From existing cameras to operational evidence

Most factory CCTV proves nothing on its own

A typical Malaysian plant already records dozens of cameras around the clock. The footage exists. What does not exist is an answer to the questions supervisors actually carry between shifts: was the station attended during operating hours, was the emergency exit clear at 2 a.m., was the receiving bay tidy at handover, did the crew follow the procedure that the customer audit will ask about next month.

Those questions are usually answered by a person walking the floor, a spot check, or a long afternoon scrubbing through recordings after something has already gone wrong. The camera estate is treated as a deterrent and an archive, not as an operational instrument.

AI video analytics for a Malaysian factory does not need to start with new cameras. It needs to start with one plainly stated question per scene.

Start with the question, not the camera

Wyse Envision runs scheduled checks against plain-language rules that you define. The rule is the product. “Is the station attended during operating hours?” “Is the exit blocked?” “Is the bay clear of goods before the next truck?” A rule written like that can be checked by a supervisor, by a vision-language model, and by an auditor, and all three can agree on what a pass and a fail look like.

That is also why the first activity in any engagement is not installation. It is a site and scenario readiness study: which cameras can actually see the condition, whether the angle, lighting and resolution are enough to judge it, who owns the response when an exception fires, and what the manual baseline costs today.

The output of a good first month is not a dashboard. It is a written go / no-go for one scenario, on your cameras, with a named owner.

What “evidence” means in practice

Recording is not evidence. Evidence is a specific frame or clip, with a timestamp, attached to a specific exception, reviewed by a named person, with a recorded decision. Wyse Envision produces that chain for every check it runs:

• Scheduled check — the rule runs on the cameras that were validated for it, at the times that matter for the shift.
• Exception with evidence — when the scene does not match the rule, the platform attaches the evidence and routes it to the workflow owner.
• Human decision — a person approves, dismisses or escalates. Wyse Envision is L2 decision support: consequential actions are approved by humans.
• Management record — the check, the evidence and the decision become a report that survives the shift change and the audit.

Typical first scenarios for a plant

Scenario	Plain-language rule	Who owns the exception
Work presence / duration	“Station 4 attended during operating hours”	Production supervisor
Restricted zone	“No presence in the chemical store outside the maintenance window”	EHS
Blocked exits	“Exit B clear at every check”	Facilities / EHS
Shift condition & housekeeping	“Line 2 tidy at handover”	Shift lead
SOP verification	“Lock-out step visible before machine access”	Production / quality, with human review

Each of these is one camera-visible condition with a measurable manual baseline. None of them requires the plant to replace its estate, and none of them is sold as “AI safety for the whole factory”. One validated scenario first; the rest follow the evidence.

What we deliberately do not claim

No generic accuracy figure. Accuracy depends on the camera, the scene and the rule, so it is measured during your pilot against test cases both parties agree in advance. No biometric processing in the standard offer. No promise that every camera works automatically — suitable RTSP/IP streams are confirmed in the readiness study. And no autonomous action: people approve consequential decisions.