Free Articles, Free Web Content, Reprint Articles
Tuesday, June 2, 2020
Free Articles, Free Web Content, Reprint ArticlesRegisterAll CategoriesTop AuthorsSubmit Article (Article Submission)ContactSubscribe Free Articles, Free Web Content, Reprint Articles

Artificial Intelligence for Retail in 2020: 12 Real-World Use Cases

For decades, traditional analytics have worked perfectly fine for the data-driven retail industry. However, Artificial Intelligence (AI) and Machine Learning (ML) have introduced an entirely new level of data processing which leads to deeper business insights. Data scientists could open a new world of possibilities to business owners extracting anomalies and correlations from hundreds of Artificial Intelligence/Machine Learning models.

Between 2013 and 2018, Artificial Intelligence startups raised $1.8 billion in 374 deals, according to CB Insights. Amazon can take credit for these impressive numbers, because they made business leaders change their minds about Artificial Intelligence in the retail market – both physical stores and e-commerce strategies to stay ahead of the competition. At the moment over 28% of retailers are already deploying Artificial Intelligence/Machine Learning solutions, which is a sevenfold increase from 2016 when the number was only 4%.

Using AI in the Retail Industry: SPD Group Use Case


Here at SPD Group, we know how retail businesses could benefit from AI because we have practical experience. We developed a system for product suggestions based on tracking a customer’s location and actions in a store. It was aimed to boost sales for shop owners while improving customer satisfaction by providing smart recommendations.


This project started as an idea to improve the CRM in a supermarket. We had a goal to implement a customer identification system that did not require physical id cards and connect it to the existing CRM process. To achieve that we had to analyze videos from cameras, identify customers in frames, and track each customer’s position in the store to match it to the location of the products. In addition to, the system also needed the capacity to alert staff when a customer is standing too long in one location so personnel could assist him or her if needed. All this valuable information is gathered to determine the products each customer prefers to create future offers for him or her.


Building this solution started by using the existing security cameras in the store and setting up just a few additional cameras. We had used the YOLO model with pre-trained weights because of its effectiveness in identifying people. The Tracklet Association method comes into play when someone has the goal to track multiple objects. This method processes and slightly improves the information from YOLO, the distinct similarities of the same visitor (this is called appearance embedding), forms tracklets, and groups them with the help of the network flow graph. Simply put, our system is able to interact with multiple customers now. We calculated the geometry of the cameras to determine their scope. Then, by implementing the perspective conversion, our system is now able to receive 2D coordinates on the location of certain customers.


After this system was installed in the store, the owners obtained an entirely new level of insights. With all of this information about customer’s preferences integrated into the CRM, they can predict demand for a particular product. More than that, business owners can come up with much more effective personal offers and promotional offers with adjusted price strategies for different groups of customers. Eliminating physical gift cards improved the shopping experience and customer satisfaction. Now, personnel can offer personalized discounts or ask about the experience with the last purchase — making the customer feel even more welcome.

 That was our experienceFeature Articles, but what about global smart retail trends?

Explore more here:

Source: Free Articles from


I am a tech writer at SPD Group

Home Repair
Home Business
Self Help

Page loaded in 0.271 seconds