Video and vision
How to Turn Existing CCTV Cameras Into Retail Intelligence
To turn existing CCTV cameras into retail intelligence you keep the cameras, add a small computer in the store that reads their streams, and change what the output is: not footage for a guard to scroll through, but events a manager can act on. Most Indian stores can do this with the recorder already installed, provided it exposes streams and the cameras can see the floor. The catch is that camera events are only half of intelligence; the other half is what was said to the shopper the camera saw.
Audit the setup you have
| Factor | Good enough | Not good enough | Fix |
|---|---|---|---|
| Recorder | NVR or DVR that exposes RTSP or ONVIF streams | Analogue DVR with no network stream | Add an encoder, or replace the recorder |
| Resolution | 1080p or better on the entrance and counter cameras | Below 720p, heavy compression | Swap the two cameras that matter most |
| Mount height and angle | 2.5 metres or higher, tilted down at the floor | Counter height, or along a wall | Re-aim before buying anything |
| Entrance coverage | One camera sees the full doorway from inside | Door half out of frame or seen at a slant | Move or add one camera |
| Zone coverage | Each display zone seen from above by one camera | Zones seen edge-on, shoppers overlapping | Accept zone dwell for open zones only |
| Lighting | Even light, no window glare across the lens | Backlit doorway, reflections in glass | Blinds, film on glass, or evening validation |
| Frame rate | 5 to 10 frames per second available per stream | Recorder throttles sub-streams to 1 or 2 | Use the main stream for two cameras only |
| Network | Wired link between recorder and edge box | Recorder on a distant Wi-Fi | Put the edge box next to the recorder |
The edge build
- Place a small computer, fanless if the store is dusty, beside the recorder and connect it by cable.
- Point it at the RTSP sub-streams of the cameras that passed the audit. Two to four cameras per store is the usual first set.
- Configure lines and zones per camera view: entrance line, counter zone, two or three display zones, a staff zone.
- Run detection and tracking on the box. Frames are processed in memory and discarded. Nothing is stored except events.
- Send events upstream over the store's existing internet: a few kilobytes an hour, which survives a poor connection and can queue when it drops.
- Validate every measure by hand for a week, then switch the report on.
From footage to events
- Entries and exits by hour, with staff and re-entries excluded.
- Group arrivals: two or more tracks entering together.
- Dwell by display zone, with a threshold so a pass-through does not count.
- Wait at the counter, and how many were waiting at once.
- Time to first staff contact for each shopper track, and the tracks that never got one.
- Unattended walk-outs: tracks that ended at the exit with no staff proximity.
- Display hotspots: which zones held the longest dwells in a week.
Intelligence is the join, not the feed
A list of walk-outs is a count. It becomes intelligence when it sits beside the reason. The camera can say that eleven shoppers left the premium fridge zone unattended between 6 and 8 pm on Saturday. The consented conversations from that floor in that window can say that the two advisors on duty were both tied up in finance discussions, that the exchange offer was not mentioned once, and that the customers who did get attended asked about a rival's price. What the camera sees surrounds the sale; what the conversation records is the sale. Joined by time and zone they explain the evening. Joined by identity they would not explain it any better, and would break the law.
Borentis captures consented in-store conversations on the advisor's phone today and scores them on the playbook. ShopperDNA, on the roadmap, is the vision product that would take events from a store's existing CCTV and join them to those conversations by time and zone.
What changes for the store manager
- A morning view of yesterday: footfall by hour, wait at the counter, unattended walk-outs, next to the conversation coverage and the most-skipped playbook step.
- A staffing case based on when shoppers went unattended, not on when the manager felt busy.
- A display decision based on where dwell happened versus where conversations happened.
- A recovery list for the hours when walk-outs peaked, drawn from the conversations in which a number was taken.
Privacy and retention
Nothing here identifies a shopper. No facial recognition, no age or gender, no emotion. Events are anonymous tracks, kept as counts by hour and zone. Footage retention follows the security policy that already exists. The entrance notice is updated to say that cameras are also used for anonymous counting, and that sales conversations are recorded only with consent. Under India's DPDP Act this is the low-risk configuration, and the edge design keeps it that way because no frame leaves the store.
Frequently asked questions
Will this work with the DVR brands common in India?
Most recorders from mainstream brands sold in the last few years expose RTSP or ONVIF streams. Check the model; older analogue units may need an encoder.
Do I need the cloud for retail intelligence from cameras?
Not for the video. Processing runs on the edge box in the store. Only events go to a dashboard, which can be cloud-hosted without any video ever leaving the premises.
Will analytics interfere with security recording?
No. The edge box reads a second stream from the recorder; recording continues unchanged.
How many cameras does a store need?
Two well-placed cameras, entrance and counter, produce most of the value. Two to four more on display zones add dwell and hotspots. Beyond eight, the marginal camera rarely adds a measure a manager will act on.
Related reading
- How to use CCTV for retail analytics
- How AI can monitor retail stores
- DPDP consent notice template for stores
- ShopperDNA, on the roadmap
Where Borentis applies this
- Walk-in Recovery: The customer who left is still yours.
- Execution Scorecards: See the floor before the P&L does.
Borentis is the Agentic Operating System for Customer Interactions, built for Indian retail floors: consented one-tap capture on the advisor's phone, every conversation scored against your playbook with the evidence behind every number, leads created when a number is heard, and coaching from your own best conversations.