See every store at once. Without watching anyone.
Aproksha Retail Vision is computer vision for multi-store retail that surfaces shrink and unusual activity across every location, with no facial recognition. It reasons about behavior, not identities.
- 0
- faceprints stored
- 100%
- runs on-prem
- 1 GPU
- many stores
- 7-day
- auto-purge
The problem
Shrink is a fleet problem you can't watch your way out of.
A single store already has more camera-hours than anyone will ever review. Multiply that by every location and the footage stops being evidence — it becomes a place incidents go to hide.
The usual answer is face recognition, which trades your customers' privacy for a database you now have to defend. Under the DPDP Act that is a liability, not a feature.
The moments that matter — a back room entered after close, a till left open, the same pattern across three stores — are small, rare, and spread thin. You need something that watches all of it and only speaks up when it should.
How it works
One box per store, no surveillance creep.
01
Connect your cameras
Your existing CCTV feeds plug in over RTSP. There is no new hardware on the shop floor and no rip-and-replace.
02
Motion wakes the model
A motion gate sends only meaningful clips to the box, where activity and roles are inferred anonymously against the zones you drew.
03
You read summaries, not tape
Each store reports up in plain language, flagged by priority, with a snapshot and an 8-second clip — so a small team can watch a large fleet.
What it flags
Reliable signals, not flaky guesses.
We deliberately don't claim automated “theft detection” — one bad week of false alarms is an uninstall. ARV watches the zones you care about and hands a human the evidence.
After hours
any zoneAny person present while the store is meant to be closed. The single most reliable signal there is.
Restricted zone
restrictedA body enters a back room, stockroom, or office that should be empty. You get the clip, not a guess.
Dwell / loitering
any zoneSomeone lingers in a watched zone past a threshold you set — a shelf, a till, a high-value aisle.
Unattended counter
counterA till or counter is left with nobody at it for too long during open hours, when it should be staffed.
Why it is different
Loss prevention that respects the people in frame.
No facial recognition, ever
By design. The system reasons about behavior and roles, not identities. There are no faceprints, no biometric database, and nothing to leak.
Anonymous role inference
It tells staff, customers, and delivery apart without knowing who anyone is. Enough context to spot what matters, without the surveillance.
Plain-language summaries
Instead of hours of footage, you get readable summaries of what happened, where, and when, across every store, ranked by what needs a look.
Every store at once
One dashboard spans all locations. Patterns invisible store by store — coordinated shrink, repeat incidents — surface across the whole fleet.
Motion-gated by design
Cameras only wake the pipeline when something moves, so one modest box covers several cameras. Efficient enough to run at real scale.
On-prem or your cloud
It runs on a small box in your back office, or in your own cloud. Video never has to leave your premises — the right answer for sensitive footage.
Privacy by design
Built to be defensible, not just clever.
False accusations kill a retail relationship instantly. Human-in-the-loop isn't a limitation here — it's the product. Everything below is on by default.
Bodies, not faces
No face recognition and no biometric database. It reasons about behavior and roles, never identities.
Video stays in the store
All AI runs on a box inside the shop. Raw footage never leaves the premises — only clips and summaries reach you.
Evidence auto-purges
Clips are deleted after seven days unless you flag one, so nothing lingers longer than it needs to.
A human always decides
ARV hands you the moment and the clip; a person judges what it means. No automated accusations, ever.
Pricing
One box, whole store, a flat monthly fee.
Priced per store, by camera count — the only thing that separates the tiers. Plus a one-time setup and edge box from ₹4,000.
Watch
After-hours monitoring, up to 2 cameras. Alerts only while the store is closed.
Single
One camera, every rule, full-time. The lean single-choke-point plan.
Standard
Up to 4 cameras — the whole-store default. Back room, till, entrance, and one more.
Pro
Up to 6 cameras with headroom for larger formats and busier floors.
FAQ
The questions retailers ask first.
Does it use facial recognition?
No — never. Aproksha Retail Vision reasons about behavior and roles anonymously. There are no faceprints and no biometric database, which is exactly what keeps it privacy-first and legally light.
Do I need to buy new cameras?
No. It runs on your existing CCTV via the RTSP sub-stream. One small box is added in the back office; nothing on the shop floor changes.
Where does the video go?
Nowhere. All processing happens on a box inside your store. Raw footage never leaves the premises — only short evidence clips and plain-language summaries reach you.
What does it actually flag?
After-hours presence, restricted / back-room intrusion, loitering in a watched zone, and an unattended counter. It does not claim to detect "theft" — it surfaces the moments worth a look and hands you the clip.
Is it legal to run in India?
Yes, with the usual monitoring signage and a stated security purpose. Because there is no face recognition and clips auto-purge, it stays well clear of the heavier biometric category under the DPDP Act.
Run it across your stores.
We are onboarding pilot partners now. Bring a few locations and we will stand it up on your feeds, on your hardware, and show you what it catches — during a silent run, before a single alert goes out.