Most SMB Azure bills do not get worse because of one bad decision. They get worse because of dozens of small, reasonable-looking ones, made by different people, over a period of months. By the time finance flags the number, the easy savings are already months old.
Here are the five lines we most often find quietly compounding when we run a cost review for a UK SMB. None of them are exotic. All of them are fixable in under a fortnight.
1. Unattached managed disks
When a virtual machine is deleted, its operating-system disk goes with it, but data disks routinely do not. They sit, attached to nothing, billing month after month. We have seen estates of fewer than fifty VMs carrying north of three terabytes of orphaned premium SSD.
The fix: a monthly report of unattached disks, with a 30-day grace period before deletion. Five minutes to set up, often hundreds of pounds a month back.
2. Reserved-instance drift
Reservations were sensible when they were bought. Then workloads moved region, were resized, or quietly retired — and the reservation kept paying for capacity nobody is using. Microsoft will not refund this automatically; they will let it run its three years.
The fix: a quarterly reservation review against actual utilisation. Where drift is significant, exchange or scope reservations to where the consumption actually is.
3. Egress that nobody planned for
Data going into Azure is free. Data coming out is not. The cost rarely matters until somebody enables a backup-to-on-prem, or replication to another region, or a SaaS integration that quietly pulls a few gigabytes a day. The line item is buried inside "Bandwidth" and almost never owned.
The fix: tag and alert on egress per workload. Move chatty integrations into the same region where possible.
4. Premium-tier services on dev environments
Production sets the template. Someone copies the IaC into a dev subscription. Suddenly your dev environment is on premium storage, geo-redundant backups and Standard_DS-series VMs that idle 22 hours a day. Dev typically costs SMBs more than it should by a factor of two to four.
The fix: a separate cost target for non-production, with cheaper SKUs, auto-shutdown schedules and aggressive backup retention.
5. The "we will tidy it up later" tier
Every estate has them: test resource groups from 2022, a half-built proof of concept, a load balancer somebody added during an incident. None of them are loud. All of them are paid for.
The fix: a monthly cost-anomaly review, plus an owner field on every production resource group. If an item has no owner, it has no business running.
A sensible operating rhythm
For most SMBs, the answer is not a heavyweight FinOps platform. It is a small set of recurring habits, owned by one person, reviewed quarterly by finance. We can usually set that up — and recover the first three months of savings — inside a two-week engagement.
