Video and vision
How AI Can Monitor Retail Stores: Cameras, Checklists and Conversations
AI can monitor retail stores in three ways that are often confused: through cameras, which produce anonymous events about traffic, waiting and attention; through checklists and photo audits, which verify that the store looks and runs as planned; and through consented conversations, which score whether the floor sells as trained. Each answers a different question. The mistake is to buy one and expect the answers of all three, or to let camera monitoring drift into watching people rather than measuring a store.
Three things worth monitoring
A store can fail in three places. Traffic can arrive and go unattended. The store can be out of standard: display down, offer signage missing, stock gap. And the conversation can miss the playbook: no demo, wrong offer, objection left standing, no number taken. Cameras see the first. Checklists and photos see the second. Only the consented conversation sees the third, and it is the one that decides revenue.
Layer one: cameras as an event source
- Footfall by hour, groups, wait at the counter, dwell by zone, time to first contact, unattended walk-outs, display hotspots.
- Runs on the store's existing two to eight cameras through an edge box; frames stay in the store, events leave.
- Produces counts, not people. No facial recognition, no age or gender estimation, no emotion reading.
- Limits: occlusion, crossing, low mounts, glare and evening light; no reasons, no words.
Layer two: checklists and photo audits
- Opening and closing, hygiene, display and offer compliance, stock on shelf, with a photo as proof and a timestamp.
- AI checks the photo against a reference: is the standee up, is the price tag current, is the bay full.
- Staff-facing; no customer data; the most mature layer in Indian retail today.
- Limits: self-reported timing, and a perfect checklist says nothing about how the store sells.
Layer three: the consented conversation
- Captured on the advisor's phone with the customer's consent, in Hindi, English and Hinglish; audio deleted after transcription.
- Scored against the playbook step by step: greeting, discovery, demo, offer, objection, ask, follow-up, with the transcript line behind every score.
- Produces the reasons: the objection in the customer's words, the rival named, the offer misquoted, the number taken or not.
- Limits: coverage depends on the advisor tapping record, so coverage is reported beside every score.
Privacy-safe versus biometric monitoring
| Technique | Data produced | Position under India's DPDP Act | In the Borentis approach |
|---|---|---|---|
| Footfall and zone events | Anonymous counts by hour and zone | Notice at entrance; low risk | Yes, via ShopperDNA, on the roadmap |
| Staff-shopper proximity | Time to contact, unattended visits | Notice; staff informed; low risk | Yes, via ShopperDNA, on the roadmap |
| Photo audits of the store | Images of displays and shelves | No customer data | Yes, through operations checklists |
| Consented conversation capture | Transcript and scores; audio deleted | Per-conversation consent evidence | Yes, live today |
| Facial recognition | Identity of customers | Biometric; consent not realistic at a door | No |
| Age and gender estimation | Estimated demographics per visitor | Biometric processing; high risk | No |
| Emotion recognition | Claimed mood per face | Biometric and unreliable | No |
| Continuous staff surveillance | Individual staff movement logs | Employee data; trust cost | No; staff appear only as a proximity rule |
One scorecard, joined by time and zone
Monitoring earns its keep when the three layers land on one page per store. Saturday 6 to 8 pm: 140 walk-ins, 11 unattended exits from the premium zone, offer signage confirmed up at 5 pm, 32 consented conversations of which 9 skipped the exchange offer and 4 recorded a rival's price. The camera reports the conditions the sale happened in; the conversation reports the sale. Neither layer knows which shopper was which, and the scorecard does not need it, because the hour and the zone are the join.
Borentis today captures consented conversations on the advisor's phone and scores them on the playbook, and joins them to operations checklists on one scorecard. ShopperDNA, on the roadmap, is the camera layer: anonymous events from existing CCTV joined to those conversations by time and zone.
What AI monitoring should not become
Watching individual staff on camera all day is surveillance, not monitoring, and it costs more in trust than it returns in data. The useful staff measure is a rule, was a staff member within reach of a shopper and how soon, reported as a store number. Coaching comes from the conversation, with the transcript line, not from a clip of someone standing still.
Frequently asked questions
Can AI monitor a store remotely?
Yes, as events and scores on a dashboard. Video itself should stay in the store on an edge device; a manager needs the counts and the transcript lines, not a live feed.
Can AI monitor store staff performance?
It can measure how quickly shoppers were attended, as a store number, and how each consented conversation ran the playbook, with evidence. It should not log individual staff movement.
Is AI store monitoring legal in India?
Anonymous camera events under notice, staff-facing checklists and consented conversation capture are each defensible under the DPDP Act. Facial recognition and demographic profiling of customers are not part of the approach.
Which layer should a retailer start with?
If conversion is unknown, cameras for the denominator. If conversion is known and low, conversations for the cause. Checklists usually exist already and should be joined to whichever comes next.
Related reading
- Turn existing CCTV cameras into retail intelligence
- How computer vision is used in retail
- What is multimodal AI for retail?
- DPDP consent notice template for stores
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.