Construction site theft is a downtime problem, not a materials problem

1. The downtime math nobody runs

A roll of $4,000 worth of copper wire walks off a site. The site superintendent files a police report, calls insurance, calls the supplier, reschedules the electrical rough-in. Three crews stand around for two days while replacement material gets sourced. The general contractor eats per-diem for a project manager and a site super at $1,200 a day combined. The owner pushes the substantial completion date by a week.

The stolen copper is $4,000. The downtime is $50,000 to $150,000 depending on project size, contract penalties, and whether the delay slips a permit window. On commercial work with liquidated damages clauses, a single one week slip can be six figures by itself.

This is why every conversation about jobsite security that starts with 'is it worth it for a few thousand dollars of materials' is the wrong frame. The materials are the loss leader. The schedule is the actual asset under attack.

2. Why most existing site cameras do not help

Most active jobsites already have cameras. There is a battery powered solar trailer in one corner, two or three IP cameras zip tied to a fence, and a DVR humming inside the construction office trailer. The system records 24/7 to a hard drive that nobody touches.

When something goes wrong, the site super spends two hours scrubbing through footage to find a 30 second clip. Often the angle is wrong, the lighting is wrong, or the suspect already left the frame. By the time anything is identified, the site has been compromised for 12 to 48 hours.

The cameras work. The recording works. The intelligence layer between the recording and a real time decision is the part that does not exist on most sites.

3. Retrofit vs greenfield, and why retrofit usually wins on a live site

The pitch from new system vendors is to rip everything out and install a fresh AI camera system. On an active build, this is rarely practical. The cabling is run, the trailer DVR is wired, the PoE switch in the office is loaded, and the crew has zero appetite for a security install crew on site for two weeks.

The retrofit alternative is a small edge AI device that taps the existing DVR's HDMI multiview output, the same composite the on site monitor is already showing. The device runs object detection on every camera tile in real time, applies per camera context rules (this zone is restricted after 1900, the lay down yard is hands off after dark), and sends a clip to the on call phone the second the rule fires.

Hardware is in the $400 to $500 range one time, monthly software runs around $200 per site, install is one HDMI cable and a power outlet. Cyrano is one device shaped like this; there are a handful of others in adjacent shapes.

4. Real time intercept vs after the fact review

The first question to ask any site security vendor is whether the system can interrupt an incident or only document it. Most camera DVRs are documentation only. They tell you what happened after the truck has already left the site.

Real time intercept means the system fires when the trespass is in progress, not after. A clip pings on call ops within 5 to 15 seconds. On call ops calls dispatch, the local non-emergency line, or a roving guard service that has the site keys. Most theft attempts on construction sites are exploratory: they walk away the moment a flashlight or a siren shows up.

Documented industry data on retrofit AI on existing camera feeds shows a 60 to 80 percent reduction in incident-to-loss conversion when the alert lands during the attempt rather than the morning after.

5. When greenfield is actually the right answer

Greenfield (rip and replace) is the right call when the existing system is genuinely broken: cameras failing, recordings missing, no light at night, no IR coverage. If the cameras are doing their job and the only missing layer is intelligence, retrofit is faster and cheaper by an order of magnitude.

A useful rule of thumb: count the existing cameras and check the DVR's recording. If you have 8 or more cameras and the DVR has continuous footage from the last 30 days, retrofit. If you have 3 cameras and the DVR has been broken since last winter, greenfield.

6. What to evaluate when choosing a retrofit AI overlay

Three concrete questions cut through the marketing on retrofit AI products. First, does it tap the DVR via HDMI multiview, or does it require RTSP streams from each camera? On a live jobsite RTSP is rarely accessible without re-cabling, and HDMI is always available on the back of any consumer or commercial DVR.

Second, does the AI run on device or in the cloud? On site means it works during the inevitable internet outage. Cloud means a wired ethernet drop and a monthly cellular bill or fiber pull.

Third, what is the rule grammar? You do not want to alert on every motion event in the lay down yard during the day. You want different rules per zone per time window. The richer the rule grammar, the fewer false alerts the on call gets at 3 in the morning.

7. Adoption pattern on active construction sites

The pattern that works on multi site GCs is to start with one site, the one with the highest theft history or the one closest to the project manager's office. Install on a Friday afternoon, run it through one weekend, and review the alert history on Monday. If the rules are tuned right, the on call should get one or two alerts the first weekend (and zero false positives by week two).

From there, a small contractor with 6 to 10 active sites typically rolls out across the portfolio over a quarter, $400 in hardware per site and ~$200 a month in software per site. Even at a single prevented break in event the math works for a year.