Azure overspend is rarely one dramatic mistake. It is a slow accumulation of reasonable decisions that nobody revisited. A VM sized generously at design time, a disk tier chosen before anyone measured, a proof of concept that quietly became permanent. None of it looks wrong in isolation, and together it can double a bill.
Right-size against measured usage, not design assumptions
Machines are almost always sized before anyone knows what the workload really does. That is unavoidable at design time and inexcusable a month later, once you have real telemetry. The single highest-return cost action is reviewing VM sizes against actual CPU and memory usage after the first month of production.
One review, most of the savings
Set a calendar reminder for four weeks after go-live to review sizing. That one review typically finds more savings than any negotiation.
Reservations and savings plans
- Anything running continuously is a candidate for a reservation, which trades commitment for a materially lower rate
- One-year terms are the sensible starting point for an SMB, since three years is a long commitment for a business that may change shape
- Right-size first, then reserve. Reserving an oversized machine locks in the waste
- Savings plans offer more flexibility than reservations if your workload mix is likely to change
Turn things off
| Resource | Common waste | Fix |
|---|---|---|
| Dev and test VMs | Running 24/7 for a 40-hour work week | Auto-shutdown schedule |
| Orphaned disks | Left behind when a VM is deleted | Monthly sweep for unattached disks |
| Old snapshots | Accumulating indefinitely | Retention policy with automatic expiry |
| Public IP addresses | Reserved and unused | Audit and release |
| Premium storage | Chosen by default, not by need | Tier against measured IOPS |
| Log retention | Verbose logging kept forever | Set retention to what you would actually query |
Storage tiering is the quiet one
Storage rarely appears on anyone's cost radar because each item is small. In aggregate it is often a significant share of the bill. Cold data sitting on premium disks is pure waste, and moving it to a cool or archive tier is usually a configuration change rather than a project.
Budget alerts wired to a person
Azure will happily let you spend. Budget alerts are free, take minutes to configure, and are the difference between noticing a runaway resource in days rather than at invoice time. Route them to a named individual, not a shared mailbox nobody reads.
A monthly rhythm that actually works
- 1Review the top five cost lines and ask whether each is still needed
- 2Check for unattached disks, unused public IPs and stale snapshots
- 3Compare spend against budget and investigate any variance over ten percent
- 4Review right-sizing recommendations rather than dismissing the notification
- 5Confirm auto-shutdown schedules are still applied to non-production resources
Azure bill drifting upwards?
We review Azure spend against actual usage, find what is running that should not be, and set up the alerts that stop it happening again.
