AI-Powered Risk Intelligence: Smarter Insights, Safer Workplaces

It's your friend. It's your helper, but [AI] is not your superior.

- Liz Hancock

About this session

A tender document that would normally take three days took three hours. A CCTV system flagged an ergonomic risk safety teams had missed entirely, and incidents dropped by sixty five percent once it was fixed. AI isn't replacing the safety professional's job. It's clearing the tedious eighty percent of it so there's more time left for the twenty percent that actually needs a person.

In this session of the Safety Voices webinar series, host Craig Bleakley sits down with Liz Hancock to map out where AI genuinely earns its place in a health and safety function, where it quietly falls short, and why the line between the two isn't about the technology at all. It's about what still needs a human on the shop floor.

Where it's safe to use AI in safety

Liz is direct about where the technology earns its keep.

• Policy/program writing — turning a multi-day drafting job into a few hours of review and personalisation.
• Hazard identification — spotting risks in a photo or video of a workspace, then suggesting controls to design them out.
• Drafting tenders — pulling together detailed, industry-specific documentation from a clear brief.
• Cutting red tape — helping operational teams own their own hazard identification and controls, rather than routing everything through a safety advisor.
• Where appropriate — the caveat that runs under everything else on the list.

The seventy percent rule

Liz returns to one number throughout the session: AI will get you seventy, maybe eighty percent of the way there. Never further.

It's fast at hazard identification, drafting policies, writing tenders, generating board reports, and spotting trends across years of incident data that a person would never have time to cross-reference by hand. But it also confidently flagged a shiny floor as a spill and a bookshelf as an unauthorised ladder in the same afternoon. It doesn't install the guard rail, mount the signage, or have the honest conversation with a colleague about why they're not following a safe system of work. It's not a source of truth. It's a first draft.

The practical takeaway isn't caution for its own sake. It's knowing exactly which seventy percent to hand over, and which thirty percent stays firmly with a person.

Key takeaways from the session:

1. The best use cases are the ones nobody notices

Drafting policies, writing tenders, building board reports, reviewing existing procedures for what's out of date. None of it is glamorous, but it's the work that used to eat entire days, and giving it back is what actually frees up a safety professional's time for the floor.

2. Consultation is what turns AI from surveillance into trust

Rolling out AI-powered CCTV without involving worker reps, unions or legal first reads as spying. Rolling it out as a project, with blurred faces, a clear "this isn't for discipline" commitment and worker input from day one, turns the same technology into something the workforce actively champions.

3. Never let AI touch the moment itself

Emergency response planning is a strong use case. Responding to an emergency in progress is not. The plan needs to already exist and already be known before anything happens; that's not a moment to start prompting a chat function.

4. Gut still wins

When AI-generated insight conflicts with frontline experience, Liz goes with the person on the shop floor every time. No output replaces the conversation of actually asking someone whether the safe system of work is being followed, or why not.