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.md

GitHub

FinOps 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:

  1. Pull 30-day CPU and memory utilization (CloudWatch / Cloud Monitoring)
  2. If average CPU < 30%: downgrade to next smaller instance family
  3. If CPU > 70% regularly: consider upgrade or horizontal scaling
  4. 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, dev
  • Team: engineering, data, platform
  • Project: service name or project ID
  • CostCenter: 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:

  1. Compile 12 months of actual usage data (not sticker price)
  2. Research committed use discounts available (typically 30–60% off on-demand)
  3. Get quotes from at least 2 competitors
  4. Negotiate based on actuals, not forecasts
  5. 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.