5 Key Applications of AI Video Analytics in Retail: From Customer Heatmaps to Anti-Theft Tracking

Dozens of cameras are already installed, but shrinkage still remains high every year

A regional manager at a supermarket chain once said something that stayed with us:

“We have 360-degree video coverage across the entire store, but every monthly inventory count still shows that shrinkage is not going down.”

The problem is not the number of cameras. The problem is that traditional surveillance systems only “record what has already happened.” When an incident occurs, teams review footage afterward. Even when something is visible on screen, it is often already too late. That is the ceiling of a passive monitoring system.

AI video analytics changes this model. Having video footage does not equal having insight. True insight means the system can warn you before an incident escalates and provide data before decisions are made. Below are five of the most practical AI camera applications in retail, and the real operational shift behind each use case.

A convenience store chain in northern Taiwan wanted to measure the actual sales impact of changing new-product displays at endcaps. In the past, store managers relied mainly on intuition, and the results of a merchandising change could only be assessed after month-end inventory review. There was no clear answer to which aisle was a dead zone, which endcap attracted the most traffic, or which time periods had the highest customer density.

Heatmaps convert customer movement captured by cameras into a visual “temperature map,” where red indicates high-traffic areas and blue indicates low-traffic zones. Updated automatically every day, heatmaps help answer questions such as:

• Which endcap has the longest dwell time? Is it worth placing promotional products there?

• Which aisle is a dead zone? Is the issue caused by store layout or weak product appeal?

• How does traffic distribution change throughout the day? How should staffing be arranged?

After implementation, the retailer found that dwell time near the refrigerator area close to the entrance was much lower than expected. After adjusting the display strategy, sales improved noticeably. The merchandising decision cycle was shortened from monthly to weekly, shifting decisions from gut feeling to data.

A supermarket chain already had traditional surveillance cameras installed across the entire store, but product shrinkage remained high. Security staff had to monitor a large number of video feeds manually. Theft cases could only be confirmed by reviewing footage after the event, leaving no way to intervene before the incident unfolded.

Loitering detection: Define specific zones. When AI detects a person staying in the area beyond a preset duration, it automatically notifies security staff without requiring someone to watch the screen continuously.

Cross-camera tracking: When a suspicious person moves across different camera views, the system automatically hands off tracking between cameras and generates a complete movement path. Security staff no longer need to manually switch between feeds.

After implementation, abnormal activity alerts in storage-related areas changed from “discovered after the fact” to “reported in real time.” Security teams can now arrive while the event is still happening, turning shrinkage prevention from passive review into active intervention.

Warehouses, offices, VIP rooms, and other restricted areas require access control. Traditional access cards can be shared or borrowed, and manual verification is time-consuming. Managing a blacklist in real time is almost impossible with human supervision alone.

• Employee entrances: Registered personnel are automatically granted access, while unregistered individuals trigger an alert.

• Warehouse control: Areas limited to specific job levels can be protected, with entry and exit records stored automatically.

• Blacklist management: If someone with a previous theft record enters the store, the system can immediately alert staff.

Facial recognition involves biometric personal data. Its collection and use must comply with Taiwan’s Personal Data Protection Act, with proper legal basis and clear on-site notification. Before implementation, a compliance review of the deployment environment is recommended.

When customers wait more than three minutes at checkout, the likelihood of customer drop-off rises significantly. However, shift managers cannot monitor both the sales floor and checkout area at the same time. Relying on human judgment to decide when to open another checkout lane is always a step behind actual demand.

AI detects the number of people waiting at the checkout area in real time. When the queue exceeds a preset threshold, such as more than five people, the system automatically sends a notification to the shift manager. The same logic can also be applied to service-counter waiting time, fitting-room queues, and entrance crowd control.

This feature is especially useful as a value-added proposal for system integrators. A client may originally ask only for security, but by adding customer experience optimization, the ROI story expands across both security and operations departments.

Heatmaps tell you “where people gather.” Journey analysis tells you “where people come from, where they go next, where they stop, and where they eventually leave.” Combining these two dimensions creates a more complete understanding of in-store behavior.

Intelli-VMS uses Multi-Feature ID to identify people based on appearance characteristics such as clothing color, accessories, and body shape. It does not require facial recognition. Combined with cross-camera tracking, the system can reconstruct the full movement path of an individual shopper inside the store. This helps answer questions such as: How attractive is the promotional zone? What is the conversion rate after customers enter? Which part of the store layout creates friction?

In the past, this type of data was usually available only to large retail chains using expensive dedicated systems. AI cameras make it possible for mid-sized retail chains to build this capability at a more reasonable cost.

You are not selling cameras. You are selling a retail decision-making system.

At this point, you have seen five retail application scenarios. Now return to one key question: with the same AI video system, why do some system integrators get their quotes cut, while others win contracts across an entire retail chain?

The difference is the proposal framework. For retail clients, the decision-maker is often the VP of Operations or the Store Operations Director. They do not care about a specification sheet. They care about whether this system can give them a dashboard every morning that helps them make decisions.

• 1 Start with pain points, not specifications Do not open the proposal with a spec sheet. Start by asking: “Is your biggest challenge shrinkage control, staffing, or the inability to measure merchandising effectiveness?” This makes the client feel that you are solving their problem, not simply selling a product.

• 2 Package security and operational data into one ROI story “One system for anti-theft, customer flow analysis, and checkout staffing optimization.” Security, merchandising, and operations teams can all see the budget as relevant to their own goals. A cross-department proposal is much harder to cut.

• 3 Start with a pilot and let the data close the deal Do not begin with a full-chain deployment proposal. Start with one or two pilot stores, then return after three months with data. How much did shrinkage drop? Which display adjustment improved sales? These numbers are more persuasive than any presentation deck.

Book an Intelli-VMS Retail Site Assessment

Dexatek professional technical team provides on-site assessment services across Taiwan.

FAQ

Heatmap analysis is mainly processed by the backend AI platform. In general, IP cameras that support ONVIF Profile S can be connected. However, installation angle is important. A vertical overhead ceiling view provides the best results, while angled installation may affect people-counting accuracy.

No. Intelli-VMS uses Multi-Feature ID to identify people based on appearance characteristics such as clothing color, accessories, and body shape. It does not require facial recognition, which reduces privacy impact and makes compliance easier to manage. Facial recognition is a separate functional module that can be enabled depending on site requirements.

Under Taiwan’s Personal Data Protection Act, video recording may involve the collection of personal data and must have a legal basis or proper notification to data subjects. In practice, retail stores commonly use visible notices such as “This premises is under video surveillance” to fulfill notification obligations. Retention periods should be defined by company policy with automatic deletion cycles. A compliance review is recommended before implementation.

A practical starting point is the combination of anti-theft loitering detection and heatmaps. Anti-theft detection provides direct shrinkage-control value and is easy to explain to management in ROI terms. Heatmap data supports merchandising decisions, allowing marketing and merchandising teams to experience the system’s value as well. This combination spans both security and operations, making the proposal more persuasive.

POS integration is currently on the Intelli-VMS product roadmap and has not yet been officially released. Once available, it can compare AI customer traffic data with sales data to analyze visitor-to-purchase conversion rates. For specific project requirements, contact a Dexatek technical consultant to confirm the latest development status.

© 2026 Dexatek Technology. All Rights Reserved.