Nearly every site security brief starts the same way. Someone asks how many cameras the budget covers. It is the wrong opening question, and it sets the cost of every later mistake.
Camera count is an output of the design. Ask for it first and you buy hardware, then bend the site around it.
Start with the risk register, not the site plan
Before a single camera position gets marked, write down what the site loses and how. Break-ins through the eastern boundary. Fuel drawn off after hours. Contractor vehicles leaving heavier than they arrived. Each entry names an asset, a route and a time window.
Then rank them by value at risk. The ranked list decides where the money goes. Without it, coverage gets spread evenly across a site where the risk is not spread evenly at all.
Even coverage on uneven risk is the most common way to overspend and still get robbed.
Scene design beats spec sheets
A camera does one job well and every other job badly. Decide the job per position before choosing the lens.
- Detect. Something is moving in the area. Wide field, low pixel density
- Recognise. This is a person, in a uniform, carrying an object
- Identify. This is a specific face or a specific plate, good enough for evidence
Identify positions need pixel density on target, controlled light and a tight field. Detect positions need coverage and analytics. Mixing the two produces a site full of cameras recording motion nobody uses in an investigation.
Light is the part budgets forget. A high-specification camera in bad light performs worse than a modest camera in good light. Walk the site after dark before signing the design.
Analytics belong in the design, not the demo
AI video analytics work well when the scene suits them. Line crossing on a clean perimeter is reliable. Behaviour analysis in a crowded yard with moving plant is not, at least not without tuning and patience.
Write the analytic rule into the design next to the camera position, with the expected false alarm rate and who responds. A rule with no response plan generates noise, and noise trains a control room to ignore alerts.
Network and power decide uptime
Most security system failures are not camera failures. They are switch failures, fibre breaks, power events and undersized backup.
- Size the backup for the real outage length, not the brochure figure
- Separate the security network from the corporate one, or segment it properly
- Plan fibre routes away from the trenches contractors dig every year
- Specify storage on retention days at the recording rate you plan to run
- Put monitoring on the infrastructure itself, so a dead camera raises an alarm
A site finding out about a failed recorder during an investigation has no system. It has a budget line.
Write a specification suppliers answer to
A good specification states outcomes and constraints, then leaves brand choice open where it makes sense. It fixes pixel density per position, retention days, response times, integration points and commissioning tests. It stays silent on the model number unless the standard demands one.
Suppliers price the same scope. Comparisons become real. And the site keeps the option of a second supplier three years later.
Commission against the design, not the invoice
Acceptance testing walks each position and confirms the job it was designed for. Identify positions get tested with a face and a plate at the design distance, at night. Analytics get tested with a real crossing, a real loiter, a real vehicle.
Anything failing the test gets fixed before sign-off. After sign-off, it becomes a maintenance argument nobody wins.
The order matters
Risk register, scene design, infrastructure, specification, commissioning. Camera count falls out of step two. Handled in the right order, a site spends less and detects more. Handled backwards, it spends the budget twice.