Opscraft
Cloud cost control (FinOps) · Lesson 2 of 6 · 4 min read

Visibility first

Allocation: who spends what

You can't optimise what you can't attribute. Build attribution in layers:

  • Accounts are the strongest boundary: a separate AWS account per team and environment (via Organizations) makes most costs attributable automatically.
  • Cost allocation tags: tag every resource with keys such as team, service, env, cost-centre. User-defined tags must be activated in the Billing console before they show up in cost reports, and they only apply from activation onwards.
  • Enforce tagging: Terraform default tags at the provider level, tag policies in Organizations to standardise keys and values, SCPs or Config rules to block or flag untagged resources.
  • Shared costs (NAT gateways, shared clusters, support plans) need an agreed split, for example proportional to each team's usage.
Enlarge diagramaccountsper team and environment: automatic attributioncost allocation tagsteam, service, env — activated in BillingKubernetes allocationnamespace and labels via Kubecost / split cost datashared costsagreed split rules: NAT, support, platform
Figure 1. Build cost attribution in layers, from strongest to finest.
Swipe sideways to see all of it, or tap to open it full screen.

Tools

Tool Use
Cost Explorer Interactive charts by service, account, tag; forecasts; rightsizing and commitment recommendations
Data Exports (CUR 2.0) Line-item billing data to S3, queried with Athena or loaded into a BI tool; can export in the FOCUS format
FOCUS The FinOps Open Cost and Usage Specification: a common schema for billing data across AWS, Azure, GCP and SaaS
AWS Budgets Alerts when actual or forecast spend crosses a threshold; budget actions can apply restrictive policies
Cost Anomaly Detection Machine-learning alerts for unusual spend by service, account or tag, usually within a day
Compute Optimizer Rightsizing recommendations for EC2, Auto Scaling groups, EBS, Lambda, ECS on Fargate and RDS
Trusted Advisor Checks for idle and underused resources

Set up budgets and anomaly detection on day one, per account and per team, routed to the people who can act, not only to finance.

Showback and chargeback

  • Showback: each team sees its costs in a regular report or dashboard. Builds awareness without accounting overhead.
  • Chargeback: costs are actually billed to each team's budget. Stronger incentive, but needs accurate allocation and agreed rules for shared costs.

Most organisations start with showback and move to chargeback once tagging and shared-cost splits are trusted.

Kubernetes cost allocation

An EKS cluster appears on the bill as EC2 instances, EBS volumes and load balancers, not as teams. To split it:

  • Kubecost or the open-source OpenCost allocate node, storage and network cost to namespaces, labels and workloads, based on resource requests and usage.
  • AWS split cost allocation data for EKS adds pod-level costs to the cost and usage report, allocated by namespace and labels.
  • Allocation is based mostly on requests: a pod requesting 4 vCPU and using 0.2 is charged for 4, because those 4 can't be scheduled for anyone else. This makes over-requesting visible.
  • Idle cluster capacity (nodes not filled by requests) is a cost too; report it separately so the platform team owns it.
Enlarge diagramteam-a / api0.3 of 4.0 vCPU usedteam-b / worker1.6 of 2.0 vCPU usedteam-c / batch0.4 of 3.0 vCPU usedGrey: requested and paid for. Green: actually used.
Figure 2. In Kubernetes you pay for what pods request, not what they use.
Swipe sideways to see all of it, or tap to open it full screen.