Turning It Off

Nobody plans the end of a video system. The result is cameras running for years after the project that justified them, holding data nobody is managing.

Why it matters

The basis expires. A justification tied to a specific problem does not survive the problem being solved.

The data does not expire by itself. Recordings, detections, templates and training data persist under whatever retention was configured, or none.

Nobody owns it, so nothing is reviewed, evaluated or deleted.

It still creates obligations: access requests, breach exposure, and an answer required if someone asks what it does.

And it still looks like a control, so people assume something is watching when nothing is.

Deciding to stop

Review each deployment annually against its stated purpose.

Does the purpose still exist?

Is the system still performing? If it has not been evaluated in two years, the honest answer is unknown.

Is anyone acting on the output? A system whose alerts nobody reads has already stopped.

Would you deploy it today, given what you now know? The most useful question and the least often asked.

The checklist

Stop the processing, and confirm it stopped rather than assuming.

Delete the data, by category: recordings, detections, templates, training data, exports, backups.

Verify the deletion, including backups, with a record.

Revoke access for everyone who had it.

Remove or repurpose the cameras, and where they remain for plain recording, say so.

Update the notices. Signage describing analysis that no longer happens is a small dishonesty that accumulates.

Tell the people who were subject to it, which is rarely done and is the decent thing.

Record that it was decommissioned, when, and what was deleted.

Templates specifically

Biometric templates have their own destruction obligations under several statutes, with defined schedules.

Deletion on the person leaving, not on the system ending.

Verified, with a record, because this is the item most likely to be the subject of a claim.

Including any copies held by the vendor, which requires asking and getting an answer in writing.

The vendor exit

What happens to your data in their infrastructure?

Ask before signing, not at termination.

Deletion within a stated period, confirmed in writing.

Whether your footage was used to train their models, and what that means for data you asked to be deleted — the honest answer is frequently that it cannot be removed from a trained model, which is a reason to have refused that use at the start.

Export of your own detection history, if you want to keep the operational record.

The abandoned deployment

The common case, worth naming.

A pilot that ended without a decision, still running.

A system from a project team that dissolved.

Cameras added for an event three years ago.

Find them: inventory every camera and every analytics deployment, with its purpose, owner and review date.

Anything without an owner or a current purpose is a decommissioning candidate, and the inventory is usually the first time anyone has counted.

The annual question

One question per deployment that produces most of the decommissioning decisions.

Would you deploy this today, given what you now know?

Not: is it working. Not: is anyone complaining.

Given the current purpose, the measured performance, the obligations and the alternatives now available.

A no is not a criticism of the original decision, which was made with different information.

Ask it in the annual review and record the answer, because nobody asks it unless the process asks it.

Finding the abandoned deployments

Every organisation has some, and they are invisible until counted.

Pilots that ended without a decision, still running.

Systems from a project team that dissolved.

Cameras added for an event years ago.

Features enabled during a trial and never disabled.

The inventory finds them, which is the argument for building it even where no new deployment is planned.

Anything without an owner or a current purpose is a decommissioning candidate, and clearing them removes obligations at no operational cost.