Nvidia adds AI workflows to retail to help combat shrinkage

Nvidia is looking to help retailers reduce losses from theft and fraud, as well as optimize in-store experience with a series of new retail artificial intelligence (AI) workflows announced today.

The new retail workflows are being announced ahead of the National Retail Federation (NRF) 2023 conference that gets underway in NYC on Jan. 15. With the Retail AI Workflows, Nvidia is aiming to provide retailers with AI technology that will solve challenges that they are facing in 2023. The new workflows are built with the Nvidia Metropolis Microservices framework for building AI applications.

One key challenge retailers face today is the issue of shrinkage, which is a term that is used to describe inventory and revenue losses due to misplaced and damaged products, as well as losses due to fraud and theft. The NRF’s 2022 Retail Security Survey reported that 65% of retail shrinkage is theft related.

“This [shrinkage] is a problem that has increased even more because of inflation and rising grocery product prices,” Azita Martin, VP and GM of AI for retail, CPG and QSR at Nvidia said during a press briefing.

Improving and securing the self-checkout experience 

One of the many ways that fraud occurs in retail today is at the front of the store, where self-checkout has increasingly become the norm.

Martin described one type of fraud that occurs called “ticket switching,” which occurs when a price tag or label for a lower priced item is substituted and placed on a higher price item. For example, if a shopper scans the price tag for a pack of gum, when in fact the item is a bottle of detergent.

The new Retail Loss Prevention AI Workflow has an algorithm that has been trained on product recognition and will identify if there is a potential anomaly with the product scanning process at checkout. If an anomaly is detected, the point of sale terminal can be shut down and the system can alert a retail associate to assist the customer.

How Nvidia trained its AI to spot retail fraud

Nvidia is using a number of techniques to train the Retail Loss Prevention AI Workflow system.

Martin said that the first thing Nvidia did was to train the model with hundreds of the most frequently stolen products, to help improve product recognition.

“This algorithm was trained to recognize products in a variety of sizes and shapes,” Martin said. “So we literally bought tens of thousands of dollars of products like steak, Tide, beer and razors, which are the most common items that are stolen, and we trained these algorithms.” 

Going a step further, the Nvidia Omniverse digital twin platform was used to help create synthetic data to further improve accuracy. Martin explained that the synthetic data generation enables developers to generate thousands of product image variations. 

“We probably didn’t buy every size of Tide and every packaging size of beer, but with synthetic data generation we were able to scale the data that trains the algorithms,” she said.

The Retail Loss Prevention AI Workflow also benefits from a continuous active learning approach. With active learning, Martin said that every time that the cashier or a customer is scanning a product for checkout, the AI is capturing additional new products or new packaging and takes that additional data to continue to improve the model accuracy.

Cleanup on aisle 3 as multicam tracking goes AI

Beyond helping to reduce fraud at the checkout, Nvidia is also rolling out AI workflows to help retailers use cameras to track in-store activity, as well as optimize store layouts.

The new Multicamera Tracking AI Workflow is a system that has been designed to help retailers track users in a store to see how they shop. Martin said that the goal is to better understand the customer journey around the store, with the ability to track objects from camera to camera.

With another workflow, the new Retail Store Analytics Workflow, Martin said that developers will be able to build a store analytics dashboard that provides insights about customer shopping preferences. The analytics workflow will also be able to generate a heatmap, identifying the most popular aisles and traffic patterns in the store.

“This is incredibly important in optimizing the merchandising, how the store is laid out and where the products go,” Martin said.

Originally appeared on: TheSpuzz