Construction site AI: preventing the RIDDOR report before it happens

The active measures question on a construction site

HSE investigations on construction sites do not ask whether the rules were written down. They ask what active measures were in place. The signs were up. The toolbox talk happened. Permit to work was signed off. And the worker still entered the exclusion zone around the operating crane. The question the investigator asks is: what stopped that happening in real time?

For most construction sites in 2026, the honest answer is nothing. Cameras record. The footage gets reviewed after the incident has already happened. The RIDDOR report goes in. The HSE investigation starts. The conversation moves to what changes for next time.

AI behaviour detection changes the answer. The cameras are not just recording. They are flagging the moment the exclusion zone is breached, the moment the PPE is missing, the moment the fire watch is unattended. Active measures, in real time, on existing cameras.

The five highest-value detection categories on a construction site

1. Exclusion zone breach

Personnel entering ATEX zones, plant areas, crane swing zones, or live electrical zones beyond permit boundaries. This is the single most common HSE prosecutable failure on a UK construction site.

2. PPE non-compliance

Missing hard hat, hi-vis, eye protection, or flame-retardant clothing on entry to controlled zones. The average HSE fine for PPE non-compliance is significant and rising.

3. Hot works fire watch

Smoke and early-stage fire detection during permit-to-work windows. The cleaner detection is during the watch window itself, but auto-escalation if the fire watch staff member leaves post is the higher-value layer.

4. Person down and accident detection

Slips, trips, falls from height, person-down events. The response speed for medical intervention is the difference between a near-miss and a fatality.

5. After-hours access

Personnel on site outside shift windows. The perimeter approach and access anomaly is typically the first sign of theft, vandalism, or unauthorised entry that becomes a Saturday morning RIDDOR report.

What this looks like in practice

A 50-camera UK construction site running AI behaviour detection typically generates 3 to 5 actionable alerts a day during active works. Each alert routes to a site manager or H&S lead with a clip and location. Response time is measured in seconds rather than the hours it takes to find the same incident in recorded footage after the event.

Over a 12-month period, we typically see a meaningful reduction in RIDDOR-reportable incidents on sites running active monitoring versus comparable sites running reactive CCTV alone. The insurance and HSE positions improve accordingly.

What it does not do

AI behaviour detection does not replace the site manager, the toolbox talk, the permit-to-work process, or the H&S culture. It is a layer that catches what those systems miss. The cleanest deployments are sites where the existing safety culture is strong and the AI layer is adding a real-time enforcement capability rather than substituting for missing procedures.

The HSE conversation

HSE investigators are increasingly asking what active measures are in place. Documented active monitoring with a timestamped evidence log changes that conversation. The investigator can see, on the day of the inspection, exactly what the system detected, when, and how the site responded.

For sites in scope of CDM and the wider construction regulatory framework, this is increasingly the differentiator between a site that the HSE views as well-managed and a site that the HSE views as exposed.

The starting point

If you want to see what active monitoring looks like on your specific site, book a discovery call. We walk through the camera layout, the highest-risk zones, and what deployment would actually look like in two weeks rather than two months.