The Azure bill nobody could explain

Reading 30 days of billing data.

audit.loglive

Numbers from the public sample report of a fictional company.replay

The scenario#

A small SaaS team on Azure watched its bill climb month after month. Nobody had changed much, so nobody could say why.

The subscription had grown the way most do: a few people with Contributor access, test environments created for a sprint and never removed, and a production setup sized for a launch that never got the traffic it was planned for.

Symptoms#

• Spend up steadily for a year, with no matching growth in usage.
• The cost view grouped by service showed “Virtual Machines” and “Storage” as the biggest lines, which explains nothing.
• No tags, so no way to tell production from leftovers.

How it was found#

The fix starts with cost per resource, not per service. Joining 30 days of actual cost to the resource inventory shows exactly what each thing costs. Then a handful of Resource Graph queries find the usual suspects.

// disks left behind after their VM was deleted
Resources
| where type =~ 'microsoft.compute/disks'
| where properties.diskState =~ 'Unattached'
| project name, resourceGroup, sku = tostring(sku.name), sizeGb = toint(properties.diskSizeGB)

The same approach finds VMs that are stopped but still allocated, public IPs attached to nothing, App Service plans with no apps, and non-production VMs that run 24 hours a day. CPU metrics over 14 days show which running machines are far bigger than they need to be.

What fixed it#

1. Week 1: deallocate stopped VMs, delete unattached disks, idle IPs and empty plans.
2. Week 2: confirm old snapshots and parked VMs with their owners, then remove them.
3. Weeks 2 to 3: auto-shutdown for dev and test, resize oversized VMs, move non-prod disks off Premium.
4. Week 4: buy reservations for what’s left, sized on the smaller footprint.

# stop paying for a VM that's already stopped
az vm deallocate --ids "$VM_ID"
# shut dev VMs down at 7 pm every day
az vm auto-shutdown --ids "$VM_ID" --time 1900

Result#

In the public sample report this scenario is based on, the plan cuts $3,066 a month from $9,094, about 34% of spend. Commitments are counted at 70% so the headline isn’t inflated.

Lessons#

• Look at cost per resource. Cost per service hides the answer.
• Clean up before you commit. Reservations bought on a wasteful footprint lock the waste in.
• Tag by environment and owner from day one, and the next audit takes an hour.