Docs / FinOps / Recommendation Approval and Rollback Safety

Recommendation Approval and Rollback Safety

Published · Last updated: May 2026 · 3 min read

Who this is for

Anyone acting on FinOps recommendations (right-sizing, deleting idle resources, releasing unattached IPs) who wants to do so safely.

What you will complete

Review a recommendation, understand the risk level, approve or dismiss it, and know what rollback options exist.

Before you begin

  • Navigate to FinOps → Waste or the Recommendations section.

What a recommendation contains

Each recommendation card shows:

  • Category: The type of waste (idle server, right-size, orphan volume, etc.)
  • Resource name: The specific server, volume, IP, or snapshot
  • Cloud provider and account
  • Estimated monthly saving: How much you would save by acting on this recommendation
  • Evidence: What data led to this recommendation (e.g., "CPU averaged 4% over 30 days")
  • Suggested action: What to do (resize to X instance type, release the IP, delete the snapshot)

Step-by-step: review and act on a recommendation

  1. Go to FinOps → Waste or the Recommendations section.
  2. Read the recommendation card fully — especially the evidence section.
  3. Ask yourself these questions before acting:
  • Is this resource actually unused, or does it serve a purpose I am not aware of?
  • Is there a team member who might know why this resource exists?
  • Is there a recent backup or snapshot if I need to recover?
  1. If you want AI Pilot's analysis: click Ask AI Pilot or go to AI Pilot and ask "Please analyze the recommendation to right-size [server-name]. Is it safe to proceed?"
  2. If confident: take the action in the dashboard (resize, stop, or delete).
  3. If unsure: click Dismiss to remove the finding temporarily (it will reappear in the next cycle if the condition persists).

Safety rules before acting on each recommendation type

Right-size (server resize):

  • Create an on-demand backup before resizing. Resizing a live server involves a restart.
  • Verify the new instance type is available in the server's current region.
  • Test performance after resize — smaller instances may struggle under peak load.
  • Rollback: if the resize causes performance issues, resize back up. Both directions are possible.

Idle server (stop or delete):

  • Confirm with the team who last used it.
  • Check the audit log for when it was last active.
  • Stop first (do not delete) — stopping preserves data but saves compute cost. Monitor for 7 days.
  • If no one needs it after 7 days of being stopped: delete.
  • Rollback: if stopped, restart. If deleted, there is no rollback — ensure a backup exists first.

Orphan volume (delete):

  • Take a snapshot of the volume before deleting it.
  • Verify in your cloud provider console that no server references this volume.
  • Rollback: restore from the snapshot you took before deletion.

Orphan snapshot (delete):

  • Verify the snapshot is older than your retention policy and the source server no longer exists.
  • Rollback: not possible after deletion. Confirm before deleting.

Unattached IP (release):

  • Verify no DNS records or applications point to this IP.
  • Rollback: a new IP can be allocated, but it will be a different IP address. DNS records would need updating.

Dismissing a recommendation

If a recommendation is a false positive (the resource is intentionally idle, or you have already handled it):

  1. Click Dismiss on the recommendation card.
  2. The card is removed from the active list.
  3. CloudAIPilot will not show the same recommendation again for 30 days. If the condition persists, it will reappear.

What success looks like

  • You acted on a recommendation, the resource is resized or removed, and the monthly saving is reflected in the next billing cycle's FinOps dashboard.
  • No service disruption occurred because you followed the safety checklist.

Related articles