Skip to content

Retail · in production

RetailVision in retail

The store’s existing CCTV now flags suspicious activity in real time, with video evidence.

6

existing cameras

641k+

AI detections logged

980

alerts ranked for review

Before

  • One person can’t watch 10–15 camera feeds at once.
  • Theft was found weeks later, in stock counts.
  • Rule-based CCTV analytics caused alert fatigue.

After

  • Running live, around the clock, since September 2026.
  • Staff review seconds of video, not hours.
  • Runs on one small GPU PC in the store.

How it works

1

Existing CCTV

RTSP / NVR

2

Detect & track

People, products, bags

3

Understand actions

Pose + behaviour models

4

Safety rules

Checkout zone, cooldowns

5

Alert + clip

Staff review & decide

Cross-camera tracking

Follows each shopper with an anonymous, temporary ID.

Behaviour recognition

Spots pick-up, concealment and hand-offs.

Evidence clips

Every alert comes with a short pre/post video.

Privacy by design

No facial recognition. Video stays in the store.

Have a similar problem?

30 minutes with an engineer. You'll know if it's doable, and roughly what it costs.

More case studies

US market intelligence

AI agents that map every home-utility business in the US, city by city.

KYC & liveness verification

Automated ID checks, liveness detection and face matching, deployed in production.