How would you identify workloads that are running and serving no purpose, and how would you make removing them safe?

What is being tested

Whether you can find waste with evidence and remove it without causing an incident.

Finding them

1. Traffic data. A service with no inbound requests for 90 days is the strongest single signal, and it is rarely consulted. Load balancer and mesh telemetry answer this definitively.

2. Deployment activity. Nothing deployed for a year, no commits to the repository, no owner responding — a compound signal that is stronger than any one part.

3. Dependency analysis. Nothing calls it, and it appears in no service catalogue entry as a dependency.

4. Orphaned infrastructure, which is easier: unattached volumes, idle load balancers, unused addresses, empty clusters, snapshots of instances that no longer exist.

5. Non-production environments from projects that ended. Frequently the largest single category and the easiest to confirm.

6. Ask. Publish the candidate list to owners with a deadline. The process of asking removes a meaningful amount of spend on its own, because people remember things they had forgotten.

Making removal safe

1. Verify before deleting. A workload with no traffic may be a disaster recovery standby, a quarterly or annual job, a compliance archive, or a break-glass path used once a year. These are exactly the things that look like waste and are not, and deleting one is how a cleanup programme loses its mandate permanently.

2. Notify the owner with a window. Fourteen days, with a clear statement that no response means deletion. If nobody responds, that is itself evidence.

3. Disable before deleting. Stop the workload, remove it from routing, and leave it for a further period. If something breaks, restoring is immediate. This two-phase approach converts an irreversible action into a reversible one and is worth the extra time.

4. Snapshot before deleting where practical, retained for a defined period.

5. Record what was removed and when, so a later incident can be correlated with it.

Making it stick

One-off cleanups regress within a year. What holds:

• Automated detection running continuously, reporting to owners rather than to a central team.
• Ownership tags enforced at creation, so "who owns this?" always has an answer.
• Expiry by default on non-production resources — created with a lifetime, extended deliberately, deleted automatically. This is the strongest single control and it prevents the category rather than cleaning it up.
• Deletion as the default for the unclaimed, after notice.

What a strong answer adds

That waste elimination is the right place to start precisely because it involves no trade-off — it is the only cost work that is uncontroversial. But it should not be where the programme ends: the larger savings are architectural, and a programme that only ever demonstrates small easy wins never earns the mandate to attempt them.