Your cloud bill is probably out of control, and here's how to fix it. The engineering teams that understood cloud cost management before 2020 had a significant operational advantage as budgets came under scrutiny.

Tagging as the foundation

Without consistent resource tagging, cloud costs cannot be allocated to teams, products, or environments. The mandatory tags that every resource should carry: team, project, environment, cost-centre, and created-date. Enforcement: Azure Policy with DeployIfNotExists effect (automatically apply tags at resource creation), deny policies for resources that lack required tags, and Azure Cost Management tag filters for per-team reporting.

Azure Cost Management and budgets

Azure Cost Management provides cost analysis by resource group, tag, service, and subscription. Budget alerts notify when actual or forecasted spend exceeds configured thresholds. The operational discipline: weekly cost reviews by team, monthly budget variance reports, and engineering-owned cost dashboards that show per-service spend trends. Cost anomaly detection identifies unexpected spend spikes before they compound.

The right-sizing workflow

Azure Advisor's cost recommendations identify over-provisioned VMs, idle resources (unattached disks, unused public IPs, idle load balancers), and Reserved Instance opportunities. The right-sizing workflow: Advisor recommendations reviewed monthly, automated reports to team leads, sizing changes scheduled for low-traffic windows, and a before/after comparison of cost and performance metrics. The typical outcome of a right-sizing engagement is 20-30% cost reduction for mature workloads.

Reserved Instances and Savings Plans

Reserved Instances for known-stable workloads (production AKS node pools, production databases) provide 40-70% discounts over pay-as-you-go pricing. Azure Savings Plans provide flexible compute commitments that apply to any VM usage regardless of SKU or region. The discipline: right-size before committing to reservations (committing to an oversized instance locks in the overprovision), review utilisation reports monthly, and convert on-demand spend to reservations as workloads stabilise.