Every warehouse owner who calls us asks the same first question: which camera should I buy?

It's the wrong question, and answering it honestly usually saves them money. In almost every warehouse we've deployed on, the cameras were already fine. What was missing was anything watching them.

Start with what you're actually losing

Warehouse loss is rarely a midnight break-in. It's the loading bay during a shift change. It's a pallet that leaves on the wrong vehicle. It's small, repeated removals by someone with a legitimate reason to be standing there.

None of those are solved by a sharper picture. They're solved by something noticing the behaviour while it happens.

So before you look at a single spec sheet, work out where your stock actually goes. In our experience the answer clusters in four places:

  • The loading bay and dock, during shift changes and outside dispatch hours
  • High-value SKU aisles, during and after picking hours
  • The staff gate, where headcount and attendance stop matching reality
  • The vehicle gate, where the register is a notebook nobody reads back

What an AI CCTV camera for warehouses actually needs

Once you know the zones, the hardware question gets much smaller.

Coverage beats resolution. A 2MP camera pointed correctly at a dock will do more for you than a 4K camera mounted high in a corner, seeing everything and identifying nothing. Every extra metre of height costs you the angle you need to recognise a face.

Face height at the gate. If you want attendance or identity, one camera has to be at roughly face height, facing arrivals. This is the single most common thing missing from warehouse camera estates, and it's usually one camera, not forty.

Usable infrared. Most intrusions happen at night. A camera whose IR washes out at eight metres is a camera that records a white blur. Check this on your existing cameras before buying anything new.

An RTSP stream. This is the boring one that decides whether software can help you at all. If your DVR or NVR exposes RTSP — and nearly everything installed in the last decade does — then AI can run on the cameras you already own.

The part the spec sheet doesn't cover

Here's what changes a warehouse: the footage becoming searchable.

Right now, investigating a discrepancy means scrubbing timelines across dozens of channels. It takes days, which in practice means it doesn't happen — the loss gets absorbed and written off.

With detection running, you filter instead: this camera, this zone, people only, between these hours. A three-day review becomes a three-minute one. That single change is what determines whether your camera estate is a deterrent or an archive.

Dwell time is the signal that matters most

If we could give a warehouse manager exactly one alert type, it would be loitering.

A person standing near high-value stock for noticeably longer than picking requires is the most reliable pre-theft indicator there is. It's also completely invisible to motion detection, which reacts to movement — and a person assessing a shelf is, by definition, not moving much.

What we'd actually recommend

Audit before you buy. Take your camera layout, and for each of the four zones above ask: can a camera see it, at a usable angle, with usable light? Usually the answer is yes for three of them and no for the gate.

Then add intelligence to what's there. Adding AI to 60 existing cameras costs less than replacing six of them, and it addresses the actual failure — that nobody is watching — rather than the imagined one.

If your cameras genuinely can't do the job, we'll tell you that too. It's a shorter conversation than most vendors want to have, but it's the one worth having.

Related reading: AI CCTV for warehouses · Loitering detection · Nobody can watch 120 camera feeds