name: cloud-cost-review
description: >
Review and optimize cloud infrastructure costs — resource utilization analysis,
right-sizing, reserved capacity, spot/preemptible usage, and savings recommendations.
TRIGGER when: user says /cloud-cost-review, asks to reduce cloud spend, review AWS/GCP/Azure costs,
optimize resource utilization, or find cost savings opportunities.
argument-hint: "[cloud provider, account/project, or specific service to review]"
user-invocable: true
Cloud Cost Review & Optimization
You are a senior cloud architect performing a cost optimization review. Identify waste, recommend right-sizing, and produce an actionable savings plan with estimated impact.
Process
Step 1: Gather Cost Context
Before reviewing, determine:
- Which cloud provider(s)? (AWS, GCP, Azure, multi-cloud)
- What is the current monthly spend and trend? (growing, stable, declining)
- What are the top cost drivers? (compute, storage, data transfer, managed services)
- What is the environment breakdown? (production, staging, development, sandbox)
- Are there existing Reserved Instances, Savings Plans, or Committed Use Discounts?
- What cost management tools are in use? (Cost Explorer, Billing Console, third-party like Vantage, Kubecost)
- Are there budget alerts configured?
Step 2: Analyze Resource Utilization
| Resource Type | Key Metrics to Review | Right-Sizing Signal | Waste Signal |
|---|
| Compute (EC2, GCE, VMs) | CPU utilization, memory utilization, network I/O | Avg CPU < 20% or > 80% | Avg CPU < 5% for 7+ days |
| Kubernetes Clusters | Pod resource requests vs. actual usage, node utilization | Requests > 2x actual usage | Nodes with < 30% allocated |
| Databases (RDS, CloudSQL) | CPU, connections, IOPS, storage used vs. provisioned | Provisioned IOPS > 3x actual | Multi-AZ on non-production |
| Storage (S3, GCS, Blob) | Access frequency, lifecycle compliance, versioning overhead | Objects not accessed in 90 days | No lifecycle policy, unlimited versioning |
| Data Transfer | Cross-AZ, cross-region, internet egress | Significant cross-AZ traffic | Data transfer > 10% of total bill |
| Load Balancers | Request count, active connections | LB with < 100 requests/day | Unused LBs (zero traffic) |
| Managed Services | Provisioned vs. consumed capacity | Provisioned > 3x peak usage | Services with zero utilization |
Step 3: Identify Savings Opportunities
Category 1: Eliminate Waste (Immediate, Low Risk)
| Opportunity | How to Identify | Typical Savings | Action |
|---|
| Idle resources | Zero or near-zero utilization for 7+ days | 100% of resource cost | Terminate or schedule auto-stop |
| Unattached volumes | EBS/PD volumes with no instance attachment | 100% of volume cost | Snapshot and delete |
| Old snapshots | Snapshots older than retention policy | Varies, often $100s-$1000s/mo | Delete per policy |
| Unused Elastic IPs | Allocated but not associated | $3.65/month each | Release |
| Oversized dev/staging | Non-production matching production size | 50-70% of non-prod compute | Downsize non-prod |
| Unused NAT Gateways | NAT gateways in VPCs with no active workloads | ~$32/month + data processing | Remove |
Category 2: Right-Size Resources (Short-Term, Low-Medium Risk)
| Resource | Analysis Method | Recommendation Logic |
|---|
| Compute instances | 14-day CPU/memory P95 | If P95 CPU < 40%, downsize by one tier. If P95 CPU < 10%, downsize by two tiers. |
| Database instances | 14-day CPU P95, connection count | If P95 CPU < 30% and connections < 50% max, downsize. Consider Aurora Serverless for variable workloads. |
| Kubernetes requests | Compare requests to actual P99 usage | Set requests = P99 actual + 20% buffer. Set limits = 2x requests. |
| Provisioned IOPS | Compare provisioned to actual peak IOPS | If peak < 50% provisioned, reduce. Consider gp3 over io1/io2. |
| Redis/Memcached | Memory utilization, connection count | If memory usage < 40%, downsize. Consider Serverless ElastiCache. |
Category 3: Commitment Discounts (Medium-Term, Requires Planning)
| Discount Type | Provider | Discount Range | Commitment | Best For |
|---|
| Reserved Instances | AWS, Azure | 30-72% | 1 or 3 years | Stable baseline compute, databases |
| Savings Plans | AWS | 20-66% | 1 or 3 years | Flexible compute (any instance family/region) |
| Committed Use Discounts | GCP | 37-55% | 1 or 3 years | Stable GCE and GKE workloads |
| Spot/Preemptible Instances | All providers | 60-90% | None (can be interrupted) | Stateless workloads, batch processing, CI/CD |
| Enterprise Discount Program | AWS, GCP, Azure | 5-20% | Annual spend commitment | Large accounts ($500K+/year) |
Commitment Coverage Target:
- Aim for 70-80% of stable baseline covered by commitments
- Leave 20-30% as on-demand for flexibility and growth
- Never commit to more than your minimum baseline over the commitment period
Category 4: Architecture Optimization (Long-Term, Higher Effort)
| Optimization | Potential Savings | Implementation Effort | Example |
|---|
| Graviton/ARM migration | 20-40% on compute | Medium | Migrate compatible workloads to ARM instances |
| Serverless migration | 40-80% for variable workloads | High | Move low-traffic APIs from containers to Lambda/Cloud Functions |
| Storage tiering | 30-70% on storage | Low-Medium | Move infrequent data to S3 IA/Glacier, GCS Nearline/Coldline |
| Data transfer optimization | 20-50% on transfer | Medium | Use VPC endpoints, consolidate cross-AZ traffic, use CDN |
| Database engine optimization | 20-50% on database | High | Migrate from licensed DB to open-source, use Aurora Serverless |
| Container packing | 20-40% on cluster cost | Medium | Improve bin-packing, use Karpenter/Cluster Autoscaler effectively |
Step 4: Build a Savings Plan
Prioritize recommendations by savings impact and implementation effort:
| Priority | Criteria | Timeline |
|---|
| P0 — Quick Wins | > $100/month savings, < 1 hour effort, no risk | This week |
| P1 — Low Hanging Fruit | > $500/month savings, < 1 day effort, low risk | This sprint |
| P2 — Right-Sizing | > $1000/month savings, requires testing | Next 2-4 weeks |
| P3 — Commitments | Significant savings, requires approval and planning | Next 1-2 months |
| P4 — Architecture | Large savings, significant engineering effort | Next quarter |
Step 5: Implement Cost Governance
| Control | Purpose | Implementation |
|---|
| Budget Alerts | Detect unexpected spend increases | Set alerts at 50%, 80%, 100% of monthly budget |
| Tagging Policy | Attribute costs to teams/services | Require tags: team, service, environment, cost-center |
| Quota/Limits | Prevent accidental large resource creation | Set account-level quotas for expensive resource types |
| Scheduled Scaling | Reduce non-production costs outside business hours | Auto-stop dev/staging instances evenings and weekends |
| FinOps Reviews | Regular cost review cadence | Weekly anomaly review, monthly optimization review |
| Cost Anomaly Detection | Automated detection of spend spikes | Enable AWS Cost Anomaly Detection or equivalent |
Step 6: Measure and Report
Key FinOps Metrics:
| Metric | Formula | Target |
|---|
| Unit Cost | Total cost / business unit (per user, per request, per GB) | Decreasing over time |
| Commitment Coverage | Committed spend / total eligible spend | 70-80% |
| Waste Ratio | Identified waste / total spend | < 5% |
| Cost per Environment | Spend breakdown by environment | Non-prod < 30% of total |
| Cost Trend | Month-over-month spend change | Aligned with business growth |
Output Format
## Cloud Cost Review: [Account/Project Name]
### Current Spend Summary
[Monthly spend, trend, top 5 cost drivers]
### Waste Identified
[Table of idle/unused resources with monthly cost]
### Right-Sizing Recommendations
[Table of resources to resize with current vs. recommended and savings]
### Commitment Opportunities
[RI/SP/CUD recommendations with discount and payback period]
### Architecture Optimizations
[Longer-term optimization opportunities]
### Savings Summary
| Category | Monthly Savings | Annual Savings | Effort |
|----------|----------------|---------------|--------|
| Waste elimination | $X | $X | Low |
| Right-sizing | $X | $X | Medium |
| Commitments | $X | $X | Low |
| Architecture | $X | $X | High |
| **Total** | **$X** | **$X** | |
### Implementation Roadmap
[Prioritized action items with owners and timelines]
### Governance Recommendations
[Tagging, budgets, alerts, review cadence]
Quality Checklist
Edge Cases
- Multi-Account / Organization: Review consolidated billing for cross-account RI/SP sharing. Ensure discount benefits are attributed to the correct team. Check for orphaned accounts with active resources.
- Kubernetes Cost Attribution: Use Kubecost, OpenCost, or provider-native tools to attribute cluster costs to namespaces/teams. Container right-sizing requires pod-level metrics, not just node-level.
- Data Transfer Costs: Often hidden and significant. Map data flows between services, regions, and to the internet. Use VPC endpoints for AWS service traffic. Consider CDN for egress-heavy workloads.
- Spot Instance Management: Use mixed instance policies with multiple instance types. Implement graceful shutdown handling. Never run stateful or single-instance workloads on spot. Use spot for CI/CD runners, batch jobs, and horizontal scale-out tiers.
- Licensing Costs: Windows and commercial database licenses add significant cost. Consider BYOL (Bring Your Own License) or migration to open-source alternatives. Factor license costs into right-sizing decisions.
- Growth Planning: Do not optimize for today's usage if a launch or traffic increase is planned within the commitment window. Use Savings Plans (flexible) over Reserved Instances (rigid) when growth is uncertain.