Using Power Automate for Azure cleanup
Automate Azure cleanup reports, VM stop schedules and expiry reviews. Check connectors, licences, permissions and safeguards before taking action.

Power Automate can help turn a cloud cleanup list into repeatable work: identify a candidate, ask its owner, carry out the action they agree to and record the result.
Start with the decisions in your cloud-cost review. Automation needs those decisions before it can safely stop or remove anything.
Confirm the tools and permissions first
Microsoft lists the Azure VM connector and Azure Resource Manager connector as Premium in Power Automate. Check the licence covering your flow, available actions, region and connection identity before designing it.
The Resource Manager connector can list and read resources and perform management operations. Metrics, billing detail and some inventory queries may need separate supported APIs or tooling. Verify the complete route instead of assuming one connector supplies every input.
Restrict the connection’s Azure permissions to the resources and actions it needs. Separate read-only discovery from a connection that can change resources. Assign owners for the flow and its connections, and decide who receives failures or can disable execution.
Workflow one: schedule agreed development VMs
Use an explicit list of development machines whose owners have agreed to a stop schedule. Confirm dependencies, overnight jobs, backups and a restart process. A missing Production tag or a low CPU reading doesn’t mean a machine is disposable.
A scheduled flow can read the eligible machine’s current state, check whether it’s on the exceptions list and invoke the intended operation. Test it on a non-critical machine, then confirm that the application is usable after restart.
The VM connector distinguishes Power off from Deallocate. Powering off can leave compute charges running; deallocation releases the compute allocation. Disks and other chargeable resources can remain, and existing commitments may affect the bill. Check the effect on the next bill before quoting a saving.
Workflow two: report unattached disks for review
Use a scheduled inventory job to produce a candidate list with resource IDs, owners, size and the reason for review. Microsoft’s unattached-disk guidance describes examining attachment information and reviewing candidates before deletion.
An unattached disk may be retained for recovery, migration or a planned deployment. Ask the owner to confirm its purpose, required retention and any backup or restore dependencies. Don’t make the absence of an attachment a delete instruction.
Power Automate can route that inventory to an approval process even when another tool gathers the detailed disk information. Keep the original resource identifier in the decision record so similarly named resources can’t be confused.
Workflow three: review temporary-resource expiry
Record an owner, purpose and review date when a temporary resource is created. A scheduled flow can identify overdue items and ask the owner to renew it or confirm cleanup. Missing owners and invalid dates should produce an exception for review.
Before acting, reread the resource and confirm that the owner’s decision still fits its current state. Check required records, dependencies and recovery arrangements. Restrict execution to explicit resource types and scopes; don’t delete a whole resource group because one member is overdue.
Set a defined approval deadline and keep resources when a request is unanswered or fails. Microsoft’s approval guidance describes a separate design for workflows that need to run beyond 30 days. Don’t treat an expired flow run as permission to proceed.
Test failures as well as successful cleanup
Begin with flows that only report intended actions. Build that behaviour explicitly; there isn’t a universal report-only switch that makes every management action safe.
Test an unavailable API, lost connection permission, denied or unanswered request, changed tags and a repeated run. Log the resource ID, who signed off, the operation and the verified result. Check the resource afterwards, because a submitted request can still fail.
Document how to switch the flow off quickly and review its first active runs. Expand only when the owners can explain the candidate selection and the failure handling.
Talk to us if you need help designing an Azure cleanup process with sign-off and recovery steps.


