Azure Advisor Cost Recommendations: Enterprise Implementation Guide

Azure Advisor's cost recommendations carry a credibility problem: Microsoft advertises the headline savings figure, which routinely overstates practical realisation by 2–3×. I've seen enterprises make budget commitments based on an Advisor headline of $2.4M in annual savings, then realise $900K after validation — a 38% realisation rate. This guide covers how to read Advisor recommendations accurately, prioritise actions by confidence and ROI, and extract the genuine savings while filtering the noise.

How Azure Advisor Generates Cost Recommendations

Azure Advisor analyses resource utilisation data from Azure Monitor and billing data from Cost Management to generate recommendations across five categories. For cost specifically, Advisor uses 7–30 days of utilisation metrics depending on the recommendation type. Right-sizing recommendations use 7–14 days of CPU utilisation. Reserved Instance recommendations use 30 days of continuous resource deployment data.

The critical limitation: Advisor uses CPU utilisation as its primary signal for right-sizing. It does not analyse memory utilisation (only available through Azure Monitor Agent with additional configuration), application-level throughput metrics, or peak vs average utilisation curves. A VM running at 3% average CPU but with 15-minute spikes to 85% for a nightly batch job will appear as an idle VM in Advisor's view — but resizing it would cause the batch job to fail.

The Five Cost Recommendation Categories

Right-Sizing Recommendations: The Validation Framework

Never implement a right-sizing recommendation on a production workload without application team validation. The three-step validation process:

Step 1 — Classify the workload type. Identify what application runs on the VM. Monitoring agents, log forwarders, authentication agents, and backup agents have legitimate low CPU profiles. Classify these as "exempt" and dismiss the Advisor recommendation as Not Applicable. Development and test VMs with no recent commits or test runs are candidates for deletion, not right-sizing.

Step 2 — Analyse peak utilisation, not average. Advisor shows average utilisation. Pull 30 days of CPU data from Azure Monitor and look at the 95th percentile utilisation. A VM with 3% average and 85th percentile of 35% can be safely right-sized to a SKU that handles 35% load. A VM with 3% average and 98th percentile of 92% should not be resized — the average is misleading.

Step 3 — Right-size in non-production first. For application workloads, implement the right-sizing recommendation in the development environment, run a representative load test, and verify performance before applying to production. For batch-processing workloads, simulate the peak batch load on the smaller SKU. This adds 1–2 weeks to implementation but eliminates the risk of production incidents from premature right-sizing.

The Right-Sizing Prioritisation Matrix

Reserved Instance Recommendations: High Confidence, Act Quickly

Reserved Instance recommendations are Advisor's most reliable cost reduction signal. When a VM series has run continuously in a specific region for 30 days, the probability it will continue running is high enough that a 1-year reservation generates meaningful savings at acceptable risk.

The standard RI recommendation discount schedule:

• 1-year reservation: ~30–40% discount vs pay-as-you-go (varies by VM series)
• 3-year reservation: ~55–65% discount vs pay-as-you-go
• 1-year Savings Plan (compute): ~20–30% discount with workload flexibility
• 3-year Savings Plan (compute): ~40–50% discount with workload flexibility

Advisor will recommend the option that maximises your savings — typically 3-year reservations for production workloads that have shown 90 days of stability. Before accepting a 3-year recommendation, validate that the workload will exist in 3 years: check the application roadmap for planned migrations to PaaS services or application retirement.

Unattached Managed Disks: The Easy Win

Unattached managed disks are the highest-confidence cost reduction in Advisor. A managed disk not attached to any VM has no business justification — it's either orphaned (the VM was deleted but the disk wasn't), a deliberate data preservation snapshot (which should be converted to a lower-tier storage), or an oversight in disk cleanup procedures.

The implementation process is simple:

1. Export the list of unattached disks from Advisor
2. For each disk, query the last attachment date using Azure Resource Graph
3. Disks detached more than 30 days ago: tag as "PendingDeletion" and notify the last known VM owner
4. After 14 days with no objection, delete the disk
5. For disks detached less than 30 days ago (recent VM deletion): confirm with application team before deletion

This process consistently recovers 85–90% of Advisor's unattached disk savings estimate. The typical enterprise with 3–5 years of Azure usage has $30,000–$150,000 in unattached managed disk spend annually — all completely recoverable.

Integrating Advisor with Your FinOps Workflow

Advisor recommendations should feed into a structured FinOps workflow, not be acted on ad-hoc. The recommended cadence:

Weekly: Export Advisor cost recommendations via the Cost Management API or Azure Resource Graph. Feed into a Power BI dashboard showing total identified savings by category, team, and subscription. Route new high-value recommendations (>$5K/month) to the relevant engineering team via ServiceNow ticket or Teams notification.

Monthly: FinOps team reviews all open recommendations with engineering leads. Approve or dismiss each recommendation with documented rationale. Track "dismissed" recommendations separately — a growing dismissed pile indicates systematic issues (either over-aggressive recommendations or cultural resistance to optimisation).

Quarterly: Measure Advisor Cost Score progress. Set a target improvement (e.g., increase from 55% to 70% over the quarter). Calculate actual savings realised from Advisor-sourced actions against the quarter's identified savings. The Advisor realisation rate tells you whether your validation and implementation process is working.

See our Azure FinOps Advanced Governance guide for the full governance framework that Advisor integrates into.

Advisor Score: What It Measures and What to Target

The Advisor Cost Score measures the percentage of Advisor's identified cost savings opportunities that you've implemented. A score of 100% means you've acted on every recommendation Advisor has surfaced. In practice, targeting 75–85% is the right range for most enterprises — above 85% means you may be implementing recommendations that don't survive operational validation, creating technical risk; below 60% means significant recoverable savings are sitting unactioned.

The score is calculated as: (Sum of impact-weighted recommendations implemented) ÷ (Sum of all impact-weighted recommendations) × 100. High-impact recommendations have more weight in the score calculation, so implementing the top 20% of recommendations by savings will move the score dramatically more than addressing the long tail of small recommendations.

Frequently Asked Questions

How accurate are Azure Advisor cost recommendations?

Right-sizing recommendations have a 40–60% practical realisation rate. Reserved Instance recommendations achieve 85–95% realisation for stable workloads. Unattached disk recommendations achieve 80–90% realisation. Always validate right-sizing recommendations against application team knowledge and peak utilisation data before implementation.

What data does Azure Advisor use to generate recommendations?

Advisor uses 7–30 days of CPU, disk, and network utilisation data from Azure Monitor. Right-sizing uses 7–14 days; RI recommendations use 30 days. Recommendations update weekly. Advisor does not analyse memory utilisation without Azure Monitor Agent — a significant gap for memory-intensive workloads.

Can I dismiss Azure Advisor recommendations?

Yes. Dismiss for 7, 14, 30, or 90 days, or permanently mark as Not Applicable. Use dismissal for workloads with legitimate low-utilisation profiles to reduce noise in the recommendations list and improve signal-to-noise ratio for your team.

What is the Azure Advisor score?

A 0–100% measure of how well you've implemented Advisor's recommendations. Target 75–85% for enterprise FinOps maturity — above 85% risks implementing recommendations that fail operational validation; below 60% indicates significant unrealised savings.

Should I always follow Azure Advisor right-sizing recommendations?

No. Validate against application team knowledge and peak utilisation data before implementing on production. 20–30% of "idle" VM recommendations involve workloads with legitimate low-average but high-peak utilisation profiles that should not be resized.