finops
Optimize cloud spend across AWS, GCP, and Azure using a structured FinOps framework covering waste detection, instance rightsizing, Reserved Instance and Savings Plan analysis, cost allocation tagging
SKILL.mdskills/finops/SKILL.md
GitHubFinOps Skill
When to Activate
Trigger when the user asks to:
- Optimize cloud costs or analyze billing
- Rightsize instances or services
- Analyze reserved instance or Savings Plan coverage
- Build a cost allocation or showback report
- Detect cloud waste or idle resources
- Set up cloud budgets and alerts
- Plan scaling costs or vendor contracts
Waste Detection Checklist
Run this checklist monthly:
AWS
- Unattached EBS volumes (still accruing GB-month charges)
- Unattached Elastic IPs (charged when idle)
- EC2 instances with CPU < 20% sustained (over-provisioned)
- RDS instances with connections < 10% of max connections
- S3 buckets with no lifecycle policies (paying for old logs)
- CloudWatch custom metrics you no longer use
- NAT Gateway idle time (replaced by VPC endpoints?)
- Public S3 buckets (security + cost: data transfer charges)
GCP
- Unused static external IPs
- Idle Cloud SQL instances
- Unused persistent disks
- Oversized machine types
- Load balancers with no backend traffic
Azure
- Orphaned managed disks
- Idle virtual machines not deallocated
- Over-provisioned VM size
- Blob storage with no lifecycle policy
Universal
- Snapshots older than 30 days that could be deleted
- Data transfer charges — is egress higher than expected?
- CDN misconfiguration causing origin hits
- Autoscaling not enabled on variable workloads
Rightsizing Framework
For each EC2/GCE/Azure VM:
- Pull 30-day CPU and memory utilization (CloudWatch / Cloud Monitoring)
- If average CPU < 30%: downgrade to next smaller instance family
- If CPU > 70% regularly: consider upgrade or horizontal scaling
- Apply changes during low-traffic window; monitor for errors
Rightsizing is the highest-ROI FinOps action — a typical 30% reduction in compute spend.
Reserved Instance / Savings Plan Analysis
Decision Tree
Is usage steady-state (>1 year, consistent daily patterns)?
YES → Buy Reserved Instances or 1-year Savings Plans
NO → Use Savings Plans with no commitment, or on-demand
(flexible, convertible if patterns change)
What coverage ratio to target?
Baseline (steady-state prod): 60–80% reserved
Dev/test/staging: 0% reserved (on-demand or spot)
Variable prod: savings plans with no commitment
Coverage Ratio Targets
| Environment | Target Coverage |
|---|---|
| Production (steady) | 60–80% |
| Production (variable) | 30–50% (no commitment) |
| Staging / Dev | 0% |
| Disaster recovery | 0% |
Cost Allocation
Tag everything with these dimensions:
Environment: prod, staging, devTeam: engineering, data, platformProject: service name or project IDCostCenter: department or business unit
Build a monthly showback report by team:
- Total spend
- Top 3 services by spend
- Month-over-month change
- Waste identified and savings actioned
Budget Alert Thresholds
Set budgets at:
- 50%: informational — check if expected
- 80%: warning — investigate immediately
- 90%: action required — freeze non-essential provisioning
- 100%: alert — escalate to engineering lead
Set budgets at service level AND total account level.
Vendor Negotiation
Before any renewal:
- Compile 12 months of actual usage data (not sticker price)
- Research committed use discounts available (typically 30–60% off on-demand)
- Get quotes from at least 2 competitors
- Negotiate based on actuals, not forecasts
- Multi-year terms for >20% additional discount — only if usage is predictable
Key ask: credits over price increases. Most vendors prefer credits to structural discounts.
Weekly Cost Trend Report Template
FINOPS REPORT — Week of [date]
─────────────────────────────────────────
Total cloud spend (MTD): $[X]
vs. prior week: +/-$[Y] (+/-[Z]%)
vs. monthly budget: $[X]/$[budget] ([P]%)
TOP SPENDING SERVICES
1. EC2: $[X] — [any异常?]
2. RDS: $[X] — [any异常?]
3. S3: $[X] — [any异常?]
ANOMALIES DETECTED
- [service]: spike on [date] — [root cause or investigate]
WASTE IDENTIFIED THIS WEEK
- [resource]: $[savings]/mo — action taken / action planned
FORECAST TO END OF MONTH
$[X] projected vs. $[budget] budget — [on track / over / under]
─────────────────────────────────────────
Per-User Scaling Cost Model
For every service that scales with users, build:
Cost model: $[X] per 1,000 MAU
Breakdown:
- Compute (EC2/RDS): $[A] per 1,000 MAU
- Storage (S3/RDS): $[B] per 1,000 MAU
- Egress/CDN: $[C] per 1,000 MAU
- Third-party APIs: $[D] per 1,000 MAU
Implication: 10,000 → 100,000 MAU = $[X * 100] incremental/month
Use this model to set pricing tiers and justify infrastructure investment.