Your cameras already know more than who came through the gate
Modern CCTV analytics produce a stream of operational data as a by-product of watching a site. Most businesses throw all of it away, and the ones that do not have to get the privacy question right first.
CCTV gets bought as a deterrent and reviewed as evidence. Something goes missing, someone scrubs back through the footage, and the system has done its job.
That is a camera used as a filing cabinet. The interesting thing about a modern system is that while it is watching, it is also producing structured data, and that data has almost nothing to do with security.
What the analytics actually emit
Ask what a current camera or video management system can output, rather than what it displays, and the list is longer than most people expect. Object classification, so a person, a vehicle and a forklift are distinguished rather than lumped together as motion. Counts across a line. Direction of travel. Dwell time in a region. Occupancy. Queue length. Number plates where the camera is positioned for it.
None of that is footage. It is a small, cheap, structured record: a class, a timestamp, a zone, a count. It is the kind of thing a database is good at, and it is generated whether or not anyone ever looks at it.
The operational questions it answers
Once you treat that as telemetry rather than a security feature, the uses stop being about theft.
- How long does a truck actually spend on site? Arrival at the gate, time at the weighbridge, time at the dock, departure. Not what the schedule says, what happened.
- Where is the queue forming? Vehicles waiting at a weighbridge, staff waiting at a time clock, product banking up between two stations on a line.
- Which docks and bays get used, and when? Occupancy over a season tells you whether the constraint is space or scheduling before you spend money on concrete.
- Is the traffic pattern on the floor what you designed? Forklift movements rarely match the layout drawing after a year of small adjustments.
- Is PPE being worn in the areas that require it? As a trend to manage, not a disciplinary tool.
Every one of those is an operations or capital question, and the sensor is already installed and already paid for.
Why nearly all of it is discarded
The data sits inside the video management system. The people who would use it work in operations and look at the ERP, a spreadsheet, or a dashboard. The two are never connected, so the metadata ages out with the footage on a thirty day retention cycle and nobody notices, because nobody knew it existed.
The fix is unglamorous integration: pull the event stream out, land it somewhere that keeps it cheaply, and put it next to the production and dispatch data it needs to be compared with. A count of trucks is mildly interesting. A count of trucks alongside tonnage received and labour hours on shift is a management report.
The other half of the fix is deciding what question you are answering before you start. Camera analytics will happily generate an enormous volume of events nobody has asked a question of, and that is how these projects turn into a dashboard nobody opens.
Now the part that has to come first
This is data about people, and in New Zealand that puts it squarely inside the Privacy Act 2020. Being genuinely useful and being careful are not in tension here, but the order matters: decide the privacy position before you turn the analytics on, not after someone asks.
A few things we would hold to.
Count, do not identify. Most of the operational value above needs a class and a timestamp, not a person. “Fourteen people entered this zone” answers the question. Knowing which fourteen usually does not, and collecting it anyway means you are now holding personal information you had no need for.
Facial recognition and other biometrics are a different conversation entirely. They are not a feature you enable because the camera supports it. If someone is proposing it, the purpose needs to be specific, the necessity needs to be arguable, and the decision belongs with the business rather than with whoever is configuring the system.
Purpose limitation is not paperwork. Cameras installed for site safety and security, then quietly repurposed to monitor individual staff productivity, is the failure mode that ends up in a complaint and deservedly so. If a new use is worth having, it is worth telling people about.
Be honest in the signage and honest with staff. People are far more relaxed about a system that measures how long trucks wait than about one they suspect is watching them, and the difference is almost entirely whether it was explained.
Keep the metadata longer than the footage, on purpose. Video is heavy and sensitive. An anonymous count is neither. Retaining a year of counts while retaining thirty days of video is both cheaper and more defensible than the reverse.
Where this lands
If you already have cameras across a site, you have an operational sensor network that was sold to you as insurance. Getting more out of it is mostly an integration job and a set of decisions about what you are willing to collect.
Both halves matter. The value is real, and so is the obligation attached to it.