Video and vision
How Computer Vision Is Used in Retail: Nine Applications and Their Limits
Computer vision is used in retail to turn camera images into measurements: people counted, shelves checked, queues timed, displays verified, incidents flagged. Nine applications cover almost everything sold under the label, and they differ widely in how well they work on the cameras a store already has and in how much personal data they touch. Three work well almost anywhere, three need the right camera, and three should be avoided or done differently in an Indian store.
The nine applications
| Application | What it measures | Main limit | Privacy |
|---|---|---|---|
| Footfall counting | Entries and exits by hour | Re-entries, staff, groups at the door | Anonymous |
| Queue management | People waiting, time waited | Browsers near the counter | Anonymous |
| Zone dwell and heatmaps | Where shoppers spend time | Needs top-down views; crowding breaks tracks | Anonymous |
| Staff-shopper engagement | Time to first contact, unattended visits | Proximity is not a conversation | Anonymous with staff zone |
| Shelf monitoring | Out-of-stocks and gaps | Needs a close, well-lit camera per bay | No people involved |
| Planogram and display compliance | Does the display match the plan | Small changes and reflections | No people involved |
| Self-checkout and loss prevention | Unscanned items, suspicious patterns | False positives; staff trust | Can drift into identification |
| Demographics | Estimated age and gender | Unreliable and unnecessary | Biometric processing; avoid |
| Emotion and attention | Claimed mood or interest from faces | No scientific basis at store distance | Biometric processing; avoid |
Three that work almost anywhere
Footfall counting, queue measurement and staff-shopper engagement run on the entrance and counter cameras most stores already have. They need a clear line or zone, a staff exclusion rule and a week of validation. Their outputs are counts by hour that a manager can act on the same day: who was unattended, how long the queue ran, when to staff up.
Three that need the right camera
- Zone dwell and heatmaps: fair from a wall camera in an open zone, good only from a near top-down mount. Most showrooms add one camera per key zone if the first measures earn it.
- Shelf monitoring: the camera has to face the bay squarely from a couple of metres in even light. It is a grocery and pharmacy application more than a showroom one.
- Planogram and display compliance: works on a fixed view against a reference image; a photo audit from a phone often does the same job for less, which is how most Indian chains do it today.
Three to avoid or do differently
Demographics, emotion reading and face-based loss prevention process biometric data about people who did not agree to it. In India the DPDP Act makes that a consent question with no good answer at a shop door, and the analytical gain is small: an estimated age band does not change what the advisor should have said. Loss prevention can still use vision without identification, by flagging unscanned items at a self-checkout as a pattern rather than by recognising a repeat offender's face.
Everything a store needs from computer vision can be produced as behavioural events, anonymous tracks with timestamps and zones, on an edge device that never sends a frame out of the store. That is the design worth insisting on with any vendor.
Why none of the nine explains a sale
All nine describe the setting of a purchase decision: the traffic, the wait, the shelf, the display, the moment a staff member arrived. None of them can hear the decision being made. Whether the customer left because the price was quoted wrong or because the finance step was skipped is in the conversation, which is a separate, consented instrument. Read camera events and conversations by the same hour and the same zone and conversion starts to have causes. Tie them to a person and you gain nothing and lose the legal ground.
Borentis captures consented in-store conversations on the advisor's phone today, in Hindi, English and Hinglish, and scores them on the playbook. ShopperDNA, on the roadmap, would bring the first four applications above from a store's existing cameras and join them to those conversations by time and zone.
Choosing for an Indian store network
- Start with entrance and counter on existing cameras; validate for a week.
- Add engagement and unattended walk-outs with a staff rule.
- Add one top-down camera per key zone only where a display decision depends on it.
- Keep planogram checks on photo audits unless the format is grocery or pharmacy.
- Decline demographics and emotion; put it in writing with the vendor.
Frequently asked questions
What is the most common use of computer vision in retail?
Footfall counting, by a wide margin, because it runs on the entrance camera every store already has and produces the denominator for conversion.
Can computer vision recognise products a shopper picks up?
On a close, dedicated shelf camera, sometimes. On ordinary security cameras, no; hands and products are too small in the frame.
Is computer vision in stores legal in India?
Anonymous counting and shelf checks under a notice are the conservative position under the DPDP Act. Face-based identification, demographics and emotion analysis of customers are not, and are not needed.
Does computer vision replace mystery shopping?
It replaces the counting parts: queues, waits, whether the display was up. It cannot judge the sales conversation, which needs either a shopper or, better, the consented conversation itself.
Related reading
- What is retail video analytics?
- How AI detects customer behaviour in stores
- What is shopper intelligence?
- Is recording customers legal in India?
Where Borentis applies this
- Execution Scorecards: See the floor before the P&L does.
- Playbook Adherence: Your playbook, finally observed.
- Walk-in Recovery: The customer who left is still yours.
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.