The problems in a shop
Shrinkage shows up as a number at stocktake, weeks later. Something walked out the front, a refund was processed with no customer there, or a stockroom carton was opened and never counted. The till is where cash and goods meet, and where staff are most exposed.
Then there is overnight. A door forced at 2am is found at 8am, and the footage sits on a recorder nobody has opened since install. You also want plain answers: how many people came in on Saturday.
What we install
We walk the shop first, including the back door and stockroom — usually thinner on cover than the floor.
- Counter and till. A camera over the counter that sees hands, goods and customer, plus one behind it facing the queue.
- Floor and entry. An entry camera at head height for usable faces, with wider cover down the aisles and over high-value displays.
- Back of house. Stockroom, rear door, bin area and staff entrance — where stock leaves without a sale.
- Recorder, storage and network. We size the NVR and drives to your camera count and retention, cable and mount, and set up remote access so you can check the shop from home or from the car. Live viewing uses a local restream, so extra viewers do not load the recorder. Installation sets out each step.
What the AI layer watches for
Footage stays on the recorder in the shop; analysis runs on our own server, not a third-party cloud.
- After-hours intrusion alerts. A scheduled sweep looks at the cameras. If a person appears outside trading hours, or on a day you have marked closed, the people you nominate get an email with the camera frame attached. A cooldown stops repeat emails for the same event.
- Door monitoring and line crossing. Lines across the entry, rear door or stockroom doorway, so door activity is logged as a searchable event.
- Customer counting. People counts per camera over time, to compare hours and days against sales.
- Stock movement checks. Goods-movement events can be cross-checked against your point-of-sale records, so a movement with no matching sale or inventory deduction is flagged for review.
The system flags and alerts. It does not recognise faces, identify individuals or prevent theft, and it will not catch everything. It cuts hours of vision down to the few events a person should look at. AI analytics lists each capability.
What you see day to day
When the stocktake is short, you search the activity timeline for the stockroom camera across the hours that matter. Each event carries a clip of roughly ten seconds either side — usually enough to see whether a carton went to the floor or out the back door.
With the live view open on the office screen, boxes are drawn around people and vehicles on the stream the browser already has. How it works follows one event from camera to inbox.
Practical constraints
- Glass and backlight. A shopfront camera facing the street fights daylight, so position matters more than specifications.
- Signage and staff notice. Surveillance and workplace surveillance laws differ by state, audio is treated more strictly than video, and cameras must not cover toilets or change rooms. Tell staff and put up signs. This is general information, not legal advice — check your state's rules.
- Processing budgets. Per-camera and hourly analysis budgets keep costs predictable, so you decide which cameras get the attention. The till and stockroom usually earn it; the back fence rarely does.
- Staff safety is not a duress alarm. Alerts tell you after the fact, and are no replacement for a panic button or a second person on a late shift.
- Licensing. In several states installing security equipment is a licensed activity. Ask any installer for their licence and insurance details, including ours.