Start with visibility
You can't optimize what you can't see. The first step is to tie every euro to an application and an owner: minimal tagging linked to the CMDB, daily billing exports from Cost Management, and a dashboard that leadership and application owners can both read. That is often where the first anomalies show up: non-prod costing more than prod, resources with no owner, a finished project still running.
Rate optimization: pay less for the same consumption
These levers don't touch the applications. They are the fastest to put in place.
| Lever | Typical saving | Good to know |
|---|---|---|
| Reserved Instances (1 or 3 years) | ≈ −50% on a prod VM | Commitment on a SKU and a region: ideal for the stable baseline |
| Azure savings plan for compute | A bit less than an RI | Hourly commitment in €, flexible across SKUs and regions |
| Azure Hybrid Benefit | The cost of the Windows Server or SQL Server license | Reuses licenses you already own with Software Assurance |
| Reserved capacity (storage, databases, Fabric) | Depends on the service | Once volumes are stable |
| MACC | Negotiated discount | Overall consumption commitment with Microsoft |
The golden rule: clean up before you commit. You don't commit for three years to an oversized or useless VM. We detail the method on our Reserved Instances and savings plans page.
Usage optimization: consume only what is useful
Rightsizing
A VM whose CPU and RAM stay below 50% at P95 over 30 days can move down one SKU, which roughly halves its cost. The latest SKU generations (v5 and later) are also often cheaper for the same performance.
Automatic Stop and Start on demand
Non-prod VMs and VDIs don't need to run at night or on weekends. A fixed-schedule Stop/Start saves 50 to 60%, almost as much as a reservation, and the two don't stack. The real lever is an automatic Stop every evening and a Start on demand: the VM only restarts when someone needs it, and savings reach about 90%. On Azure Virtual Desktop, the "Start VM on Connect" feature works exactly like this for virtual desktops.
Cleaning
Unattached disks, stale snapshots, unused public IPs, empty App Service Plans, forgotten POC environments: removing what's no longer used is 100% savings on the resource. In non-prod, Premium SSD disks can often move to Standard SSD.
The services that drift with volume
Some services cost little on day one, then a fortune at scale: Log Analytics and Azure Monitor, Microsoft Sentinel, Azure Backup, Synapse, Fabric. They are often the biggest savings pool on a mature Azure bill. We detail further down the levers we activate on each one, together with the experts in charge. On logs, costs can sometimes be divided by 4.
Why Azure Advisor isn't enough
Azure Advisor is a good starting point, but its rules are generic and know nothing about your business context, the criticality of your applications or your load peaks. We apply rules defined with your IT department, on your real metrics, and we then measure the real savings after implementation. See our recommendation engine.
Our approach on Azure
- See: read-only (Reader) access to billing, dashboard, cost mapping by application.
- Reduce: quick wins from the first month, then a remediation plan across all services.
- Stay in control: FinOps governance, monitoring and alerting, budget forecasting.