Jobsite Camera Placement and AI Retrofit: Watching Footage That Was Never Watched

1. Why placement drifts on active jobsites

Camera placement on a construction site is a moving target. The site logistics plan locks the original positions to the lay-down area, the gate, and the trailer location on day one. Three things tend to break that plan within a few weeks:

• Phasing. The lay-down area moves once foundations cure. Tower cranes get erected and re-positioned. The original sight lines stop matching the highest-value zones.
• Trailer relocation. The job trailer where the DVR lives moves at least once on most projects. PoE runs get extended, sometimes spliced, and the resulting cable plant is often improvised.
• Trade-driven obstruction. Scaffold goes up, weather screen goes up, mock-ups get staged. Cameras that had clear lines of sight start staring at plywood within days.

By week three or four, several of the cameras are functionally useless even though the recorder shows green lights across the board.

2. The job trailer DVR reality

Walk into a construction trailer at any point during a typical day and the security system looks the same. A 9 or 16 channel DVR humming away under a desk. A small monitor in the corner showing a multiview that nobody is watching. A keyboard caked in concrete dust. Footage retention set to about 14 to 30 days, after which the system overwrites itself.

The number of hours of footage being generated per day on a typical mid-size jobsite is staggering. The number of hours actually reviewed is roughly zero, except after an incident. The system is doing exactly what it was specified to do, which is record. It is doing none of the thing the project actually needs from it, which is detect.

3. Retrofit AI vs replace the system

When intelligence finally enters the conversation, the contractor usually gets pitched a full system replacement. New IP cameras with built-in analytics, a new VMS server, sometimes a managed service contract. Capex routinely lands in the $80k to $150k range for a mid-size project, and the install timeline is measured in weeks.

The retrofit path runs on a different cost curve. The hardware sitting in the trailer keeps recording exactly the way it always did. A small device hangs off the DVR's HDMI output, processes whatever multiview is being displayed on the trailer monitor, and runs detection on each tile. Install takes minutes. The capital outlay is a fraction of replacement, and the recurring cost slots into the security line item rather than going through capital approval.

The trade-off is that retrofit constraints the AI to whatever the DVR can already see. If a camera was poorly placed, AI will not fix that. The flip side is that bad placement is a 15 minute fix on the camera bracket, not a six week replacement project.

4. Flagging anomalies in real time

Once detection is running, the alert design is what determines whether the system actually changes outcomes. The pattern we see work on jobsites is:

• Off-hours-first detection. Most jobsite events that matter happen between 8pm and 5am. Schedule rules suppress alerts during normal crew hours and enable them after.
• Lay-down zone priority. Material lay-down areas are where most loss originates. Zone-level rules prioritize alerts here ahead of perimeter.
• WhatsApp delivery. Alerts go to the existing site WhatsApp group, not a new app. The super and the foreman are already there 18 hours a day.
• Evidence link in the alert. One tap to open the clip with the bounding box already drawn.

The result is a system that catches the events the team cares about and ignores normal crew activity, instead of the inverse.

5. What contractors are actually asking for

Adoption today skews toward retrofit on existing infrastructure rather than full replacement. The reasons line up consistently across general contractors:

• Capex is already spent. The cameras and DVR are already on the books. Replacing them feels like writing the same check twice.
• Procurement cycles are long. Requesting capital approval on a new VMS adds weeks of process. Retrofit lives inside an existing line item.
• Pilot risk is low. A retrofit on one site can be tested for a couple of weeks. A full replacement is hard to undo.
• Most pain is detection, not capture. The cameras already see the events. The miss is the detection layer.

Greenfield projects with new owners and a real capex authority sometimes still go full replacement, especially on flagship jobs. Everywhere else, retrofit is winning.

6. Getting started without ripping anything out

The cleanest first step is a two week trial on the highest-loss site you have:

• Plug into the DVR HDMI port. No camera changes, no network changes.
• Configure zones for the lay-down area, perimeter, and gates. Set off-hours schedules for the first round.
• Route alerts to the existing site WhatsApp group.
• Track everything for two weeks. Real events caught, false positives, and time to acknowledge.

The two week window is usually enough to see the value clearly. After that the conversation shifts from whether retrofit is worth it to which other sites get it next.