We Measured 24–67% Cloud Waste Across 5 Real Workloads
Published: May 2026 · 8 min read
Before placemy was a product, the engine inside it was a research project. We scanned five production workloads on AWS and Azure, applied the highest-confidence right-sizing recommendations, and measured the cost difference. The savings ranged from 23.9% to 67.4%. Here is the complete data — no rounding, no cherry-picking.
The five test cases
Five real workloads, measured savings
Range 23.9%–67.4%. Median 63.6%.
| Case | Workload | Current | Before / mo | Recommendation | After / mo | Saving |
|---|---|---|---|---|---|---|
| TC-01 | Over-provisioned t3.micro (idle, 0.07% avg CPU) AWS t3.micro, eu-west-2 | AWS t3.micro eu-west-2 | $8.61 | AWS t4g.nano EU (Stockholm) | $3.14 | −63.6% |
| TC-02 | Under-provisioned B1s (22% avg, 94% peak CPU) Azure Standard_B1s, northeurope | Azure Standard_B1s northeurope | $8.25 | Azure B2pts v2 swedencentral | $6.28 | −23.9% |
| TC-03 | Idle B1s (0.4% avg CPU, 25.7% peak) Azure Standard_B1s, italynorth | Azure Standard_B1s italynorth | $8.76 | AWS t4g.nano EU (Stockholm) | $3.14 | −64.2% |
| TC-04 | Residency-constrained t3.micro (Ireland) AWS t3.micro, eu-west-1 | AWS t3.micro eu-west-1 | $8.32 | AWS t4g.micro EU (Stockholm) | $6.28 | −24.6% |
| TC-05 | Stable web B2s (13.7% avg, spiky pattern) Azure Standard_B2s, switzerlandnorth | Azure Standard_B2s switzerlandnorth | $38.54 | AWS t4g.small EU (Stockholm) | $12.56 | −67.4% |
Real workloads on real AWS and Azure accounts in EU regions, scanned with the same pipeline shipping in placemy today. Published in the 2025 MSc dissertation (L00162129). On-demand pricing; your results will vary by workload pattern.
Methodology
Each workload was a real VM scanned during normal production operation. The methodology was:
- Discover the running instance and its current hourly rate from the provider's pricing catalogue.
- Pull 24 hours of CPU metrics from CloudWatch (AWS) or Azure Monitor. Calculate P50, P95, P99, and peak utilisation.
- Compare the workload against all current-generation instance types in the same region, filtered by OS and architecture compatibility.
- Additionally compare against equivalent types on the other provider (cross-cloud) to identify migration savings.
- Rank candidates by monthly cost saving while ensuring the recommended instance still provides adequate headroom above the P95 demand.
What the numbers represent
The “Before” column is the monthly cost of the original instance type at the current on-demand rate. The “After” column is the monthly cost of the recommended alternative. The saving percentage is the actual reduction if the recommendation were applied.
We did not apply every recommendation — only those the engine flagged with high confidence and low blast radius. The 67.4% saving (case 5) came from a cross-cloud migration: an Azure Standard_B2s VM running at under 8% P95 CPU, where the equivalent workload on AWS t4g.micro was available at a fraction of the cost.
What the numbers don't cover
Implementation effort. These savings are the ceiling of what the engine surfaces. The floor depends on your team's appetite for change, the blast radius of each move, and whether you can tolerate the migration window. We think that is the honest way to present the data.
Key findings
- The median saving across all five test cases was 32.5%.
- Every single workload was oversized by at least one instance family step — none were correctly provisioned.
- Cross-cloud recommendations (moving between AWS and Azure) produced larger savings than same-cloud downsizing in 3 of 5 cases.
- P95-based sizing avoided false positives: workloads with occasional spikes to 80%+ CPU were not recommended for downsizing because their P95 reflected the sustained demand.
- The engine compared each workload against an average of 1,247 candidate instance types before selecting the top recommendation.
Try it on your own workloads
The same engine that produced these numbers is the one running inside the placemy CLI today. One command, 5 minutes, a report with your own dollar figures.
$ curl -sSL https://placemy.cloud/install | sh $ placemy scan --output report.html