Cost Management

Budgets, reservations, and common cost mistakes

Cost Management

Budget Alerts

Set these on both subscriptions immediately. Budget alerts are free and take 5 minutes to configure.

Thresholds

Setting Your Initial Budget

If you don’t know your expected spend, set the budget at 2x your current monthly burn. You’ll get a 50% alert when you hit your actual expected spend, which is a good sanity check. Adjust after 2-3 months of data.

Tagging Strategy

Required Tags

Enforced via Azure Policy (deny mode):

Optional Tags (Recommended)

Tag Inheritance

Azure Policy can auto-inherit tags from resource groups to child resources. Deploy the built-in policy Inherit a tag from the resource group for environment and team. This way you only need to tag the resource group, and resources inside automatically get tagged.

Common Startup Cost Mistakes

1. Forgotten Dev Resources

The problem: Someone spins up a Standard_D4s_v5 VM to test something. Three weeks later it’s still running. Multiply by 5 engineers.

The fix:

• Set auto-shutdown on all non-prod VMs (7 PM local time, no auto-start)
• Review Azure Advisor cost recommendations weekly (it flags idle resources)
• Deploy a policy that requires the auto-shutdown tag on VMs in non-prod

resource autoShutdown 'Microsoft.DevTestLab/schedules@2018-09-15' = {
  name: 'shutdown-computevm-${vm.name}'
  location: location
  properties: {
    status: 'Enabled'
    taskType: 'ComputeVmShutdownTask'
    dailyRecurrence: { time: '1900' }
    timeZoneId: 'Eastern Standard Time'
    targetResourceId: vm.id
  }
}

2. Over-Provisioned Databases

The problem: “Let’s start with S3 to be safe” turns into $600/month for a database handling 10 requests/minute.

The fix:

• Start with the smallest tier. Always.
• Azure SQL: S0 (~$15/month) for dev. S1 or elastic pool for prod.
• Cosmos DB: Serverless for dev (pay per request). Autoscale provisioned for prod.
• Azure Database for PostgreSQL: Burstable B1ms (~$13/month) for dev. General Purpose for prod.
• Scale up is instant. Scale down requires no downtime for most services.

3. Not Using Reserved Instances

The problem: Paying on-demand prices for workloads that have been running for 6 months.

The fix:

• After a workload runs for 3+ months and isn’t going away, buy a 1-year reservation
• Typical savings: 30-40% on compute (VMs, AKS nodes, App Service plans)
• Start with 1-year commitments only — don’t lock into 3 years at your stage
• Azure Advisor shows RI recommendations based on your actual usage

4. Premium Storage Where Standard Works

The problem: All VMs provisioned with Premium SSD because “it’s the default.”

The fix:

• Non-prod: Standard SSD (E-series disks). 1/3 the cost of Premium.
• Prod with low IOPS: Standard SSD is often sufficient.
• Prod with high IOPS: Premium SSD or Ultra Disk.
• Managed disks are billed whether the VM is running or not — right-size them.

5. Ignoring Spot VMs

The problem: Paying full price for fault-tolerant workloads.

The fix:

• CI/CD agents: Run on Spot VMs (up to 90% discount). If evicted, the job retries.
• Batch processing: Spot VMs or Spot node pools in AKS.
• Dev/test environments: Spot VMs with Deallocate eviction policy.

Setting maxPrice: -1 (or spot_max_price = -1 in Terraform) means “pay up to the current on-demand price.” Azure will only evict the VM when it needs the capacity back for on-demand customers — never because of price fluctuations. This is the recommended default for most Spot workloads because it maximizes uptime while still getting the Spot discount (typically 60-90% off).

resource vmss 'Microsoft.Compute/virtualMachineScaleSets@2024-03-01' = {
  properties: {
    virtualMachineProfile: {
      priority: 'Spot'
      evictionPolicy: 'Deallocate'
      billingProfile: {
        maxPrice: -1  // pay up to on-demand price, evict only when Azure needs capacity
      }
    }
  }
}

If you want to cap the price (e.g., only run when the Spot price is below $0.05/hr), set maxPrice to that value instead. The VM will be evicted if the Spot price exceeds your cap or if Azure needs the capacity.

6. Not Using Azure Dev/Test Pricing

The problem: Paying full price for non-prod Windows VMs and licensed services.

The fix:

• Create your non-prod subscription as a Dev/Test subscription (or convert it)
• Benefits: No Windows OS license charges on VMs, discounted rates on several services
• Requires Visual Studio subscribers (most startups with MSDN/MPN have these)

Automated Cost Reporting

Budget alerts catch spikes, but you also need regular visibility into where money goes. Azure Cost Management provides built-in scheduled exports — no extra tooling required.

Set Up Cost Exports

In the Azure Portal: Cost Management → Exports → Add

Or via CLI:

# Create a weekly cost export for the subscription
az costmanagement export create \
  --name "weekly-cost-export" \
  --scope "subscriptions/<SUBSCRIPTION_ID>" \
  --type "ActualCost" \
  --timeframe "MonthToDate" \
  --storage-account-id "/subscriptions/<SUB>/resourceGroups/<RG>/providers/Microsoft.Storage/storageAccounts/<SA>" \
  --storage-container "cost-exports" \
  --schedule-recurrence "Weekly" \
  --schedule-status "Active" \
  --recurrence-period-from "$(date -u +%Y-%m-%dT00:00:00Z)" \
  --recurrence-period-to "2030-01-01T00:00:00Z"

What to Do With the Data

• Small team (< 10 engineers): Weekly portal review is enough. Set up exports as a backup audit trail.
• Growing team (10-50): Import exports into Power BI with the Cost Management connector for team-level dashboards.
• Enterprise path: Feed exports into a FinOps tool (e.g., Azure FinOps toolkit, or third-party like Infracost).

Cost Anomaly Alerts

Azure Cost Management also supports anomaly detection (preview). Enable it to get automatic alerts when daily spend deviates significantly from the baseline — catches issues that fixed-threshold budget alerts miss.

Cost Management → Cost alerts → Anomaly alerts → Enable for each subscription.

Cost Monitoring Routine

Weekly (5 minutes)

1. Open Cost Management in Azure Portal
2. Check cost trend — is it flat, growing, or spiking?
3. Review Azure Advisor cost recommendations — usually 2-3 quick wins
4. Check for resources with no tags (use Resource Graph: resources | where tags == {})

Monthly (30 minutes)

1. Review cost by resource group and tag
2. Compare actual vs budget
3. Check RI utilization (if you have reservations)
4. Review and resize over-provisioned resources
5. Delete old snapshots, unattached disks, unused public IPs

Useful Azure Resource Graph Queries

// Resources with no environment tag
resources
| where isnull(tags.environment) or tags.environment == ""
| project name, type, resourceGroup, subscriptionId

// Unattached managed disks (wasting money)
resources
| where type == "microsoft.compute/disks"
| where properties.diskState == "Unattached"
| project name, resourceGroup, sku.name, properties.diskSizeGB

// Public IPs not attached to anything
resources
| where type == "microsoft.network/publicipaddresses"
| where isnull(properties.ipConfiguration)
| project name, resourceGroup

Azure Savings Plans vs Reserved Instances

Recommendation: Start with Savings Plans for flexibility. Switch to RIs for workloads that are very stable (same VM size for 6+ months).

See Also

• Architecture Decisions — Why two subscriptions, budget design
• Troubleshooting — Budget start date format errors
• Graduation Guide — When to add cost management tooling