Map
Map existing CCTV coverage to the store's commercially significant zones — no new hardware
The situation
ElAraby Stores is one of the Middle East's most recognised consumer electronics retailers, operating large-format flagship stores carrying everything from smartphones to home appliances. Their commercial performance is tied directly to how well the physical store converts a few percentage points of foot traffic into purchase decisions — and across a high-footfall environment, that margin is worth real revenue.
What the commercial team could see, and what it couldn't:
- Strong transactional data — sales by SKU, category trends, stock movement
- No aisle-level engagement data — no way to tell which zones attracted browsers versus serious shoppers
- No demographic insight — no mechanism for understanding which segments gravitated to which parts of the store
- Invisible conversion friction — products picked up, examined, and returned, with no way to track it systematically
- Merchandising decisions made on a quarterly cadence, based on supplier recommendations and occasional manual floor walks
A previous attempt with third-party footfall counters gave aggregate entry and exit numbers, useful for staffing but with nothing at the zone or demographic level. The fundamental gap remained — and it needed to close without new hardware, a lengthy IT programme, or any disruption to daily trading.
The challenge
The challenge was to turn the CCTV infrastructure ElAraby already owned into a commercial intelligence layer — without new camera hardware, without disrupting daily trading, and without a lengthy IT project standing between the insight and the shop floor.
Our approach
Every design decision followed one principle: work entirely within the existing store environment. The cameras were already there — what was missing was the intelligence layer that could turn a raw video feed into structured, commercial-grade insight.
Our objective was to:
Model
Deploy purpose-built models for demographics, heat maps, and behavioural events
Deliver
Turn raw detections into a continuously updating dashboard for store managers
Recommend
Translate detected patterns into specific, prioritised merchandising actions
What we did
1
Mapped CCTV coverage to the zones that mattered commercially
We identified the zones with the greatest commercial significance — main entrance flows, premium display areas, high-margin category aisles, checkout adjacencies, and service counters — and built feed routing and processing pipelines for each, with a phased plan to extend coverage further.
Designed with privacy and trading continuity in mind:
- Near-real-time frame processing without storing raw footage
- No new camera hardware — built entirely on ElAraby's existing infrastructure
- No disruption to daily trading during rollout
2
Deployed a suite of purpose-built deep learning models
Four models worked against the processed feeds, each targeting a specific commercial question rather than a generic video-analytics feature set.
The four models:
- Demographics — gender and estimated age-group classification, per zone
- Heat maps — dwell time and zone-to-zone transition tracking
- Behaviour — pick-up, price-check, and return-to-shelf detection, flagging high pick-up-to-return ratios as friction signals
- Patterns — repeat-session behaviour that recurs across days, separating structural issues from daily noise
3
Delivered insight through a dashboard, not a data dump
Zone-level heat maps, demographic breakdowns, and behavioural summaries updated continuously during trading hours and compiled into a structured weekly report for store managers and the central commercial team.
The commercial team received their first actionable report within two weeks of go-live — before the full store rollout was even complete.
4
Built an automated recommendation layer on top of the data
The weekly report didn't just present numbers. It translated detected patterns into specific, prioritised merchandising actions — which products to move, which zones to reconfigure, which aisles warranted promotional investment, and which high-intent segments were underserved by current placement.
This is what turned data into decisions:
- Prioritised, ranked merchandising actions — not raw charts
- Segment-level guidance on underserved high-intent customers
- A weekly cadence replacing quarterly, instinct-led reviews
How it works
From an existing camera feed to a prioritised merchandising action — no new hardware, no disruption to trading.
1
Existing CCTV feeds mapped
Coverage is mapped to the store's most commercially significant zones.
No new camera hardware2
Privacy-safe frame processing
Frames are processed in near-real-time for behavioural signals, without storing raw footage.
No footage stored3
Demographics, heat maps & behaviour
Four purpose-built models classify demographics, dwell time, and product interaction events.
4 purpose-built models4
Continuous manager dashboard
Zone-level heat maps and behavioural summaries update throughout trading hours.
Updated during trading hours5
Prioritised recommendations
Detected patterns are translated into specific, ranked merchandising actions each week.
Weekly structured report“
We always suspected there was more revenue sitting on the floor than our numbers showed. Atomic Loops gave us the visibility to act on it. The first insight alone paid for the engagement.
ElAraby Stores
“
Before this, I was walking the floor twice a week and writing notes. Now I open a report on Monday morning and I know exactly what to change and why. It has made my job meaningfully different.
ElAraby Stores
~22%
In-store revenue uplift
Within 3 months of deployment
31%→41%
Conversion rate
After merchandising changes informed by AI insight
<2 hrs
Weekly reporting effort
Down from ~18 hours of manual observation
100%
Store floor coverage
Versus 3 manually sampled zones previously
The outcome
What changed within the first quarter:
- In-store monthly revenue grew by approximately 22% over the three months following deployment
- Overall in-store conversion rate improved from roughly 31% to 41%
- Weekly manual observation and reporting time fell from ~18 hours to under 2 hours
- Store floor insight coverage went from 3 manually sampled zones to 100% of the floor
- Merchandising review cadence moved from quarterly to weekly
- An unplanned find — a premium appliances aisle in the wrong zone — drove a measurable category revenue lift within four weeks of relocation
Frequently asked questions (FAQs)
1) What is in-store AI vision analytics?
It's a system that processes video from store cameras in real time to generate structured commercial insight — foot traffic patterns, dwell time, demographics, and behavioural signals — rather than raw footage a person has to watch.
2) Can retail heat maps work with existing CCTV cameras?
Yes. A heat-map and analytics layer can typically be built directly on top of the CCTV infrastructure a store already has, without installing new camera hardware.
3) How does AI classify customer demographics in a store?
A deep learning model processes video frames to estimate attributes like gender and age group for detected customers, aggregated per zone rather than tied to any individual identity.
4) What is dwell-time analysis and why does it matter for merchandising?
Dwell-time analysis measures how long customers spend in each zone. Zones with high dwell time but low conversion often point to a merchandising or positioning issue worth investigating.
5) How is customer privacy protected in video-based retail analytics?
Processing can be designed to extract behavioural signals from frames in near-real-time without storing raw footage, so the system generates aggregate insight rather than a searchable video archive.
6) What is conversion friction and how can AI detect it?
Conversion friction is when customers engage with a product — picking it up, examining it — but don't buy. AI can flag high pick-up-to-return ratios as a friction signal worth investigating.
7) How quickly can a retailer see results from in-store AI vision?
In ElAraby's case, the commercial team received their first actionable report within two weeks of go-live, before the full store rollout was complete.
8) How does Atomic Loops help retailers turn CCTV into commercial insight?
By building the intelligence layer on top of infrastructure retailers already own, and translating raw detections into prioritised, specific merchandising recommendations rather than dashboards alone.
Industry Services
Education
Focus areas - Education
© 2026 London School of DIGITAL BUSINESS.
All Rights Reserved