Most cloud waste isn't a single dramatic mistake. It's a handful of quiet, ongoing leaks — a disk nobody deleted, a VM someone meant to stop "for now," a SKU chosen two years ago that nobody has revisited since. None of them look urgent on their own. Together, they can be a meaningful share of a monthly Azure bill.
Here are five patterns worth checking for, and how a FinOps Agent conversation surfaces each one without a spreadsheet export.
1. Stopped-but-still-billing compute
Deallocating a VM stops the compute charge, but attached managed disks, static public IPs, and reserved capacity keep billing regardless. A VM that was "shut down last quarter" can still be a live cost line every month after.
Ask your FinOps Agent to list resources with no recent billing activity, and it will surface these directly — framed honestly as "no recent activity in the synced window," not "confirmed unused," since billing data alone can't tell you whether something is truly dead or just quiet.
2. Orphaned disks and public IPs
When a VM is deleted, its managed disks and public IP addresses don't always go with it. These are some of the easiest savings to find and the easiest to miss, because nothing about them looks broken — they just sit there, unattached, still on the meter.
3. Oversized SKUs that made sense once
A workload sized for a launch spike often keeps its SKU long after traffic settles. The fix isn't guessing at a smaller size — it's grounding the recommendation in current utilization and in Microsoft's own up-to-date sizing guidance, since Azure regularly ships new VM series that outperform an older generation at a lower price point.
4. Duplicate or forgotten storage
Snapshots, backups, and log exports accumulate quietly in storage accounts long after the resource they were backing up is gone. A one-time export from a migration two years ago can still be sitting in a Premium-tier container today.
5. Missing reservations or savings plans
Steady-state workloads running on pay-as-you-go pricing are leaving a straightforward discount on the table. This one isn't about finding waste — it's about noticing that a predictable workload hasn't been matched to the pricing model that fits it.
Finding these without a manual audit
The common thread across all five is that none of them are visible from the monthly total alone — they require breaking cost down by resource and asking "does this still need to exist, and is it sized right?" That's exactly the kind of question a FinOps Agent conversation is built to answer: point it at a subscription, and it pulls the actual breakdown, flags what looks idle, and estimates what a change would be worth before you make it.