agency-finops-engineer

Start with allocation: enforce required tags so spend maps to teams and services. Then eliminate idle and orphaned resources, schedule non-production environments, rightsize over-provisioned compute, and only then purchase commitments on the stable remainder.

Buy commitments only after waste is eliminated and the workload baseline is proven stable for 30-90 days. Target roughly 70-85% coverage of the steady baseline, and confirm no migration or refactor is planned during the 1-3 year term.

Split shared costs like networking and observability using a documented key, preferably usage-based or headcount-based. Enforce mandatory tags at provisioning through SCPs, Azure Policy, or GCP org policy, and audit daily until allocated spend exceeds 95%.

Rightsizing must preserve headroom required by availability and performance SLOs. The Skill treats SLOs as constraints, validates risky changes in staging, and rejects any saving that removes burst capacity needed to absorb traffic spikes.

Cross-AZ and cross-region traffic, NAT gateway data processing, internet egress, and snapshot or storage-class sprawl are commonly missed. Trace the data path rather than just compute, then apply VPC endpoints, CDN caching, and lifecycle policies.

Track unit economics such as cost per customer, per request, or per transaction rather than total spend. A rising bill with flat or falling unit cost indicates efficient scaling, while rising unit cost signals waste.