Artificial Intelligence Driven Customer Monitoring and Analysis
Motivations and problematics of the project
E-commerce has seen the start of its golden era with the rise of data science and data analysis. Brands start to get a better understanding of their customers, which leads to better customer service and marketing strategies. The physical stores, on the other hand, have struggled to adapt since there are no mechanisms in place to track and understand customer behaviors.
The purpose of this project is to fill these gaps and put in place a system that would be able to gather customers’ data while respecting their privacy and give brands better insights into their operations and sales.
● High-resolution camera footage and ultrasound sensors that acquire key information on customer behaviors.
● Video Stream subject to multiple machine learning algorithms to extract in real-time relevant data points: Age, gender, emotion and customer journey in the store etc.
● Machine learning algorithms are programmed to monitor customer activity within stores. This data allows the clients to make decisions based on their customers’ specific activities.
● Real-time activity monitoring through the data points already installed produces alerts/reports informing the client of current footfall and staff activity.
● Real time customer recognition reports to devices: reports will be generated in real-time to on the ground staff through devices such as smartwatches and or tablets.
● A multidimensional cube to merge KPIs and facilitate their analysis based on different dimensions.
● Customized interfaces which include client branding, tailored reports and dashboards showcasing key data as per their request.
The solution trains five deep learning models to detect faces, age, emotion, gender, and footfall inside the store (through motion detection). The smart camera can avoid recounting the staff by detecting the persisting presence of the person inside the store and flagging them as a staff member.
The camera does most of the heavy work and sends only the necessary information to the database, resolving the connectivity challenge. The Raspberry Pi contains the trained algorithms we developed and can be produced in mass quantities.