How Digeiz Uses AI to Enhance Visitor Journey Analysis


Created in 2015, the French start-up has developed a software solution to accurately measure pedestrian traffic indoors, in shopping centers.

AI for precision. That is the objective of a French start-up called Digeiz. Launched in 2015, the start-up has developed a solution that details with a very high level of accuracy statistics on indoor pedestrian traffic. Based on several artificial intelligence models, the software provides qualified data to shopping center managers. Digeiz stands out from its competitors by guaranteeing a reliability level of 95%. Even today, a large majority of malls base their data on customer journey analysis software solutions for traffic monitoring based on IoT sensors, which are less precise.

A solution based on several neural networks

The solution offered by Digeiz provides three different types of data. Classic counting: AI defines virtual doormats (a defined area) and counts the number of people passing through those areas. Segmentation: AI is capable of distinguishing genders, age groups, and the number of people walking together. And finally, cross-visit data: AI analyzes the visitors’ routes between different cameras to see correlations between, for example, visited shop windows and frequented stores. To obtain these three types of data, Digeiz’s software solution is based on three different neural models.

“The software is a two-part solution. On one hand, there are the servers located in the shopping center that process real-time video streams. On the other hand, there are the algorithms and neural networks that analyze the images, detect people, understand their paths, and analyze the different data streams,” explains Nicolas Bouvattier, CEO of the company. The AIs used by the software have been adapted from open-source frameworks and re-developed internally to perfectly fit the clients’ specifications. “Shopping centers represent a unique challenge in terms of computer vision: they impose strong constraints in terms of brightness. For example, managing skylights on marble is extremely complex, with very bright areas next to shadowy areas. Shopping centers also impose constraints on population density (crowds), as well as the distance between the cameras and the observed subjects. We therefore had to develop specialized, robust neural networks,” explains the manager. To train the models to recognize, differentiate, and analyze humans, Digeiz’s teams fine-tuned the models on academic datasets before expanding them with more representative data from the operating environment (shopping centers).

An on-premise deployed software

The operation of the software relies on an infrastructure installed mainly on-premise, at the client’s site. “For a shopping center, the number of cameras and servers required depends on the size of the site. We only use Nvidia servers equipped with GPU graphics cards for image processing. Each of these servers can process video streams from up to 80 cameras simultaneously. Depending on the total number of cameras present in the shopping center, we stack as many servers as necessary to handle all the streams. The shopping center can work with its usual camera installation provider to expand its existing network if necessary, based on our recommendations,” specifies Nicolas Bouvattier. Only aggregated statistical attendance and journey data are sent to the cloud to be viewed from the dashboard.

The particularity of the solution also lies in its live design. The system analyzes all streams in real-time. The only real limitation comes from GDPR and the ethical constraints that Digeiz imposes. Since the AI only uses visual characteristics such as clothing color (no facial recognition), the tool is not able, for legal reasons, to track the customer day after day within the shopping center. Finally, the second limitation is that the system only works with aggregated statistical data and not at the individual level. “In concrete terms, we are not able to individually track a customer to analyze their complete online and offline journey and make the link between their web browsing, in-store purchases, etc. This level of personalization would require lifting the anonymization of the data, which would not comply with GDPR,” justifies the business leader.

Reliable data

By basing its analysis on AI and providing reliable data, the start-up’s technical solution provides shopping center managers with practical information for decision-making. “We can precisely analyze which profiles of people are exposed to campaigns, if targeting is working, and if these campaigns subsequently generate more store traffic,” indicates Nicolas Bouvattier. Finally, the tool allows the commercialization of media space in the shopping center with a yield management approach. Very concretely, “during a new movie release or product campaign, the center can sell spaces guaranteeing a targeted audience, and thus monetize the flow of people,” in real-time.

Since its beginnings, Digeiz has come a long way and now signs contracts with major groups in the sector. The start-up, which now has around thirty employees, signed an agreement with the Unibail-Rodamco-Westfield Group in November 2023 to deploy its solution in 24 shopping centers in 10 European countries. In France, the Forum des Halles, the 4 Temps, and Vélizy 2 centers, in particular, are already equipped. More recently, the start-up has concluded a new contract with Hammerson for the deployment of the solution in 7 shopping centers in Ireland, England, and France. The next step for this gem of the French ecosystem is international expansion and the arrival in the second quarter of 2024 of a certification of the solution by an independent third party.

Leave a Reply

Your email address will not be published. Required fields are marked *