Understanding customer behavior in a physical store is critical to staying competitive, yet traditional data collection methods rarely deliver the timely, actionable insights retailers need. While e-commerce platforms measure every click, brick-and-mortar retailers and shopping malls have historically operated with far less visibility. AI-powered video analytics closes that gap: by applying computer vision and machine learning to in-store video feeds, it converts cameras into a continuous source of business intelligence.
This whitepaper explains how AI-powered video analytics is transforming the retail sector with real-time insights that drive customer engagement, optimize store layouts and streamline operations. Download the full whitepaper using the form on this page. Below you will find a practical overview of the technology: what it is, what it measures, the use cases that matter for stores and shopping malls, and how to deploy it in full compliance with the GDPR.
What is AI-powered video analytics in retail?
Video analytics is an AI-based solution that processes video signals and converts them into structured, intelligent data that supports profitable decisions. It began as a security tool, but it has evolved into something much broader: a way to measure business performance and reinvent the customer experience in physical spaces.
The difference between plain CCTV and AI video analytics comes down to what happens after the camera captures the image. A traditional CCTV system records footage that someone has to watch. An AI system, such as Flame Hypersensor, analyzes the scene in real time and outputs metrics: how many people entered, where they went, how long they stayed, whether they bought.
Modern AI brings three practical advantages over older counting technologies:
- Accuracy. AI distinguishes people the way a human does, instead of counting shapes. It filters out loitering (security staff, cleaning personnel, people standing at the door) that used to bias footfall counts.
- Anonymity. Detection can be completely anonymous: the system classifies without identifying any personal trait, which matters both ethically and legally.
- Edge computing. Algorithms can run inside the camera itself, without expensive servers and with lower maintenance costs.
AI can also monitor its own data quality: if a camera moves or a count becomes anomalous, the system detects the error and triggers alerts, so decisions are never based on faulty data.
Knowing how full the store or mall is at different times became popular during the pandemic and is here to stay.
What AI video analytics measures
The core value of the technology is a set of metrics that describe how a physical space actually performs:
- Footfall and capture rate. People counting measures how many people pass by, how many come in, and when. The ratio between passersby and walk-ins (your capture rate) reveals how well your storefront and window displays attract traffic.
- Occupancy. Precise capacity measurement shows how full the space is at any moment, supporting staffing decisions and alerts.
- Zone traffic and heatmaps. A single 360 fisheye camera can cover roughly 40-50 square meters, divided into zones with individual traffic and dwell-time measurement. Heatmaps show where customers concentrate and which areas go cold. Explore this in depth in Traffic Insights.
- Customer journey. Trajectories between zones: how many people who visited zone A continued to zone B, and how zones correlate.
- Queues. Estimated waiting times per line, so staff can react before checkout friction damages the experience.
- Demographics without biometrics. Anonymous classification by broad characteristics, with no facial recognition and no personal identification.
- Conversion. The complete funnel: passersby, visitors, buyers. Conversion analytics can be segmented by zone or demographic group to show not just how many people buy, but who and where.
Traditional CCTV vs AI video analytics
Many retailers already own cameras. The question is what those cameras produce:
| Traditional CCTV | AI-powered video analytics | |
|---|---|---|
| Primary purpose | Security recording and incident review | Business intelligence, plus security functions |
| Data produced | Raw footage that requires manual review | Structured metrics: footfall, occupancy, dwell time, conversion |
| How it is used | Reactively, after something happens | Proactively: real-time dashboards, alerts and forecasts |
| Privacy | Stores identifiable images of individuals | Can operate on fully anonymous detection, with no biometric data |
| Infrastructure | Cameras plus recorders and storage | Edge computing: algorithms run inside the camera |
| Business role | Cost center focused on loss prevention | Decision support for operations, marketing and merchandising |
Video analytics use cases for retail stores and shopping malls
These are the use cases that consistently deliver value in stores and malls, oriented to the three departments that benefit most: operations, marketing and merchandising.
1. Traffic measurement and forecasting
Accurate entry counts, cleaned of loitering bias, reveal traffic patterns and trends: how Mondays behave, what happens between 5pm and 8pm. Forecasting on top of historical data helps plan staff and opening hours before peaks arrive.
2. Occupancy management
Knowing how full the store or mall is at different times became popular during the pandemic and is here to stay. It supports staff allocation, capacity alerts and a more comfortable shopping environment.
3. Zone analytics and customer journey
Zone-level traffic and dwell times show which areas of the space attract interest and where people actually stay. Combined with journey analysis (the correlation between zones), this informs layout changes, product placement and in-store pathways designed to guide customers toward key areas.
4. Queue management
Queue algorithms estimate each person's waiting time and detect service delays per line, so managers can reorganize queues and open positions before waiting damages the store's image.
5. Conversion funnel analysis
Measuring passersby, visitors and buyers produces true conversion rates, which become even more actionable when segmented by zone or demographic group. This is the metric that connects traffic data directly to sales performance.
6. Staff planning and performance
Traffic heatmaps guide staffing decisions: more personnel at peak periods, better shift planning by hour and day. The system can even notify employees when a customer has been standing in one spot for an extended time and may need assistance.
7. Loss prevention and security
The original use case still matters: motion detection in no-entry zones during and after opening hours, shoplifting prevention and fraud detection remain standard applications of video analytics.
8. Visual merchandising and marketing measurement
Automatic high-definition photographs (taken when no people are in frame) let visual and trade marketing teams track how displays evolve across stores. Meanwhile, comparing traffic, new versus repeat visitors and conversion before and after campaigns measures the real effect of promotions, so marketing can react to poor performance in real time.
In shopping malls, the same metrics serve an additional purpose: mall operators use footfall and zone-level data to give tenants objective performance reports, optimize store placement and manage common areas. See our dedicated solutions for the retail sector and for shopping malls.
Business benefits
Across these use cases, the returns concentrate in five areas:
- Higher sales conversion, by optimizing layouts, product placement and promotions based on how customers actually behave.
- Lower operating costs, by aligning staffing with real traffic instead of guesswork.
- Better customer experience, from shorter queues to store atmospheres tested and refined with dwell-time data.
- Smarter marketing, with campaign effectiveness measured against real in-store behavior rather than assumptions.
- Customer loyalty, tracked through visitor recurrence rates that show how many customers return and how often.
A real-world example: CaixaForum
Video analytics is not limited to stores. Flame Analytics deployed its AI-powered video analytics solution across all CaixaForum centers, the cultural spaces managed by Fundación "la Caixa" in Spain, using more than 250 cameras to analyze visitor footfall and peak hours, movement across zones and exhibitions, dwell times, and traffic trends through the Flame Dashboard. An integration with Snowflake lets CaixaForum combine these insights with other data sources.
With real-time data, CaixaForum improves space allocation and visitor flow, plans exhibitions with better engagement data, and makes data-backed decisions for future initiatives. The same measurement principles apply directly to shopping malls and large retail spaces: multiple zones, variable traffic and a visitor experience that depends on managing both well.
Automatic high-definition photographs (taken when no people are in frame) let visual and trade marketing teams track how displays evolve across stores.
Privacy and GDPR compliance
Privacy is the deciding factor in most video analytics projects in Europe, and for good reason: GDPR sanctions can reach 20 million euros or 4% of global annual turnover. But the regulation does not ban video analytics. It bans processing personal data without a lawful basis, and that distinction is where compliance begins.
Technology designed with privacy at its core makes the difference. Flame Hypersensor performs its analysis without capturing a single biometric data point: no facial recognition, no personal identification, only anonymous aggregated metrics. That privacy-first approach is trusted by more than 50 shopping malls.
If you operate a mall or a retail network in Europe, read our complete GDPR compliance guide for video analytics, which covers lawful bases, the biometrics line most retailers cross unknowingly, and a 15-point compliance checklist.
How to get started
Adopting video analytics does not require rebuilding your infrastructure. A typical path looks like this:
- Define the questions. Capture rate? Conversion by zone? Queue times? The metrics you need determine the setup.
- Audit your spaces. Entrances, key zones and checkout areas define where sensors deliver the most value.
- Pilot in one location. Validate data quality and build the reporting routine before scaling.
- Roll out and integrate. Extend to the full network and connect the data to the tools your teams already use.
The whitepaper available on this page covers each stage in more depth. And if you prefer to see the technology working on real data, book a demo of Flame Hypersensor: our team will show you how it applies to your specific space.
Frequently asked questions
What is the difference between video analytics and CCTV?
CCTV records footage for security review. AI video analytics processes the video in real time and converts it into business metrics: footfall, occupancy, dwell times, queues and conversion rates. One produces images to watch; the other produces data to act on.
Does video analytics require facial recognition?
No. Solutions like Flame Hypersensor work with completely anonymous detection: people are counted and classified without identifying any personal trait, and no biometric data is captured at any point.
Is video analytics GDPR compliant?
It can be fully compliant when the technology avoids processing personal data or has a proper lawful basis for it. The GDPR does not prohibit video analytics as such. Privacy-by-design systems that produce only anonymous, aggregated metrics are the safest route; our GDPR compliance guide explains the requirements in detail.
What can shopping malls measure with video analytics?
Malls typically measure footfall by entrance and by hour, occupancy, traffic and dwell times per zone, visitor journeys between areas, and queue times. This data supports tenant reporting, store placement decisions and common-area management.
Do I need new servers to run video analytics?
Not necessarily. With edge computing, the algorithms run inside the camera itself, which avoids expensive servers and reduces maintenance costs. Data quality is monitored automatically, with alerts if any count becomes anomalous.