A general contractor we spoke with last quarter had a fairly typical 24-month security history. Year one: standard chain-link fence, motion lights, an alarm-monitored job trailer. Direct theft loss: $310,000. Year two: added two overnight guards at the largest site and upgraded the trailer cameras to 4K. Direct theft loss: $340,000. The guards cost $156,000. Net change: worse.
This is the pattern. More spend, more loss. The reason is that most jobsite security programs are designed to record theft rather than prevent it. The four numbers below are the diagnostic.
Number one: perimeter approach rate
How many times per night does a person cross your perimeter? Not how many thefts — how many approaches. A perimeter that gets approached 6 times a night will eventually produce theft, regardless of the camera count. A perimeter that gets approached 0.8 times a night almost never produces theft.
Most GCs do not know this number. The fixed-camera systems they run record everything but surface nothing. The alarm-monitored trailer triggers only on door contact, which means a fence cut at the back of the lot is invisible. You cannot manage what you do not measure.
On the Mid-Atlantic GC deployment we wrote up as a case study, the pre-deployment approach rate at the largest of three sites was estimated by the night-shift superintendent at 4 to 6 per night. Thirty days after deployment with active SOC monitoring, the measured rate was 0.8.
Number two: response time to a verified threat
This is the number where most programs collapse. The chain looks like this: a sensor or AI fires, somebody reviews it, somebody decides to escalate, somebody calls LE, LE responds, LE arrives. If you add up the realistic numbers for a typical fixed-camera-plus-third-party-monitoring stack, you get something like 12 to 18 minutes from breach to LE arrival. The theft has been over for 10 minutes.
The number that matters is the time from intrusion to direct contact with the intruder. Audio talk-down compresses that to moments. The intruder hears a live operator, by name, before the saw comes out. On the Mid-Atlantic GC, the SOC issued 187 audio talk-downs across the year. Eleven of every fourteen subjects departed immediately on first contact.
Rapid response is not an aspiration. It is a measurable property of the system you bought. Ask your current vendor what their median time from event to direct intruder contact is. If they cannot answer, that is your answer.
Number three: false alarm rate
A program that fires 20 times a night and turns out to be 19 false alarms trains everyone — operators, customers, LE — to ignore it. By month three the alarm is functionally off.
The industry baseline on pure-AI systems running unsupervised is in the 30 to 60 percent false-alarm range depending on conditions. Wildlife, weather, construction equipment, contractors arriving early, security lighting cycling. The result is alert fatigue and a slow drift toward operational irrelevance.
Human verification — an actual SOC operator reviewing the AI-flagged event before any escalation — drives false alarms below 5 percent. On our platform we measure it at 97 percent reduction. The customer does not get notified for events the operator clears. The customer does not get a 3am phone call about a raccoon. By month six, every alert the customer sees is real.
Number four: per-incident loss
This is the number every CFO asks about and almost no one tracks correctly. Direct theft loss is the easy part — what was stolen, what did it cost to replace. The full per-incident number includes:
- Replacement cost (parts, equipment, materials)
- Labor to install replacement
- Project delay cost (per-day overhead, liquidated damages risk)
- Insurance premium impact (loss-ratio-driven)
- LE callout fees in jurisdictions that bill them
- Internal admin time on police reports, claims, and superintendent rework
On a typical commercial construction site, the loaded per-incident cost is 2.5 to 4 times the direct-theft figure. A $12,000 copper pull is really a $32,000 event after the project-delay tail is honest.
When you run the four numbers together — approach rate, response time, false alarm rate, loaded per-incident loss — the math on jobsite security shifts. The Mid-Atlantic GC spent under $90,000 over twelve months on the surveillance and SOC monitoring. The avoided loss was $340,800. The insurance carrier renewed at the prior rate.
What to do tomorrow
If your current program does not produce these four numbers on a monthly report, your program is operating on faith. Three actions:
- Ask your current vendor for the monthly approach rate at each site. If they do not have it, the perimeter is not actively monitored.
- Ask for the median event-to-intruder-contact time. If it is over 90 seconds, you have a recording system, not a prevention system.
- Run a real loaded per-incident loss number for the prior twelve months. Most GCs are surprised by what falls out.
If those three questions produ