Most Azure bills have 20–35% waste. Some of it is genuinely hard to remove; most is sitting in plain sight. Here's the quarterly playbook we run with clients to keep cost under control without slowing down delivery.
The cardinal rule
Optimize after a workload is stable, not before.
Premature optimization kills delivery velocity. Build first, run for 30 days, then tune. Repeat.
Quick wins (week 1)
Open Cost Management. In one hour you'll find:
- Stopped-but-not-deallocated VMs — paying full compute. Deallocate.
- Orphaned managed disks — VM gone, disk still billed.
- Unattached public IPs — small line items, lots of them.
- Unused App Service Plans — empty plans still cost.
- Premium SSDs on dev workloads — Standard SSD is plenty.
- Old snapshots that should have been deleted.
- Defender for Cloud Standard on subscriptions that don't need it.
- Log Analytics excessive ingestion — table-level retention review.
- Premium tier services running 24/7 in non-prod.
- Test environments running 24/7. Schedule shutdown nights and weekends.
This list alone commonly recovers 5–15% in week one.
Right-sizing (week 2–3)
Use Azure Advisor recommendations as the starting point, then verify:
- VM SKUs — many workloads are 2–4 sizes over-provisioned.
- App Service plans — scale to the right tier and instance count.
- SQL DTUs / vCores — DTU graphs show the truth.
- AKS node pools — CPU and memory utilization per pool.
Be conservative. Right-size 20% at a time, monitor, then continue.
Reservations and savings plans (month 2)
Once you have 90 days of stable usage:
- Reserved Instances — 1 or 3 year commitment, biggest discount, locks SKU.
- Savings Plans — flexible compute commitment, smaller discount, much more flexibility.
Rule of thumb:
- For predictable steady-state workloads: Reservations win.
- For mixed or evolving workloads: Savings Plans win.
- Most clients land on 70% Savings Plans + 30% Reservations as the right blend.
Don't commit on month-one estimates. You'll over-commit and lose flexibility.
Hybrid Use Benefit
If you have eligible Windows Server or SQL Server licenses:
- Apply Azure Hybrid Benefit on every applicable VM.
- For SQL on Azure SQL Database / Managed Instance, AHB applies too.
Often forgotten. Often worth 30–40% on those workloads.
Storage tiering
For Blob Storage:
- Hot for actively accessed.
- Cool for infrequent (>30 days).
- Cold / Archive for compliance retention.
Lifecycle management policies move data automatically. Set them up once, save indefinitely.
Data transfer
Data leaving Azure costs. Patterns:
- Use Azure Front Door for global users — caches at the edge.
- Keep data and compute in the same region.
- Use Private Link / Service Endpoints to avoid public egress where possible.
- Watch cross-region replication costs — sometimes the simple choice is more expensive than you think.
Tagging and chargeback
You can't optimize what you can't allocate. Required tags:
costCenterenvironment(prod, non-prod, sandbox)owner(an actual person, not a team alias)application
Enforce via Azure Policy. Reject untagged resource creation.
Then build a Power BI report on Cost Management exports that shows spend per tag combination. Distribute monthly.
The 80/20 of cost reviews
Don't spend hours on everything:
- The top 10 spend lines per subscription almost always represent 80%+ of cost.
- Focus optimization there.
- The long tail is rarely worth the effort unless you find systemic patterns.
Quarterly playbook
Cycle this every quarter:
- Week 1: quick wins sweep (the 10 above).
- Week 2–3: right-sizing review and execution.
- Week 4: reservations / savings plan review and update.
- Week 5: tag and chargeback review.
- Week 6: workload-specific deep dive (rotate through compute, storage, networking, data, AI).
Track savings claimed and savings realized. Auditors love the gap; close it.
When to bring in outside help
If your bill is growing >20% per year and you're not adding equivalent business value, get an external review. A FinOps engagement should pay for itself in the first month.
FAQs
Should we use Azure Advisor or third-party tools? Advisor is solid baseline. Third-party tools (Cloudability, CloudHealth, Vantage, custom) add multi-cloud, deeper analytics and showback. Worth it once your bill exceeds ~$50K/month.
How aggressive can we be with non-prod shutdowns? Very. Most non-prod can run 8am–8pm weekdays only — saves ~70% on those workloads. Use Automation Accounts or AKS Cluster Autoscaler.
Do reservations apply to all services? Major compute services (VMs, SQL, Synapse, App Service Premium v3, AKS reserved nodes), yes. Many PaaS services no. Read the small print per service.
What about spot instances? For batch / fault-tolerant workloads, spot saves 60–90%. For anything customer-facing, don't risk it.
We run quarterly Azure FinOps engagements that consistently recover 15–30% of bill. Get a sample report — we share before any commitment.