In this industry, labour is the cost structure. When the minimum wage rises year after year and insurance premiums climb as well, simply hiring more people is no longer a road you can stay on indefinitely.
Two pressures arriving at once
The SET Finance report states the problem plainly: employers are facing a staffing shortage and rising wages at the same time. The first leaves rosters unfilled; the second raises the cost of every shift that is filled. Security feels the combination especially sharply, because what we sell is people and hours.
Ai Hsiang-Wen, Associate Vice President of WorldTrend Security, described the position in his interview:
On the shortage itself, the report also cited Yang Tsung-Pin, a PR deputy manager at a job site, who observed that after the minimum wage adjustment some people would rather take two hourly side jobs than a full-time office role. In other words, what is missing is not only money — the structure of labour supply itself has changed.
Not cutting people — letting the machine take the first look
The report then goes inside WorldTrend’s monitoring centre. The reporter’s voice-over sets out the core of the approach: folding AI inspection into the cameras already installed, so that businesses and residential communities still have someone — or rather, something — watching, without adding headcount.
The key word is “already”. The approach does not ask a community or a business to replace its cameras: existing cameras, existing cabling and existing recorders all stay, and the AI is a layer of interpretation added on top. That is why the cost can be held down.
On a set schedule the system works through the cameras one by one, producing a patrol record for each checkpoint with its state at that moment.
Readings that come back normal disturb nobody; only exceptions are pushed to the monitoring centre and the management office, so operators are not buried under alerts that mean nothing.
An operator in the 24-hour monitoring centre reviews the footage, confirms the site, and then decides whether to notify, dispatch or call the authorities. The last mile is still human.
The camera also caught the system in use: a screen of patrol records, each tagged with the site name, the checkpoint (the ramp shutter, the No. 21 gate) and the camera ID, followed by the reading. The voice-over explained what the list means — when every line reads normal, the site has been normal all day.
Another shot showed an AI reading on a phone: the system described a construction site, a blue truck parked in the centre, scaffolding and structural work visible behind it, and noted that no personnel were present in the frame. That is the difference between today’s AI video reading and the old question of whether there is a person in the picture — what comes back is a description of the site that can be read, kept as evidence and traced later.
“Unmanned” is not the answer — cutting unproductive hours is
The second half of the report turns to retail and food service — self-checkout terminals, serving robots, automated ordering — where companies use automation to cut labour cost. Security is different: we cannot go unmanned. An alarm needs someone to judge it, a site needs someone to reach it, and liability needs someone to carry it.
So our approach is not to replace people but to move them out of repetitive, low-yield work. An operator used to stare at dozens of tiled feeds in which, almost all of the time, nothing happened. That stretch of time now goes to the AI, while the operator concentrates on the events that genuinely call for judgement and response.
For clients the meaning is concrete: a community or a business no longer has to carry a full on-site shift indefinitely just to have eyes on the place overnight, yet the place is still watched overnight. For our own colleagues, the meaning is that the roster can actually be filled — which, in a shortage, matters just as much.
If you are evaluating this
- Check whether your existing kit will do — camera positions, angles and image quality determine whether any further investment is needed. In most cases what is already installed is enough.
- Ask who picks up an exception — once the AI flags something, where does it go, who is on duty, and between what hours is someone there? That is what separates a security service from a software licence.
- Make sure the records are retrievable — the time, checkpoint and reading of every inspection should be available afterwards, which helps both management committees and companies on questions of responsibility.