Finops Framework Inform Optimize Operate

Cloud FinOps Framework: Inform, Optimize & Operate

When to Use This Skill

Triggers — load this skill when:

• Cloud spend must be allocated, attributed, or forecast credibly
• A right-sizing, waste-cleanup, or commitment-purchase decision is needed
• Unit economics (cost per request/tenant/feature) must be established

Route elsewhere when:

• Capacity and autoscaling design -> scalability-high-availability-patterns
• Cluster-level node efficiency -> aws-eks-enterprise-patterns / gcp-gke-autopilot-multi-tenant
• Efficiency vs delivery-throughput metrics -> devops-metrics-dora-kpis

1. The FinOps 3-Phase Lifecycle

       +--------------+
       |   INFORM     |  -> Visibility, Cost Allocation, Unit Economics
       +------+-------+
              |
       +------v-------+
       |  OPTIMIZE    |  -> Right-sizing, Waste Reduction, Commitment Models
       +------+-------+
              |
       +------v-------+
       |  OPERATE     |  -> Continuous Governance, Automated Budget Alerts
       +--------------+

2. Mandatory Enterprise Tagging Strategy (IaC / Terraform)

Enforce default provider tags across all cloud resources to ensure $100%$ cost attribution:

provider "aws" {
  region = "us-east-1"
  default_tags {
    tags = {
      Environment     = "production"
      CostCenter      = "CC-4092"
      Owner           = "data-platform-team"
      Service         = "stream-processing"
      FinOpsManaged   = "true"
      BusinessUnit    = "E-Commerce"
    }
  }
}

3. High-Impact Cloud Cost Optimization Checklist

• Orphaned Storage Cleanup: Delete unattached EBS volumes, unreferenced Azure managed disks, and stale RDS/EBS snapshots.
• Commitment Coverage Target: Maintain $75%-85%$ coverage with Compute Savings Plans / Reserved Instances (RIs) for steady-state baselines.
• Kubernetes Cost Allocation: Deploy Kubecost / OpenCost to slice cluster costs down to namespaces, labels, and individual Pods.
• Storage Lifecycle Policies: Transition S3 / GCS buckets to Infrequent Access (IA) after 30 days and Glacier / Archive after 90 days.

4. Showback, Chargeback & Cost Spike Triage

Start with showback. Chargeback on top of unreliable allocation produces disputes about the data instead of decisions about the spend.

Shared cost is where showback loses credibility. Decide and publish the rule before the first report: split shared clusters, observability, and support by a measured driver (Kubecost / OpenCost namespace cost, request count, or storage GB), never evenly. Keep unallocatable spend visible as its own line — a hidden 30% "platform" bucket destroys trust in the whole model.

Cost spike triage, in order:

1. Scope it: Cost Explorer / BigQuery billing export grouped by service, then by usage type, at daily granularity — one dimension at a time until the delta is a single line item.
2. Date it: align the step change against deploys, autoscaling events, and commitment expiries; a step means a change shipped, a ramp means growth or a leak.
3. Name the driver: NAT gateway or cross-AZ data transfer, log ingestion volume, orphaned volumes and snapshots, idle load balancers, a retry storm, or expired Savings Plan coverage.
4. Decide: fix the driver, or accept it and re-forecast — with a named owner and a date.

Commitment purchases come last, after the waste is removed: buying a three-year commitment on an inefficient baseline locks in the inefficiency. Cover measured steady-state usage only (typically 60–80%), and let elastic peaks run on demand or spot.

5. Commitment Instruments Compared

Sequence that avoids locking in waste:

1. Remove waste first (idle, orphaned, over-provisioned) — commitments on inefficient baselines fund the inefficiency for three years.
2. Measure the trough, not the average: commit to the floor of steady-state usage, typically 60-80% coverage, and leave peaks on demand or spot.
3. Prefer the flexible instrument (Compute Savings Plans, spend-based CUDs, convertible RIs) whenever an architecture change is plausible within the term.
4. Track coverage and utilisation monthly as two separate metrics; high coverage with low utilisation means the commitment is being wasted.