3.5 Cost Optimization Opportunities
Official task mapping
SAP-C02 task 3.5: Usage and pricing knowledge; granular cost investigation and ownership reporting. This lesson applies the official task to an original design exercise; it is not an exhaustive exam syllabus.
Prerequisites
Review billing exports, commitments and safe automation. Be able to explain the relevant failure boundary and access/cost model before choosing a service.
Establish a measurable baseline
Use Cost Explorer for exploration and detailed exports for reproducible analysis. Legacy CUR and Data Exports CUR 2.0 have different schema/configuration rules. Both CSV and Parquet are relevant supported formats; resource IDs are not available for every line item. Reconcile credits, upfront fees, amortization and estimated/final data before comparing totals.
Find idle/oversized resources and avoidable transfer while preserving SLOs and retention. EBS continues billing when detached; snapshot age does not prove dispensability. Spot requires interruption tolerance. Existing commitments can make gross compute avoidance differ from actual bill reduction.
Use account/tag allocation and owners to turn an observation into an accountable change. Budgets track targets; anomalies detect unexpected behavior. Billing latency limits both.
Analysis-tool selection
| Tool | Appropriate question | Caution |
|---|---|---|
| Trusted Advisor | Which supported optimization/quota checks flag candidates? | Check/support/plan availability; recommendation is not automatic approval |
| Cost Explorer/API | Which recorded cost dimensions changed? | Delayed data, metric and API charges |
| Data Exports/CUR + Athena | Which detailed rows/allocations explain totals? | Schema, scanning, revisions and null resource IDs |
| Budgets | Is a configured objective at risk? | Not a hard cap |
| CAD | Is behavior unexpected? | Learned history and alert thresholds |
| Resource metrics | Can a change meet SLOs? | Operational metrics are not billing line items |
Cost Explorer also has programmatic APIs; “programmatic billing always means CUR” is too absolute. Choose detailed exports for repeatable row-level analysis and Cost Explorer for supported aggregate queries.
For hypothetical us-east-1 study inputs on 2026-10-12, use a before/after bill ledger rather than subtracting a sunk commitment twice:
| Selected workload scope | Before/month | After scheduling/month |
|---|---|---|
| Existing fixed commitment payment/amortization | $180 | $180 |
| Remaining On-Demand usage charges | $120 | $0 |
| Incremental monitoring | $0 | $20 |
| Total | $300 | $200 |
Assume all selected running usage is removed, the same commitment term continues, and no other eligible usage can consume its benefit. The removed workload had $300 allocated running cost, but the actual bill falls only $120 before monitoring, or $100 net. The $180 is already present in both totals; it is not an extra new penalty to subtract again from the $100. If another workload absorbs the benefit, model its reduced uncovered charges separately. A future non-renewal is a separate decision/horizon, not an immediate saving from stopping today’s instances. These are synthetic ledger amounts, not discounted-rate or coverage predictions.
A quota warning is capacity-risk evidence, not a cost saving. An unassociated IPv4 address or idle load balancer may be billable, but check dependencies and current price rules before deletion. Link every candidate to its owner, recoverability and measurement period.
Scenario and rationale
A $50,000 monthly platform bill includes 20% overnight development compute and a nearly idle reservation portfolio. Scheduling can reduce running hours, but unused commitments may absorb some or all of the expected invoice saving. First model hourly coverage and alternative eligible usage, then right-size/schedule and revise future renewal plans.
Reject reporting 20% savings solely from stopped instance hours. Reject selling Convertible RIs: Marketplace eligibility applies to qualifying Standard EC2 RIs. Reject deleting all old snapshots without retention and owner review.
Practice drill
Using supplied hourly usage and purchase data, calculate coverage, utilization, gross avoided cost and remaining commitment liability separately. Build a prioritized list with owner, change risk, expected net saving and verification window. Query a provided Parquet billing dataset only after inspecting its actual schema; no universal copied SQL schema is assumed.
References
- Official Domain 3 task guide
- Data Exports and CUR 2.0
- RI Marketplace restrictions
- Cost optimization principles
Return to Domain 3. The drill is proposed study work, not a claim of a completed deployment.