Skip to content
Back to Blog
2 min read

Cost Optimization with Azure Advisor

Azure Azure Advisor Cost Optimization FinOps Cloud Management

“Where is the money going?” is the most expensive question in cloud. Azure Advisor is the free service that answers a surprising amount of it for you. Right-sizing recommendations, idle resources, reserved instance suggestions, and the small embarrassments like “you have a Premium SSD attached to a stopped VM.” Today I’m focusing on the cost recommendations specifically, plus a Logic App pattern I use to auto-remediate the safe ones (deallocate dev VMs after hours, delete orphan disks) without a human in the loop.

Understanding Azure Advisor

Azure Advisor analyzes your resource configurations and usage telemetry to provide actionable recommendations. For cost optimization, it identifies:

  • Idle and underutilized resources
  • Right-sizing opportunities
  • Reserved Instance recommendations
  • Unassociated resources (orphaned disks, IPs)

Accessing Advisor Recommendations via API

Here is how to programmatically retrieve cost recommendations:

from azure.identity import DefaultAzureCredential
from azure.mgmt.advisor import AdvisorManagementClient
import pandas as pd

credential = DefaultAzureCredential()
subscription_id = "your-subscription-id"

client = AdvisorManagementClient(credential, subscription_id)

def get_cost_recommendations():
    """
    Retrieve all cost-related recommendations from Azure Advisor.
    """
    recommendations = []

    # Filter for cost category
    for rec in client.recommendations.list(filter="Category eq 'Cost'"):
        recommendations.append({
            "id": rec.id,
            "name": rec.name,
            "resource_id": rec.resource_metadata.resource_id if rec.resource_metadata else None,
            "problem": rec.short_description.problem,
            "solution": rec.short_description.solution,
            "impact": rec.impact,
            "impacted_value": rec.extended_properties.get("annualSavingsAmount", "N/A"),
            "currency": rec.extended_properties.get("savingsCurrency", "USD"),
            "recommendation_type": rec.recommendation_type_id,
            "last_updated": rec.last_updated
        })

    return pd.DataFrame(recommendations)

# Get and display recommendations
df = get_cost_recommendations()
print(f"Total cost recommendations: {len(df)}")
print(f"\nEstimated annual savings: ${df['impacted_value'].sum():,.2f}")
print(df.to_string())

Common Cost Optimization Scenarios

1. Virtual Machine Right-Sizing

using Azure.ResourceManager;
using Azure.ResourceManager.Advisor;
using Azure.ResourceManager.Compute;

public class VmRightSizingService
{
    private readonly ArmClient _armClient;

    public VmRightSizingService(ArmClient armClient)
    {
        _armClient = armClient;
    }

    public async Task<List<VmRecommendation>> GetVmSizingRecommendationsAsync(
        string subscriptionId)
    {
        var recommendations = new List<VmRecommendation>();
        var subscription = _armClient.GetSubscriptionResource(
            new ResourceIdentifier($"/subscriptions/{subscriptionId}")
        );

        await foreach (var advisor in subscription.GetAdvisorRecommendations()
            .GetAllAsync(filter: "Category eq 'Cost'"))
        {
            if (advisor.Data.RecommendationTypeId ==
                "e10b1381-5f0a-47ff-8c7b-37bd13d7c974") // VM right-sizing
            {
                var extProps = advisor.Data.ExtendedProperties;

                recommendations.Add(new VmRecommendation
                {
                    VmName = extProps["vmName"],
                    CurrentSize = extProps["currentSku"],
                    RecommendedSize = extProps["targetSku"],
                    CurrentCost = decimal.Parse(extProps["currentCost"]),
                    ProjectedCost = decimal.Parse(extProps["targetCost"]),
                    AnnualSavings = decimal.Parse(extProps["annualSavingsAmount"]),
                    CpuP95 = double.Parse(extProps["cpuP95"]),
                    MemoryP95 = double.Parse(extProps["memoryP95"])
                });
            }
        }

        return recommendations;
    }

    public async Task ApplyRecommendationAsync(VmRecommendation recommendation)
    {
        // Note: This requires VM restart - implement with caution
        var vmResource = _armClient.GetVirtualMachineResource(
            new ResourceIdentifier(recommendation.ResourceId)
        );

        var vm = await vmResource.GetAsync();
        var vmData = vm.Value.Data;

        // Update the VM size
        vmData.HardwareProfile.VmSize = recommendation.RecommendedSize;

        // Deallocate first (required for size changes)
        await vmResource.DeallocateAsync(WaitUntil.Completed);

        // Update the VM
        await vmResource.UpdateAsync(WaitUntil.Completed,
            new VirtualMachinePatch { HardwareProfile = vmData.HardwareProfile });

        // Start the VM
        await vmResource.PowerOnAsync(WaitUntil.Completed);
    }
}

public class VmRecommendation
{
    public string ResourceId { get; set; }
    public string VmName { get; set; }
    public string CurrentSize { get; set; }
    public string RecommendedSize { get; set; }
    public decimal CurrentCost { get; set; }
    public decimal ProjectedCost { get; set; }
    public decimal AnnualSavings { get; set; }
    public double CpuP95 { get; set; }
    public double MemoryP95 { get; set; }
}

2. Identifying Orphaned Resources

from azure.mgmt.compute import ComputeManagementClient
from azure.mgmt.network import NetworkManagementClient

def find_orphaned_disks(credential, subscription_id):
    """
    Find managed disks not attached to any VM.
    """
    compute_client = ComputeManagementClient(credential, subscription_id)

    orphaned_disks = []
    total_cost = 0

    for disk in compute_client.disks.list():
        if disk.managed_by is None and disk.disk_state == "Unattached":
            # Calculate monthly cost based on disk size and type
            monthly_cost = calculate_disk_cost(disk.disk_size_gb, disk.sku.name)
            total_cost += monthly_cost

            orphaned_disks.append({
                "name": disk.name,
                "resource_group": disk.id.split("/")[4],
                "size_gb": disk.disk_size_gb,
                "sku": disk.sku.name,
                "created": disk.time_created,
                "monthly_cost_estimate": monthly_cost
            })

    print(f"Found {len(orphaned_disks)} orphaned disks")
    print(f"Estimated monthly cost: ${total_cost:,.2f}")

    return orphaned_disks

def find_orphaned_public_ips(credential, subscription_id):
    """
    Find public IP addresses not associated with any resource.
    """
    network_client = NetworkManagementClient(credential, subscription_id)

    orphaned_ips = []

    for ip in network_client.public_ip_addresses.list_all():
        if ip.ip_configuration is None:
            orphaned_ips.append({
                "name": ip.name,
                "resource_group": ip.id.split("/")[4],
                "ip_address": ip.ip_address,
                "sku": ip.sku.name,
                "allocation_method": ip.public_ip_allocation_method
            })

    return orphaned_ips

def calculate_disk_cost(size_gb, sku_name):
    """
    Estimate monthly cost for a managed disk.
    Prices are approximate and vary by region.
    """
    price_per_gb = {
        "Standard_LRS": 0.05,
        "StandardSSD_LRS": 0.075,
        "Premium_LRS": 0.15,
        "UltraSSD_LRS": 0.25
    }
    return size_gb * price_per_gb.get(sku_name, 0.05)

3. Reserved Instance Analysis

import requests
from datetime import datetime, timedelta

def analyze_ri_recommendations(credential, subscription_id):
    """
    Analyze Reserved Instance purchase recommendations.
    """
    token = credential.get_token("https://management.azure.com/.default").token

    # Get RI recommendations from Advisor
    url = f"https://management.azure.com/subscriptions/{subscription_id}/providers/Microsoft.Advisor/recommendations"
    params = {
        "api-version": "2020-01-01",
        "$filter": "Category eq 'Cost' and RecommendationTypeId eq 'reserve-vm-instances'"
    }

    headers = {"Authorization": f"Bearer {token}"}
    response = requests.get(url, params=params, headers=headers)
    recommendations = response.json().get("value", [])

    ri_analysis = []

    for rec in recommendations:
        props = rec.get("properties", {}).get("extendedProperties", {})

        ri_analysis.append({
            "vm_family": props.get("vmFamily"),
            "region": props.get("region"),
            "term": props.get("term"),
            "recommended_quantity": int(props.get("qty", 0)),
            "current_hourly_cost": float(props.get("currentHourlyCost", 0)),
            "ri_hourly_cost": float(props.get("riHourlyCost", 0)),
            "annual_savings": float(props.get("annualSavingsAmount", 0)),
            "savings_percentage": float(props.get("savingsPercent", 0))
        })

    return ri_analysis

Automating Remediation with Azure Functions

Create an automated workflow to act on recommendations:

using Microsoft.Azure.WebJobs;
using Microsoft.Extensions.Logging;
using Azure.ResourceManager;
using Azure.Identity;

public class AdvisorRemediationFunction
{
    [FunctionName("ProcessAdvisorRecommendations")]
    public async Task Run(
        [TimerTrigger("0 0 6 * * 1")] TimerInfo timer, // Weekly on Monday at 6 AM
        ILogger log)
    {
        log.LogInformation("Starting Advisor recommendation processing");

        var credential = new DefaultAzureCredential();
        var armClient = new ArmClient(credential);

        var subscriptionId = Environment.GetEnvironmentVariable("SUBSCRIPTION_ID");
        var subscription = armClient.GetSubscriptionResource(
            new ResourceIdentifier($"/subscriptions/{subscriptionId}")
        );

        // Process different recommendation types
        await ProcessOrphanedDisks(subscription, log);
        await ProcessIdleResources(subscription, log);
        await GenerateCostReport(subscription, log);
    }

    private async Task ProcessOrphanedDisks(
        SubscriptionResource subscription,
        ILogger log)
    {
        var disksToDelete = new List<string>();
        var threshold = TimeSpan.FromDays(30); // Delete if unattached for 30+ days

        await foreach (var rec in subscription.GetAdvisorRecommendations()
            .GetAllAsync(filter: "Category eq 'Cost'"))
        {
            if (rec.Data.RecommendationTypeId == "orphaned-disk-recommendation")
            {
                var lastUpdated = rec.Data.LastUpdated ?? DateTime.UtcNow;

                if (DateTime.UtcNow - lastUpdated > threshold)
                {
                    // Auto-delete old orphaned disks
                    var resourceId = rec.Data.ResourceMetadata.ResourceId;

                    // First, create a snapshot for safety
                    await CreateDiskSnapshot(resourceId, log);

                    // Then delete
                    disksToDelete.Add(resourceId);
                    log.LogInformation($"Queued disk for deletion: {resourceId}");
                }
                else
                {
                    // Send notification for recent orphans
                    await SendNotification(
                        $"Orphaned disk detected: {rec.Data.ResourceMetadata.ResourceId}",
                        log
                    );
                }
            }
        }

        foreach (var diskId in disksToDelete)
        {
            // Delete with confirmation
            await DeleteOrphanedDisk(diskId, log);
        }
    }

    private async Task GenerateCostReport(
        SubscriptionResource subscription,
        ILogger log)
    {
        var report = new StringBuilder();
        report.AppendLine("# Weekly Cost Optimization Report\n");

        decimal totalPotentialSavings = 0;

        await foreach (var rec in subscription.GetAdvisorRecommendations()
            .GetAllAsync(filter: "Category eq 'Cost'"))
        {
            var savings = decimal.Parse(
                rec.Data.ExtendedProperties.GetValueOrDefault("annualSavingsAmount", "0")
            );
            totalPotentialSavings += savings;

            report.AppendLine($"## {rec.Data.ShortDescription.Problem}");
            report.AppendLine($"**Solution:** {rec.Data.ShortDescription.Solution}");
            report.AppendLine($"**Annual Savings:** ${savings:N2}");
            report.AppendLine($"**Impact:** {rec.Data.Impact}");
            report.AppendLine();
        }

        report.AppendLine($"\n## Summary");
        report.AppendLine($"**Total Potential Annual Savings:** ${totalPotentialSavings:N2}");

        // Send report via email or Teams
        await SendCostReport(report.ToString(), log);
    }
}

Suppressing Recommendations

Sometimes you need to dismiss recommendations that do not apply:

# Suppress a recommendation via CLI
az advisor recommendation disable \
    --ids "/subscriptions/{sub}/resourceGroups/{rg}/providers/Microsoft.Compute/virtualMachines/{vm}/providers/Microsoft.Advisor/recommendations/{rec-id}"

# Suppress for a specific duration
az advisor recommendation postpone \
    --ids "/subscriptions/{sub}/resourceGroups/{rg}/providers/..." \
    --duration 30  # days

Creating Custom Alerts

Set up alerts when new high-impact recommendations appear:

{
  "$schema": "https://schema.management.azure.com/schemas/2019-04-01/deploymentTemplate.json#",
  "contentVersion": "1.0.0.0",
  "resources": [
    {
      "type": "Microsoft.Insights/activityLogAlerts",
      "apiVersion": "2020-10-01",
      "name": "HighImpactCostRecommendation",
      "location": "global",
      "properties": {
        "enabled": true,
        "scopes": [
          "[subscription().id]"
        ],
        "condition": {
          "allOf": [
            {
              "field": "category",
              "equals": "Recommendation"
            },
            {
              "field": "operationName",
              "equals": "Microsoft.Advisor/recommendations/available/action"
            },
            {
              "field": "properties.recommendationCategory",
              "equals": "Cost"
            },
            {
              "field": "properties.recommendationImpact",
              "equals": "High"
            }
          ]
        },
        "actions": {
          "actionGroups": [
            {
              "actionGroupId": "[resourceId('Microsoft.Insights/actionGroups', 'FinOpsTeam')]"
            }
          ]
        }
      }
    }
  ]
}

Best Practices

  1. Review Regularly: Schedule weekly reviews of Advisor recommendations
  2. Automate Low-Risk Actions: Auto-delete orphaned resources after a grace period
  3. Validate Before Acting: Always verify recommendations make sense for your workload
  4. Track Savings: Measure actual savings after implementing recommendations
  5. Integrate with FinOps: Use Advisor data in your cost management dashboards

Azure Advisor is a valuable tool in your FinOps toolkit. Combined with automation, it helps maintain cost efficiency as your cloud environment grows and evolves.\n\n## Takeaways\n\nAdd a concise, personal takeaway and recommended next steps here.\n

Michael John Pena

Michael John Pena

Senior Data Engineer based in Sydney. Writing about data, cloud, and technology.