Three first checks
- Find the baseline: in Cost Management, choose the correct billing scope/date range and identify the largest services, subscriptions and unexplained changes. Record actual spend, not an assumed saving.
- Find the owner and usage: compare Advisor recommendations with workload metrics, peak demand and service-level requirements. An idle-looking disk or VM may be a recovery dependency.
- Review a reversible change: agree on a small rightsizing or scheduling trial with its owner and rollback. Delete only after retention/recovery checks; buy commitments only after stable usage is understood.
No saving percentage or Azure execution result is demonstrated by this article. Price, eligibility and contract reviews are not recorded; consult Cost Management and your actual agreement.
Prioritise the work
| Family | Typical effort | Reversibility | Prerequisite |
|---|---|---|---|
| Usage review and Advisor triage | Low to moderate | Read-only | Correct scope and representative metrics |
| Scheduling and rightsizing | Moderate | Often reversible, but downtime/capacity may matter | Owner approval, performance baseline, rollback |
| Data/resource cleanup | Moderate | Potentially irreversible | Dependency inventory, retention and restore evidence |
| Tags, policy and budget ownership | Moderate | Policy enforcement needs controlled rollout | Named owners and valid service constraints |
| Reservations/savings plans/licence benefits | Commercial review | Contract-dependent, not equivalent to stopping a VM | Stable eligible usage, terms and licence evidence |
| Region, platform and agreement changes | High | Migration/contract-dependent | Architecture, residency, reliability and finance approval |
Operational changes
VM sizing: compare CPU, memory, disk and network demand across representative periods, not just averages. VM selector and Advisor help narrow choices; a resize can require restart or encounter capacity constraints.
Start/stop schedules: agree on working hours and dependencies. Verify Stopped (deallocated) for compute-allocation savings; stopping the guest OS alone is not equivalent. Disks, reserved IPs and other resources can still be billed. Check current VM state and billing guidance .
Disks and blob tiers: choose measured IOPS/throughput/redundancy needs and data access patterns. Lower storage prices can add retrieval, transaction, rehydration or early-deletion costs. Check current blob tiers ; do not archive data that must be immediately available.
Orphan/data cleanup: inventory unattached resources and obsolete data, obtain owner/retention approval, then follow the preview-first TTL review. "Unattached" is not proof of "safe to delete".
Governance and ongoing ownership
Allowed VM sizes: use policy only after selecting supported SKUs for the actual region/workload. Test Audit before Deny and maintain an exception path; copying old Basic_A examples is not a current capacity recommendation.
Tags and policies: agree on Owner, Environment and cost-allocation tags. Verify tag support and review missing ownership before enforcing. The Policies catalogue is a starting point, not a complete cost control.
Cost Management and training: assign budget alerts and anomaly reviews to people who will act. Budgets are alerts, not hard caps. Train workload owners using Microsoft Learn and review spend continuously after changes.
Financial commitments
Reservations: exchange an eligible usage commitment for contractual pricing. Compare utilisation, scope, term, region/family flexibility and current exchange/refund rules before buying. A reservation does not stop or resize a resource. See reservation guidance .
Savings plans: evaluate eligible hourly compute spend, term and scope against a stable baseline. Unused commitment and workloads outside eligibility affect the outcome. Review savings plan documentation .
Azure Hybrid Benefit: confirm actual eligible licences, Software Assurance/subscription rights and permitted reuse before enabling it. A policy setting is not proof of entitlement. Keep evidence with finance/licensing owners; see Hybrid Benefit .
Architecture and commercial decisions
Regions: compare total costs including latency, egress, replication, availability and data-residency requirements. Use Azure Regions for orientation and verify current service pricing.
Agreements: enterprise/commercial discounts depend on your agreement and procurement terms; no universal discount or multi-year saving is promised here.
Workload choices: evaluate Spot eviction tolerance, burstable CPU behaviour, Dev/Test eligibility, autoscaling, containers and alternative compute services. Choose based on workload requirements , not a universal cheapest architecture.
Continuous review: use a Well-Architected Review and FinOps ownership. Optional inventory/automation tools such as Azure Optimization Engine still need scoped permissions and review.
Copy or print a cost review
Billing scope and baseline period:
Service/resource and accountable owner:
Observed usage and reliability requirements:
Proposed change and expected cost mechanism:
Commitment / licence / retention conditions:
Approval and rollback or restore plan:
Actual before/after measurement:
Next review: