I've sat through a lot of vendor pitches this year, and they've started to blur together. A camera watches the floor. A dashboard turns red when something looks risky. A chart in the appendix claims a percentage reduction in incidents at some unnamed reference site. It's a good pitch. It's also, in my experience, not the part that determines whether the tool actually helps anyone.
This is my own take, not a summary of someone else's reporting — I want to be upfront about that distinction, because it's the whole point of this section of the site. The automated feed elsewhere on HSE Hub tells you what happened. This is where I tell you what I think it means.
The pitch is never the problem
Every vision-AI safety vendor I've spoken with can demonstrate detection working. That's not hard anymore — the underlying computer vision has genuinely gotten good. What almost none of them can show me convincingly is what happens in month four, on the night shift, in the aisle with bad lighting and a temporary storage overflow that wasn't in the training data.
That's not a knock on the technology. It's a knock on how it gets evaluated before purchase. Nobody buys a fire suppression system based on a demo in perfect conditions; everyone seems willing to buy a detection system that way.
What I'd actually ask before signing
If I were sitting across from a vendor today, here's where I'd spend the meeting:
- "Show me your false-positive rate after 90 days in a live facility, not a pilot." Pilots are curated. Production isn't. If they can't answer this with a number, that's the answer.
- "What happens to alerts at 2 a.m. when the safety office is closed?" A detection system with no accountable human response loop isn't a safety system — it's a very expensive log file.
- "Who retrains the model when our floor plan changes?" Warehouses aren't static. If retraining requires a new contract negotiation, the system will quietly go stale.
"A tool that detects a hazard and tells no one who can act on it in time hasn't reduced risk. It's just moved the paperwork earlier."
Where I actually land on this
I'm not against the technology — some of it is genuinely useful, particularly for proximity detection in high-traffic zones where human attention naturally lapses. What I'm against is treating a dashboard as a substitute for a supervisor who knows the floor. The best deployments I've seen treat these tools as an additional set of eyes reporting to a person, not as the person.
If your organization is evaluating one of these systems right now, my honest suggestion: spend less time in the vendor demo and more time asking what happens on the worst version of an ordinary Tuesday. That's the question the sales deck never answers — and it's the only one that matters once the contract is signed.