Most retail CCTV answers one question after the fact: what happened? On a busy cosmetics, liquor, or fashion aisle, the question that protects margin is different: can we act while they are still here?
Real-time shoplifting detection is the shift from playback to floor intervention - cameras you already own, analysis on site, staff alerted before the exit.
> Quick answer: On-premises AI watches for concealment on your camera feeds and typically alerts staff in about two seconds - with who and where - so they can respond before exit.
At a glance
Concealment is quick. On a crowded Saturday, the person who slipped fragrance or a fashion accessory into a bag can be at the door before anyone finishes reviewing a clip.
Staff cannot watch every feed. Locked cabinets depress sales. Guards help but are expensive for every independent site. What moves shrink is intervention before exit - which only works if the alert arrives while the person is still in the aisle.
National context: cost of retail shrinkage in Australia. Crew tactics: organised retail crime.
Skip the jargon soup. Buyers need a path they can follow:
| Stage | What happens | Why it matters |
|---|---|---|
| 1. Camera | Existing CCTV frames enter the on-site appliance | No rip-and-replace of cameras for a pilot |
| 2. On-site AI | Concealment behaviour flagged on the local appliance | Typical alert path ~2 seconds end-to-end |
| 3. Staff alert | Phone / display with face crop and aisle context | Staff know who and where - not a vague motion beep |
| 4. Floor response | Approach per store SOP before exit | Prevention beats Monday playback |
What you should ask a vendor about that "on-site AI" step is practical, not theoretical: alert path latency, false-positive trend after week two, and whether video leaves the building for inference.
Camera placement still limits everything. Clear views of high-value bays help more than a single overhead dome that never sees the shelf. A site assessment maps coverage before go-live.
Theft behaviours on retail floors, in plain English:
Context matters. Putting a phone in a bag should not page the floor. Sliding a perfume bottle from shelf to bag should. Cosmetics, liquor, and fashion aisles all share the same need for that distinction.
Pharmacy-deep SERP piece: AI theft detection for Australian pharmacies.
If staff stop trusting the phone, ORC walks anyway. Systems earn week-two trust when noisy alerts drop and the floor still picks up the phone. Read what is a false positive and why it matters before you score vendors on a perfect demo day.
Analysis should run on a compact appliance on your LAN:
Cloud round-trips of 5 to 15 seconds often miss the intervention window. Full comparison: edge AI vs cloud CCTV. Privacy depth: on-premises AI for retail CCTV.
Managed services can improve accuracy over time without sending live video off-site for routine inference.
IntelliGuard serves Australian retail and pharmacy sites: existing cameras, on-site appliance, staff alerts with who-to-approach context and aisle location, no cloud upload for inference. Pricing $299 to $369 per store per month (excl. GST), edge appliance included, 30-day money-back guarantee.
Real-time detection is not "smarter CCTV." It is a short path from camera to staff while intervention is still possible. Architecture (edge vs cloud) and alert quality (false positives) decide whether the system survives the first month on the floor.