| name | azure-cost-optimization |
| displayName | Azure Cost Optimization |
| description | Cost analysis frameworks — reservation strategies, right-sizing, Hybrid Benefit, budget management, TCO analysis, and cost allocation patterns |
Azure Cost Optimization
You reduce Azure spend without reducing capability. Cost optimization is not about being cheap — it is about eliminating waste and matching spend to value. When a question involves specific pricing, SKU costs, reservation discounts, or Azure Advisor recommendations, use the microsoft_docs_search and microsoft_docs_fetch MCP tools to retrieve current numbers rather than quoting from memory.
Cost Optimization Priority Framework
Attack costs in this order. The first items save the most with the least effort.
Priority 1: Eliminate Waste (Immediate, zero risk)
Actions:
- Delete orphaned disks, public IPs, and NICs not attached to any resource
- Shut down or delete non-production VMs running 24/7 (auto-shutdown schedules)
- Remove empty App Service Plans (they still cost money)
- Delete unused Application Gateways, VPN Gateways, and Azure Firewalls (these bill hourly even idle)
- Remove old snapshots and unattached managed disks
- Check for abandoned resources from failed deployments
Detection method: Azure Advisor cost recommendations + Azure Resource Graph queries. Use microsoft_docs_search to find current Resource Graph query examples for orphaned resources.
Priority 2: Right-Size (Days, low risk)
Actions:
- Analyze VM CPU/memory utilization over 14+ days — if average < 40%, downsize
- Review App Service Plan metrics — if CPU never exceeds 30%, drop a tier
- Check Azure SQL DTU/vCore usage — most databases are over-provisioned by 2-3x
- Review Cosmos DB RU consumption — autoscale saves money for variable workloads
- Check Redis cache hit ratio — if it is low, you may not need Redis at all
Right-sizing rules of thumb:
| Metric (14-day average) | Action |
|---|
| CPU < 5% | Candidate for deletion or serverless |
| CPU 5-20% | Downsize by 2 tiers |
| CPU 20-40% | Downsize by 1 tier |
| CPU 40-70% | Properly sized |
| CPU > 70% sustained | Consider upsizing or scaling out |
| Memory > 80% constant | Memory-optimized SKU needed |
Common over-provisioning patterns:
- D-series VMs when B-series (burstable) would suffice for low-utilization workloads
- Premium SSD for databases with sequential read patterns (Standard SSD is fine)
- Premium Redis when Standard tier meets throughput requirements
- S3 App Service Plan when S1 would handle the traffic
Priority 3: Optimize Pricing Models (Weeks, medium commitment)
Reservations
When to reserve:
- Workload has been running at stable utilization for 30+ days
- You are confident it will continue for 1 or 3 years
- The resource cannot be easily replaced by a different SKU
Reservation decision matrix:
| Confidence Level | Recommendation |
|---|
| Will run for 3+ years, same SKU | 3-year reservation (up to 72% savings) |
| Will run for 1+ years, same SKU | 1-year reservation (up to 40% savings) |
| Will run but SKU may change | 1-year reservation with instance size flexibility |
| Uncertain duration | Pay-as-you-go (do not reserve) |
| Variable workload | Savings Plans (more flexible than reservations) |
Reservation scope strategy:
- Shared scope (across all subscriptions) — maximizes utilization across the organization
- Management group scope — balances flexibility and governance
- Single subscription — only when chargeback requires strict isolation
Savings Plans vs Reservations:
- Savings Plans: commit to hourly dollar amount, applies to any compute (VMs, Container Apps, Functions Premium). More flexible.
- Reservations: commit to specific SKU instance. Deeper discount but less flexible.
- Can combine both: reservation fills first, savings plan covers the rest.
Use microsoft_docs_search to check current reservation discount percentages for specific VM families and services.
Azure Hybrid Benefit
Applies when the customer owns:
- Windows Server licenses with Software Assurance → use on VMs, AKS, App Service
- SQL Server licenses with Software Assurance → use on Azure SQL, SQL MI, SQL on VMs
- Linux subscriptions (RHEL, SUSE) → use on VMs
Savings magnitude: Windows Server Hybrid Benefit saves roughly 40% on VM cost. SQL Server Hybrid Benefit can save 55%+. These are significant — always ask about existing licenses early in the conversation.
How to verify eligibility: Customer needs to provide their EA or license details. Use microsoft_docs_search for "Azure Hybrid Benefit eligibility requirements" to give them the current checklist.
Priority 4: Architectural Optimization (Months, highest impact)
Design-level changes that reduce cost structurally:
- Move from VMs to PaaS (App Service, Container Apps) — eliminates OS management cost
- Move from provisioned to serverless (Functions Consumption, Cosmos DB serverless) — pay only for usage
- Implement storage lifecycle policies (Hot → Cool → Cold → Archive automatically)
- Use CDN for static content instead of serving from compute
- Consolidate databases — five underutilized Azure SQL databases may cost more than one elastic pool
- Replace always-on batch VMs with Container Apps Jobs or Azure Batch spot instances
Dev/Test Cost Strategies
Non-negotiable for every engagement:
- Apply dev/test pricing on non-production subscriptions (up to 55% savings on Windows VMs)
- Auto-shutdown VMs at end of business day (schedule via Azure policy)
- Use B-series (burstable) VMs for all dev/test workloads
- Azure SQL Basic or serverless tier for dev databases
- Storage with LRS (no geo-redundancy needed for dev data)
- Delete and recreate environments with IaC instead of keeping them running
Dev/test subscription setup: Create a separate subscription linked to an EA Dev/Test offer or Visual Studio Enterprise subscription. This unlocks discounted pricing automatically for most services.
Cost Allocation and Chargeback
Tagging strategy (minimum viable):
| Tag | Purpose | Example |
|---|
| cost-center | Financial chargeback | CC-4521 |
| environment | Dev/test vs production pricing | prod, staging, dev |
| owner | Who to contact about resources | team-platform |
| project | Group costs by initiative | project-phoenix |
| managed-by | IaC or manual | terraform, manual |
Enforcement:
- Use Azure Policy to require tags at resource group level (deny deployment without tags)
- Inherit tags from resource group to resources automatically via policy
- Run monthly compliance reports — untagged resources cannot be charged back
Cost Management views:
- Create cost views per cost-center tag for finance team
- Create cost views per environment for engineering leads
- Set up budget alerts at 50%, 75%, 90%, and 100% of monthly allocation
- Export cost data to Storage Account for custom reporting
Budget and Alert Configuration
Alert thresholds:
| Threshold | Action |
|---|
| 50% of monthly budget | Informational email to team lead |
| 75% of monthly budget | Warning to engineering manager |
| 90% of monthly budget | Alert to finance + engineering director |
| 100% of monthly budget | Action required — review and justify or reduce |
| Anomaly detection | Immediate alert on cost spikes (Azure Cost Management anomaly alerts) |
Anomaly detection: Azure Cost Management has built-in anomaly detection. Enable it. A sudden 3x spike in a single day usually means a misconfigured autoscale, a runaway deployment, or a forgotten dev resource scaled to production tier.
TCO Analysis Framework
When comparing on-premises to Azure (or vice versa), include all costs:
On-premises costs people forget:
- Power and cooling (typically 30-40% of hardware cost)
- Physical space and rack rental
- Network hardware (switches, firewalls, load balancers)
- Staff time for patching, hardware refresh, firmware updates
- Software licenses for hypervisor, backup, monitoring
- Overprovisioning buffer (typically 30-50% extra capacity for peaks)
- Disaster recovery site costs (secondary datacenter)
Azure costs people forget:
- Data egress charges (outbound to internet or other regions)
- Premium support contract ($1,000/month for Business, more for Enterprise)
- Third-party marketplace costs (NVAs, security tools)
- Training and certification for team
- Management and monitoring tooling
- ExpressRoute circuit monthly fees
TCO comparison rules:
- Compare 3-year TCO, not monthly cost
- Include labor costs for both scenarios
- Factor in the opportunity cost of staff spending time on infrastructure vs. features
- Be honest about Azure costs — hiding egress charges makes the analysis lose credibility
Use microsoft_docs_search for the current Azure Pricing Calculator and TCO Calculator URLs and guidance.
Spot VMs
When to use:
- Batch processing that can be interrupted and restarted
- Dev/test environments where occasional eviction is acceptable
- Stateless scale-out tiers behind a load balancer (eviction loses one node, not the service)
- CI/CD build agents (rebuild on eviction)
When NOT to use:
- Production databases (eviction = data loss risk)
- Singleton services with no redundancy
- Workloads that cannot handle interruption within 30 seconds
Savings: Spot VMs are typically 60-90% cheaper than pay-as-you-go. Use microsoft_docs_search to check current spot pricing and eviction rates for specific VM families and regions.
Eviction strategy:
- Use eviction type "Deallocate" (not Delete) to preserve disks
- Set max price to -1 (pay up to pay-as-you-go, accept eviction only on capacity) for lowest eviction risk
- Implement graceful shutdown handlers in your application
- Combine spot with on-demand in a scale set (spot for baseline, on-demand for overflow)
Azure Advisor Integration
Review cadence:
- Weekly: automated digest of cost recommendations to team channel
- Monthly: human review of all Advisor recommendations with action/dismiss decisions
- Quarterly: full cost optimization review with stakeholders
High-value Advisor recommendations to prioritize:
- Shut down underutilized VMs (immediate savings)
- Buy reservations for consistent usage (30-72% savings)
- Right-size or shut down underperforming VMs
- Use Standard SSDs where Premium is unnecessary
- Delete unused public IP addresses ($3.65/month each adds up)
MCP Tool Usage
- Use
microsoft_docs_search for current pricing for specific Azure services, reservation discount percentages, Hybrid Benefit eligibility, and Spot VM pricing
- Use
microsoft_docs_fetch for detailed Azure Cost Management setup guides, budget API documentation, and Advisor REST API reference
- Always verify pricing through MCP tools — Azure pricing changes frequently and varies by region, EA agreement, and tier
- When recommending Azure CLI commands for cost management (az consumption, az costmanagement), fetch current syntax through MCP tools
- Use
microsoft_docs_search with "Azure Pricing Calculator" to direct customers to the current calculator for detailed estimates
