AI-driven Retail Incident Detection Platform

The customer was looking to strengthen their development capacity to support ongoing product development. Initially, they approached Altabel in search of Сomputer Vision expertise for a highly specific scope. Altabel proposed strong candidates; however, the customer ultimately selected a specialist with a narrower fit for their immediate technical requirements.

The customer’s product is a computer vision platform designed to analyze video streams from physical retail stores and detect incidents that require staff attention. The solution focuses on loss prevention, operational safety and real-time monitoring of in-store activity.

The platform is built with a strong understanding of retail processes and store operations. This is reflected in its structure, role-based access model and the emphasis on practical workflows used by store personnel and administrators.

Altabel’s Angular developer was responsible for the web interface used across the platform’s core operational workflows. The scope covered development and refinement of dashboards, alert handling and case-related UI, as well as ongoing improvements driven by evolving product and operational requirements.

The application uses a structured role-based access model with defined permissions. Administrative users can manage system functions, including company and store configuration, interface settings and translation updates.

Users can access dashboards with statistics for selected time periods across multiple companies and stores. Notifications can be enabled or disabled per store for both desktop and mobile use.

The platform includes several monitoring modules addressing different retail scenarios, including:

• automatic theft detection with product list upload and an incident calendar;

• detection of abandoned transactions when payment is initiated but not completed;

• real-time tracking of suspicious behavior;

• monitoring of activity after store closing with configurable cameras and time intervals;

• push-out detection for cases when a person leaves with a trolley;

• shelf sweep detection to identify rapid product removal from shelves.

All detected events appear in the alert section with contextual information, including a description, video from multiple cameras and a chronological activity history. Alerts can be reviewed, rejected or confirmed and converted into cases. A case may include files, information about the offender, witnesses, missing products, product value and other data required for internal reporting and police documentation. Alerts and cases can be filtered by status, application type, time range and identifier.

The platform is available via a web interface and a mobile version that provides access to alerts and cases. The mobile interface offers the same core functionality with a layout adapted for smaller screens and is implemented in JavaScript. A native mobile application is under development as a separate track.

The platform remains under active development and is being used in a live retail environment. At the time of writing, the solution has been adopted by a large retail chain in Switzerland in a non-commercial usage model. On the one hand, this allows retail teams to explore the platform’s capabilities in real store conditions. On the other hand, it enables the customer to continue developing and refining the product using live operational data.