Azure Finops

AzureFinOps Skill

Azure cost optimization through reservation analysis, waste discovery, and stakeholder reporting.

Important

• All Azure operations are READ-ONLY — zero modifications to any resource
• Primary output language: Portuguese (BR) for stakeholder responses
• Python CLI fallback: uv run python -m azure_finops when MCP/CLI auth unavailable

Context Files

• AzureToolReference.md — MCP tools + CLI commands quick reference
• ReservationInventory.md — SKU matching and reservation interpretation SOP
• PricingComparison.md — PAYG vs RI pricing methodology and savings formulas

Workflow Routing

Examples

Example 1 — Cost Validation:

"Are these January costs reservations or pay-as-you-go?"

Routes to: Workflows/ValidateCosts.md

Example 2 — Waste Discovery:

"Find orphaned disks and wasted resources across all subscriptions"

Routes to: Workflows/FindWaste.md

Example 3 — Executive Response:

"Draft a response for Jean about the January billing anomalies"

Routes to: Workflows/DraftResponse.md

---

Gotchas

• Reservation matching: SKU naming differs between usage data and purchase data — Standard_D8s_v3 in usage vs D8sv3 in RI. A direct join misses 100%.
• Savings Plan vs Reservation can stack — calculators that pick one ignore the combined savings.
• PAYG vs RI pricing: Marketplace listings are excluded from RI — surprise unreserved charges show up for SQL Server BYOL, etc.
• Cost data lag is 24-48 hours — same-day analyses miss recent activity entirely.
• RI scope (shared vs single-subscription): wrong scope means the RI applies to unintended workloads — and you discover it months later in an audit.
• Currency conversion in EA reports: exchange rate is locked at month close; mid-month estimates use a different rate than the final invoice.

Azure Tool Reference

Quick reference for all Azure tools used by AzureFinOps workflows.

MCP Tools (mcp__azure__*)

MCP Tool Invocation Pattern

mcp__azure__compute → compute_vm_list
  Parameters: subscription, resource_group (optional)
  Returns: List of VMs with SKU, location, status

mcp__azure__pricing → pricing_get
  Parameters: service_name, sku_name, region
  Returns: PAYG price, 1yr RI price, 3yr RI price

CLI Commands (No MCP Equivalent)

Reservation Orders

# List all reservation orders in tenant
az reservations reservation-order list --query "[].{name:name, displayName:displayName, createdDateTime:createdDateTime}" -o table

# List reservations within a specific order
az reservations reservation list --reservation-order-id <ORDER_ID> --query "[].{name:name, sku:sku.name, quantity:quantity, location:location, provisioningState:provisioningState}" -o table

Resource Graph Queries

# Find orphaned (unattached) managed disks
az graph query -q "Resources | where type =~ 'microsoft.compute/disks' | where managedBy == '' or isnull(managedBy) | project name, resourceGroup, subscriptionId, properties.diskSizeGB, sku.name, location" --first 1000

# Find stopped/deallocated VMs
az graph query -q "Resources | where type =~ 'microsoft.compute/virtualMachines' | where properties.extended.instanceView.powerState.code == 'PowerState/deallocated' | project name, resourceGroup, subscriptionId" --first 1000

# Find unassociated public IPs
az graph query -q "Resources | where type =~ 'microsoft.network/publicipaddresses' | where properties.ipConfiguration == '' or isnull(properties.ipConfiguration) | project name, resourceGroup, subscriptionId" --first 1000

Tool Selection Guide

Authentication Notes

• MCP tools use the active Azure CLI session (az login)
• CLI commands require az login with tenant access
• If auth expires mid-workflow, re-authenticate: az login --tenant <TENANT_ID>
• Python CLI fallback reads from exported CSV data when live auth unavailable

Pricing Comparison Methodology

Standard methodology for comparing Pay-As-You-Go (PAYG) vs Reserved Instance (RI) pricing.

Using mcp__azure__pricing

mcp__azure__pricing → pricing_get
  Parameters:
    service_name: "Virtual Machines" | "SQL Database" | "Storage" | etc.
    sku_name: "Standard_D32s_v5" | "P50" | etc.
    region: "eastus" | "brazilsouth" | etc.
    currency: "BRL" | "USD"

Always request pricing in BRL for Brazilian stakeholders. Include USD for reference if needed.

Savings Calculation Formula

Monthly PAYG Cost = hourly_rate * 730 (hours/month)
Monthly 1yr RI Cost = annual_price / 12
Monthly 3yr RI Cost = triennial_price / 36

Monthly Savings (1yr) = Monthly PAYG - Monthly 1yr RI
Monthly Savings (3yr) = Monthly PAYG - Monthly 3yr RI
Savings Percentage = (Monthly Savings / Monthly PAYG) * 100
Annual Savings = Monthly Savings * 12

Typical RI Savings by Service Type

Pricing Report Format

For each service analyzed, present:

| Attribute | Value |
|-----------|-------|
| Service | {service name and SKU} |
| Region | {region} |
| Monthly PAYG | BRL {amount} |
| Monthly 1yr RI | BRL {amount} ({savings_pct}% savings) |
| Monthly 3yr RI | BRL {amount} ({savings_pct}% savings) |
| Recommended Term | {1yr or 3yr} — {reasoning} |
| Annual Savings | BRL {amount} |

When to Recommend Each Term

Common Pricing Pitfalls

1. Comparing wrong SKU: Ensure the pricing SKU matches exactly what billing shows 2. Region mismatch: Prices vary significantly by region (brazilsouth often 20-40% more than eastus) 3. Currency fluctuation: BRL prices change with exchange rates — always use current pricing 4. Instance size flexibility: A reservation for a larger VM can cover multiple smaller ones at ratio 5. Cancelled reservations: Former reservations that were cancelled still show in order history but provide zero savings

Reservation Inventory SOP

Standard operating procedure for interpreting Azure reservation data and matching SKUs.

Order vs Reservation Distinction

• Reservation Order: The purchase container. Has an order ID, created date, billing plan.
• Reservation: The actual capacity commitment within an order. Has SKU, quantity, location, term.
• One order can contain multiple reservations (rare but possible).
• Always drill from order → reservation to get the actual SKU details.

SKU Matching Rules

Exact Match

The reservation SKU exactly matches the billing line item SKU.

Billing: Standard_D32s_v5
Reservation: Standard_D32s_v5
Result: COVERED

Instance Size Flexibility

Azure VM reservations within the same series/family can cover different sizes:

• A D16s_v5 reservation covers 2x D8s_v5 or 0.5x D32s_v5
• Flexibility applies ONLY within the same series and region
• Check the instance size flexibility ratio table for exact mappings

Common Mismatch Patterns

PaaS vs IaaS Distinction

This is the most common source of false "covered" assessments:

• IaaS SQL: VM running SQL Server → needs VM reservation + SQL license (or AHB)
• PaaS SQL: Azure SQL Database → needs SQLDB reservation (different SKU family)
• A SQLDB reservation NEVER covers an IaaS SQL VM license cost

Reservation State Interpretation

Billing Plan Impact

Monthly billing reservations are the hardest to distinguish from PAYG in Cost Management views. Always cross-reference against reservation order list.

Verification Checklist

When asked "is service X covered by a reservation?":

1. List all reservation orders: az reservations reservation-order list 2. For each order, get reservation details: az reservations reservation list 3. Match the billing SKU against reservation SKUs using rules above 4. Check reservation state is Succeeded 5. Check reservation region matches resource region 6. Check reservation quantity vs resource count 7. Report: COVERED / PARTIAL / NOT COVERED with evidence

CoverageAnalysis Workflow

Full reservation gap analysis with savings projections for uncovered Azure services.

Prerequisites

• Read ../AzureToolReference.md for tool invocation patterns
• Read ../ReservationInventory.md for SKU matching rules
• Read ../PricingComparison.md for savings calculation methodology

Steps

1. Identify High-Cost Uncovered Services

Source from either:

• User-provided service list
• Output of ValidateCosts workflow (services marked "NO" coverage)
• Cost Management top-N spenders

2. List Active Reservations

az reservations reservation-order list
az reservations reservation list --reservation-order-id <ORDER_ID>

Build complete reservation inventory with SKU, quantity, region, state, term.

3. Check Each Service for Coverage

For each high-cost service:

• Match against reservation inventory using ReservationInventory.md rules
• Classify: COVERED / PARTIAL / NOT COVERED
• For PARTIAL: explain the gap (wrong SKU, wrong tier, insufficient quantity)

4. Query Azure Advisor Recommendations

mcp__azure__advisor → advisor_recommendation_list
  Filter: category = "Cost"

Advisor may recommend specific RI purchases — capture these as independent validation.

5. Get PAYG vs RI Pricing

For each uncovered service:

mcp__azure__pricing → pricing_get
  Parameters: service_name, sku_name, region, currency="BRL"

Calculate savings using formulas from PricingComparison.md.

6. Locate Actual Resources

Verify resources exist and are active to confirm the spend is real:

mcp__azure__compute → compute_vm_list (for VMs)
mcp__azure__sql → sql_server_list (for SQL)
mcp__azure__storage → storage_account_list (for storage)

7. Present Prioritized Coverage Report

Output format — prioritized by monthly savings (highest first):

## Reservation Coverage Analysis

| # | Service | Monthly (BRL) | Coverage | Savings (1yr) | Savings (3yr) |
|---|---------|--------------|----------|---------------|---------------|
| 1 | {service} | {amount} | NO | {amount} ({pct}%) | {amount} ({pct}%) |

## Recommended Actions (Priority Order)

1. **{Service}** — {term} RI saves ~BRL {amount}/month. {reasoning}.
2. ...

## Summary

| Metric | Value |
|--------|-------|
| Total unprotected monthly spend | BRL {amount} |
| Estimated monthly savings (full RI) | BRL {amount} |
| Estimated annual savings | BRL {amount} |

Output

• Per-service coverage status with pricing evidence
• Prioritized reservation purchase recommendations
• Total savings projection (monthly and annual)
• Advisor recommendations cross-referenced against findings

DraftResponse Workflow

Generate executive stakeholder responses from Azure FinOps analysis results.

Prerequisites

• Analysis must be completed first (ValidateCosts, FindWaste, or CoverageAnalysis)
• Read analysis output from Plans/ directory or current session context

Steps

1. Load Analysis Context

Check for analysis results in order of preference: 1. Current session context (if analysis was just performed) 2. Plans/reservation-coverage-analysis-*.md 3. Plans/response-*-*.md (previous responses for context)

2. Determine Language

3. Determine Audience Tone

4. Structure Response

# Response Template

{Greeting — informal but professional}

{1-2 sentence summary answering the core question}

{Summary table: Service | Monthly Cost | Reservation Status}

{Key findings — 2-4 bullet points}

{Waste findings if applicable — orphaned resources, cleanup opportunity}

{Savings opportunity if applicable — estimated monthly/annual savings}

{Next steps — what you can do or present}

---

## Technical Detail (for internal reference)

{Detailed breakdown for the recipient's team}

5. Save Response

Save to: Plans/response-{recipient-firstname}-{subject-slug}.md

Include metadata header:

# Resposta para {Recipient} — {Subject}

**De:** {sender}
**Para:** {recipient full name}
**Data:** {date}
**Assunto:** {subject in recipient's language}

Key Principles

• Lead with the answer — don't make the reader hunt for the conclusion
• Use BRL for all amounts — this is the stakeholder's currency
• Include the evidence chain — "validated directly in Azure" builds credibility
• Offer next steps — always close with what you can do next
• Keep executive section under 300 words — detail goes in technical appendix

Output

• Formatted response ready to send/share
• Saved to Plans/ directory for future reference
• Technical appendix with full analysis detail

FindWaste Workflow

Discover orphaned and wasted Azure resources across all subscriptions.

Prerequisites

• Read ../AzureToolReference.md for Resource Graph query patterns

Steps

1. Query Orphaned Disks

az graph query -q "
  Resources
  | where type =~ 'microsoft.compute/disks'
  | where managedBy == '' or isnull(managedBy)
  | project name, resourceGroup, subscriptionId,
            diskSizeGB=properties.diskSizeGB,
            sku=sku.name, location
" --first 1000

2. Query Stopped VMs

az graph query -q "
  Resources
  | where type =~ 'microsoft.compute/virtualMachines'
  | where properties.extended.instanceView.powerState.code == 'PowerState/deallocated'
  | project name, resourceGroup, subscriptionId,
            vmSize=properties.hardwareProfile.vmSize, location
" --first 1000

3. Query Unassociated Public IPs

az graph query -q "
  Resources
  | where type =~ 'microsoft.network/publicipaddresses'
  | where properties.ipConfiguration == '' or isnull(properties.ipConfiguration)
  | project name, resourceGroup, subscriptionId, location
" --first 1000

4. Categorize by Type and Size

Group discovered waste by:

• Disk type: StandardSSD_LRS, Premium_LRS, Standard_LRS, Premium_ZRS, etc.
• Disk size: Small (<128GB), Medium (128-1024GB), Large (>1024GB)
• Subscription: Group by subscription for ownership attribution

5. Calculate Monthly Waste

For each orphaned resource, estimate cost using:

mcp__azure__pricing → pricing_get
  Parameters: service_name, sku_name, region

Sum by category and overall.

6. Verify Specific Resources (Optional)

If user asks about specific disks (e.g., "are the P50 disks orphaned?"):

mcp__azure__compute → compute_vm_get
  Check managedBy field on specific disks
  Cross-reference disk names against VM disk attachments

7. Present Waste Report

Output format:

## Orphaned Resources Summary

| Category | Count | Monthly Waste (BRL) | Annual Waste (BRL) |
|----------|-------|--------------------|--------------------|
| Unattached Disks | {n} | {amount} | {amount} |
| Stopped VMs | {n} | {amount} | {amount} |
| Unused Public IPs | {n} | {amount} | {amount} |
| **Total** | | **{total}** | **{total}** |

### Disk Breakdown

| Disk Type | Count | Est. Monthly (BRL) |
|-----------|-------|--------------------|
| StandardSSD_LRS | {n} | {amount} |
| Premium_LRS | {n} | {amount} |

Output

• Categorized waste report with BRL amounts
• Specific resource details if requested
• Recommendation: cleanup candidates vs resources to investigate further

ValidateCosts Workflow

Cross-reference billing costs against active reservations to determine PAYG vs RI coverage.

Prerequisites

• Read ../AzureToolReference.md for tool invocation patterns
• Read ../ReservationInventory.md for SKU matching rules

Steps

1. Parse Target Services

Extract the services/SKUs the user wants validated from their request or from Cost Management data.

2. List All Subscriptions

mcp__azure__subscription_list → subscription_list

Note all subscription IDs for subsequent queries.

3. List Active Reservation Orders

az reservations reservation-order list \
  --query "[].{orderId:name, displayName:displayName, created:createdDateTime}" \
  -o table

4. Get Reservation Details Per Order

For each reservation order:

az reservations reservation list \
  --reservation-order-id <ORDER_ID> \
  --query "[].{name:name, sku:sku.name, quantity:quantity, location:location, state:provisioningState, term:properties.term}" \
  -o table

5. Cross-Reference SKUs

For each target service from step 1:

• Find matching reservation using SKU matching rules from ReservationInventory.md
• Check for exact match, instance size flexibility, and common mismatches
• Check reservation state is Succeeded
• Check region alignment

6. Explain Billing Discrepancies

Common explanations for unexpected costs:

• Partial month: New resource provisioned mid-month — first bill is prorated
• View filter: Cost Management showing "Actual Cost" vs "Amortized Cost" produces different numbers
• Monthly billing RI: Reserved instances with monthly billing look identical to PAYG in cost views
• Renewal batch: Multiple reservation renewals processed on same day create a spending spike

7. Present Coverage Table

Output format:

| # | Service | Monthly (BRL) | RI Coverage | Notes |
|---|---------|--------------|-------------|-------|
| 1 | {service} | {amount} | COVERED / PARTIAL / NO | {explanation} |

Output

• Per-service coverage table with evidence
• Explanation of any billing anomalies
• List of services with no reservation coverage (candidates for CoverageAnalysis workflow)