't tell you is that the 13% you're not using isn't randomly distributed. It's concentrated in the tail: the reservations purchased 18 months ago for a product that launched, ran at low utilization, and then got deprecated. The reservations for the database tier that was consolidated. The compute reserved instances for the service that migrated to serverless. Each one burns money every hour of every day, with no workload consuming it and no mechanism alerting you that it's happening.

Those unused commitments don't fail the RI utilization metric because that metric aggregates across the portfolio. New purchasing on new workloads (where utilization is high) offsets the stranded tail. The 87% average hides the 0% on the legacy commitments. And the AI purchasing tool doesn't surface this problem because it's tasked with recommending future purchases, not auditing past ones.

I've seen portfolios where the stranded commitment tail, once isolated and totaled, came to $400,000 in annual spend paying for capacity nobody uses. The commitments were individually small: $8,000 here, $22,000 there. Nothing flagged a single one as a crisis. The aggregate was significant enough to matter to the business. It went untracked because the metric framework rewarded keeping the headline utilization rate above 85%, not eliminating dead spend.

The legacy metrics are blind to this because they were designed for a different problem. RI utilization measures whether purchased commitments are being consumed: it doesn't ask whether the workloads they were purchased for still exist. Savings plan coverage measures how much eligible spend is discounted: it doesn't ask whether the discounted workloads are still running in the expected pattern. Cost-per-unit metrics eventually reflect stranded commitments, but with a lag measured in quarters, and diluted by the rest of the spend base.

The measurement gap for SCWR is this: there is no metric that tells you the cumulative dollar value of committed cloud spend paying for capacity no longer used by any active workload. Without that metric, stranded commitments accumulate invisibly in the portfolio tail while the headline RI utilization number reassures everyone that the purchasing program is working. The financial reality of stranded commitments only becomes visible when someone runs an ad-hoc audit, and those don't happen often enough because there's no metric driving the work.

Formula

Stranded Commitment Waste Rate (SCWR) is the proposed metric for this: the percentage of committed cloud spend that is paying for capacity no longer used by any active workload. SCWR makes the stranded commitment problem specific and trackable. It's not about average utilization across the portfolio: it's about isolating the portion of committed spend that is burning money on capacity with no associated workload consuming it.

Formula: plain language:

For each active commitment, compare committed capacity to actual utilization. Where utilization falls below the committed level, the gap represents stranded capacity. Multiply the gap by the effective hourly rate for that commitment. Sum the dollar value of all stranded capacity across all commitments. Divide by total committed spend. The result is the percentage of your committed budget that is producing no value.

Formula: notation:

SCWR = Σ (C_committed,i − C_utilized,i) × P_i / Σ C_committed,i × P_i

Where:

  • C_committed,i = committed capacity for commitment i (hours, normalized compute units, or service units per period)
  • C_utilized,i = actual utilization of commitment i, capped at C_committed,i
  • P_i = effective hourly rate for commitment i
  • The numerator sums only over commitments where C_utilized,i < C_committed,iWorked Example An organization has 10 active reserved instances. Nine are running at full utilization: the committed capacity equals the utilized capacity, contributing $0 to the numerator. The tenth is a 3-year standard RI purchased 18 months ago for a service that was deprecated 6 months in. It commits to 730 instance-hours per month at an effective rate of $1.20/hour. Current utilization: 0 hours. Stranded contribution: (730 − 0) × $1.20 = $876/month = $10,512/year. If total committed spend across all 10 RIs is $120,000/year, SCWR = $10,512 / $120,000 = 8.76%. That's one stranded RI. Scale this to a portfolio of hundreds of commitments and the aggregate becomes material.

Sign convention:

Lower is better. 0% means all committed capacity is being consumed by active workloads. Higher percentages indicate more stranded spend. Track the direction of change, a SCWR that increases quarter-over-quarter indicates the portfolio is accumulating stranded commitments faster than they're being resolved.

Edge Cases

AWS allows RI sharing across accounts: attribute utilization to the sharing pool, not the purchasing account. Partially stranded commitments (commitment is 60% utilized) count the 40% gap in the numerator, not the full commitment. Commitments in their first 90 days should be excluded from SCWR: they haven't had time to stabilize, and inclusion will artificially inflate the metric during periods of heavy purchasing. Commitments sold on the marketplace should be removed from the active portfolio before calculating.

Cross-references:

Upstream: Commitment Recommendation Accuracy (CRA, GMI-IT-013). When the AI purchasing recommender makes inaccurate predictions, commitments don't match actual usage patterns. Low CRA is a leading indicator of future SCWR growth. If CRA is declining, expect SCWR to follow with a lag of 6–12 months. Upstream: Commitment Portfolio Flexibility Index (CPFI, GMI-IT-014), a rigid portfolio can't redirect stranded commitments. Low CPFI means stranded commitments, once created, stay stranded until the term expires. High CPFI means there's at least the option to exchange or convert. Lateral: Capacity Prediction Accuracy (CPA, GMI-IT-009). Demand forecast errors on the CPA side become commitment purchasing errors, which become stranded commitments on the SCWR side. CPA is the upstream cause; SCWR measures the accumulated downstream cost. When both are deteriorating simultaneously, the root problem is in the demand forecasting inputs the AI purchases from.

Proposed formula, pending validation pilots.

At a Glance

Category
AI-Era IT Operations
← STALE DEPENDENCY INFLATION RATE (SDIR)TELEMETRY SAMPLING BLIND SPOT RATE (TSBSR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment