How AI Investment in Retail Works And Where iMerit Comes In

November 30, 2020

Retail investment in machine learning AI algorithms falls into two general buckets nowadays, the customer experience bucket, which is a combination of an E-Commerce site’s search engine and customer service bots, and an operations bucket — stocking, fraud detection, and other back office functions — that in both the physical and virtual worlds have begun to rely increasingly on machines to augment if not entirely replace humans.

Much of the industry’s customer experience investment in machine learning AI aims to enhance the ability of eCommerce search engines to deliver unerringly relevant search results and product recommendations. There’s also a significant focus on sentiment analysis of customer reviews and comments, data mining, and related initiatives.

Much of the spending in AI-driven operational initiatives, conversely, has been for brick and mortar retail. A growing number of major big box retailers already rely on advanced sensor and camera-equipped robots to roam the aisles of their physical stores to run inventory checks. Robotic arms outfitted with sophisticated sensors and cameras have been trained to examine the shelves, top to bottom, detect missing (and misallocated) merchandise, and navigate the stores aisles, much as an autonomous vehicle might navigate the streets of a small town.

The algorithms that operate those robots, though, have to match the capabilities of the inventory checkers they are replacing. Not only do they need to identify empty spaces on a product shelf; in a supermarket, for example, they have to understand the difference between a can of corn and a can of peas, mistakes in price tags, and even mis-stocked products – such as cans of dog food placed in the canned fish section.

Data annotation in this scenario leverages a combination of Computer Vision techniques – the use of semantic segmentation and bounding boxes to identify and categorize individual products – sometimes Natural Language Processing to evaluate signage, and even on occasion geospatial intelligence for device navigation within the store and shelf locations. One such example: a combined discipline project employed drones inside a major mass market retailer to identify empty shelves in need of restocking, the specific products (based on shelf labeling) and even mistakes in product pricing.