In almost every Azure environment we assess, we find the same thing: VMs that are massively oversized for their actual workload. It's the single easiest win in cloud cost optimisation.
Why VMs End Up Oversized
It usually happens for understandable reasons:
- Lift and shift - On-prem specs were copied directly to Azure without considering that cloud VMs often perform better
- Safety margins - Teams spec'd for peak load plus a generous buffer "just in case"
- Performance anxiety - Nobody wants to be blamed for slow systems, so they err on the side of too big
- Set and forget - The VM was right-sized initially but the workload changed
The problem is, unlike on-prem, you're paying for every vCPU and GB of RAM whether you use it or not.
How to Identify Right-Sizing Candidates
Azure Advisor
The easiest starting point. Azure Advisor automatically identifies VMs with low CPU utilisation:
- Go to Azure Advisor in the portal
- Click on Cost recommendations
- Look for "Right-size or shutdown underutilised virtual machines"
Advisor flags VMs with average CPU below 5% over 7 days. That threshold is conservative - even VMs at 10-15% average are often good candidates.
Metrics Deep Dive
For a more thorough analysis, check the actual metrics:
- Go to the VM → Metrics
- Add Percentage CPU - look at average over 30 days
- Add Available Memory Bytes - are you using what you're paying for?
- Check Disk IOPS if you're using premium storage
A VM running at 15% CPU with 60% free memory is almost certainly oversized.
Our Approach
When we do a FinOps assessment, we pull 30-90 days of metrics and correlate them with cost data. This lets us:
- Identify patterns (is it low all the time, or just outside business hours?)
- Calculate actual savings potential in pounds
- Recommend specific target SKUs
Common Right-Sizing Opportunities
D-series to B-series - If your workload is "bursty" (occasional spikes but generally low utilisation), B-series VMs are much cheaper. A D4s_v3 costs about £115/month; a B4ms costs about £60/month.
Dropping a size - Simply going from D4 to D2 halves your cost. If you're averaging 20% CPU on a D4, you'll be at 40% on a D2 - still plenty of headroom.
Memory-optimised to general purpose - E-series VMs are pricey. If you don't actually need all that RAM, switch to D-series.
Real Example
A client had 24 VMs running their internal applications. After analysis:
- 8 VMs were running under 10% CPU average
- 6 more were under 20%
- Total spend was £4,200/month
- After right-sizing: £2,800/month
- Annual saving: £16,800
And that was just VMs - we hadn't touched storage or other services yet.
The Right-Sizing Process
- Identify candidates using Advisor and metrics
- Validate with application owners - make sure there isn't a reason for the sizing
- Test in non-prod first if possible
- Resize during a maintenance window (requires a restart)
- Monitor for a week to ensure performance is acceptable
- Repeat quarterly - workloads change
Don't Forget Reserved Instances
Right-sizing and Reserved Instances work together. There's no point buying a 3-year reservation for a D8 if you should be running a D4. Right-size first, then commit.
Want to know how much you could save by right-sizing? Our free savings snapshot identifies your oversized VMs and calculates the potential savings.