Azure cost · field note
Reserved Instances before rightsizing is how you lock in your overspend for a year.
A discount is not the same as efficiency. If the VM is twice the size you need, paying less for it is still paying for the wrong VM.
Reservations are one of Azure's easiest savings mechanisms to explain and one of the easiest to buy at the wrong time. The portal shows a discount. It cannot prove that the underlying capacity deserves to exist.
The mistake starts with a sensible instruction: reduce our compute cost. Someone sees a VM running every hour, calculates the one-year reservation saving, and buys the commitment. Finance sees an immediate improvement. The architecture remains oversized.
You have made the waste cheaper—and harder to unwind.
01 / Two different questions
Price optimisation and resource optimisation are not interchangeable
Rightsizing asks, “What capacity does this workload need?” A Reservation asks, “How much of this eligible usage are we confident enough to commit to?”
You cannot answer the second question well until you have answered the first.
VM size, count, schedule, tier, or architecture.
The discounted rate applied to matching eligible usage.
Whether the resource should run at all.
Whether the same waste returns next quarter.
02 / A simple example
The saving can be real and the decision can still be wrong
Consider a deliberately simple illustration. An oversized VM costs $1,000 per month on pay-as-you-go pricing. A one-year reservation reduces the effective cost to $700. The dashboard calls that a $300 monthly saving.
But suppose a smaller VM, validated against the actual workload, costs $500 pay as you go—and less again once the correct size is reserved. Buying first made the expensive baseline look efficient.
- Oversized pay as you go: $1,000/month
- Oversized with reservation: $700/month
- Rightsized pay as you go: $500/month
- Rightsized, then committed: potentially lower again
The exact Azure discount varies by service, term, region, offer, and configuration. The principle does not: optimise the baseline before you discount it.
03 / The order of operations
Baseline. Rightsize. Validate. Commit.
- Map the workload. Identify owner, dependencies, region, lifecycle, availability requirements, and business peaks.
- Measure representative demand. Include month-end, batch, seasonal, deployment, and failover events—not only a quiet fortnight.
- Remove and rightsize. Eliminate resources with no valid workload and resize active services with a rollback plan.
- Validate the new baseline. Watch performance, reliability, and cost long enough to distinguish stability from luck.
- Model the commitment. Buy coverage for the predictable floor, not the optimistic ceiling.
This is also why an Azure Advisor rightsizing recommendation needs workload context. The recommendation is evidence. It is not change approval.
04 / The evidence
A 24/7 resource is not automatically a 24/7 requirement
Reservation candidates are often identified from continuous usage. But continuous usage may simply mean nobody configured shutdown. A development VM that runs nights and weekends is predictable; it is not necessarily required.
Before committing, check CPU, available memory where collected, disk latency and IOPS, network throughput, application response time, queue depth, scheduled jobs, restart history, scaling behaviour, and business calendar. Then ask the more important question: does this workload still belong on this resource at all?
Reservations work best against durable, well-understood demand. Bursty growth, short-lived projects, planned migrations, and uncertain product lines need flexibility more than an impressive coverage percentage.
05 / The purchase
Only then decide how much certainty you actually have
Once the smaller baseline is stable, review eligible scope, region, VM family or service constraints, existing benefits, expected utilisation, and the likelihood of architectural change. Azure offers one-year and, for many reservation types, three-year terms. A longer term can improve pricing while increasing the cost of being wrong.
Do not use 100% reservation coverage as a vanity metric. Leave genuinely variable demand variable. Monitor reservation utilisation after purchase, because organisational changes, deployments, and migrations can strand a commitment that was sensible on day one.
In our 65% Azure cost reduction case study, the sequencing mattered: remove waste and rightsize the estate before using commercial commitments to optimise what remained.
06 / The nuance
Sometimes the decisions can run in parallel—but the baseline still needs evidence
A mature estate may already have well-observed, stable core demand and an expiring commercial deadline. In that case, a team can model commitments while a wider optimisation programme continues. Existing reservations may also influence which workloads move first.
That is not permission to skip rightsizing. It is permission to distinguish the stable floor from the uncertain edge. If you cannot name the workload, owner, lifecycle, and safe operating size, you are not buying certainty. You are buying a discount on ambiguity.
Primary references
Sources and further reading
- Microsoft: Save with Azure Reservations — official reservation scope, purchasing, and term guidance
- Microsoft: Azure Advisor cost recommendations — official context for cost and rightsizing recommendations
- Microsoft Azure Well-Architected Framework: Cost Optimization — workload cost optimization principles
Questions teams ask
Frequently asked questions
Should I rightsize Azure VMs before buying Reserved Instances?
Usually, yes. Rightsizing establishes the capacity the workload actually needs. Once that new baseline is stable, you can decide how much predictable usage is safe to commit. Otherwise, the reservation may discount an oversized VM.
Do Azure Reserved Instances reduce utilisation waste?
No. A reservation can reduce the unit price of eligible, matching usage, but it does not reduce idle CPU, excess memory, or an unnecessary VM count. It changes price, not demand.
How long are Azure Reservations?
Azure Reservations commonly use one-year or three-year terms, depending on the service and offer. The operational point is the same: validate workload shape and stable demand before making a commitment.
What should I check before reserving Azure VM capacity?
Check the workload owner, representative utilisation, required VM family and region, expected lifecycle, existing reservations, utilisation and coverage, exchange or cancellation conditions, and whether modernization or migration is likely during the term.
Are Azure Savings Plans a substitute for rightsizing?
No. Savings Plans can provide more flexibility across eligible compute usage, but the commitment still assumes a stable hourly spend. Rightsizing and removing waste should happen before sizing that commitment.