Levi’s AI drive this year will culminate in new updates that bring computer vision and visual search technology to the denim brand for the first time. An executive inside the company confirmed to WWD that the announcement will go out sometime this week, in a move teed up for the holiday season.
The goal, according to Louis DiCesari, global commercial head of data, analytics and artificial intelligence at Levi Strauss & Co., was to boost curation and discovery for shoppers navigating the site, which brims with more than a thousand products. To help, it developed the new Grid Sort feature using a combination of computer vision and neural networks.
“When you land on a product page, the order is determined by a set of rules that include things like inventory, level and tiering,” DiCesari explained. “We’re furthering that personalization by using real-time information about the user to build a profile on the kinds of products they’re interested in based on what they’ve looked at and purchased in the past. We want to boost up the products that we think the user is interested in.”
Computer vision allows machines to read and understand images, while neural networks, a form of machine learning or deep learning, takes cues from the structure of the human brain and how biological neurons fire and communicate with each other. Simply put, the latter aims to imbue machines with more human-like ways of thinking and reasoning, allowing them to identify patterns and improve over time.
“The computer vision allows us to learn from pictures, and better link our family of products and how they relate to each other,” DiCesari continued. “Behind this is a neural network that figures out which products look like other products visually and shows them to the consumer, in the order we predict will be most interesting to them.”
This sort of image-driven AI application can be more robust than relying on text-based descriptions, which can be hampered by reliance on limited attributes or categorizations. Grid Sort doesn’t have the same limitations, and it’s fast, since it’s updated continuously, nearly in real time.
The feature just entered testing in the U.S., and the plan is to roll it out globally next year. When it arrives, it will join another feature that will have launched by then: visual search.
Slated to release in November, visual search allows shoppers to search Levi’s e-commerce site by uploading photos. The tech will pull up results from the store’s assortment that looks similar to the items in the images. Like Grid Sort, it’s also driven by the company’s neural networks.
Levi’s may be emphasizing computer vision, but that doesn’t mean it’s abandoning text altogether. In fact, as recently as last month it rolled out Personalized Popular Filters, which are clickable text-based filters that live on product pages.
Like the others, it’s based on what the company knows about its customers. “[Personalized Popular Filters are] different words that narrow down the product mix, depending on what we think the consumer is most interested in,” said DiCesari. “So you might see, like stretchy jeans, or skinny jeans, or 501s, or black jeans, or rib gauge — whatever we predict the consumer is most likely to be interested in. Then the consumer can click that button and see only those products that meet that criterion.”
New shoppers start out with a standard set of generally popular filters of the moment, “but as we learn about the consumer, we update those filters,” he said. “The algorithm is updated every day, so people who come back to the website will see updated filters based on how they’ve interacted with the website.”
The latest set of e-commerce updates cap off an intense period for Levi’s AI team, which had to scramble after the upheavals of the pandemic laid to waste its road map for 2020.
According to a January blog post, “Determining optimal store inventories and predicting store traffic became nearly impossible with stores closed, and the team found themselves with a small problem: With no new data coming in, their data models were about to break.”
The team jumped into high gear, collaborating cross-regionally to rebuild its predictive models as quickly as possible. Soon after, the company decided to kick off its first AI bootcamp, its bid to cultivate more homegrown talent, as well as teach employees across the organization how to work with data in different ways.
The efforts seem notable and, for a Bay Area denim brand that has been doing business for more than 160 years, relatively futuristic.
DiCesari joked that Levi’s was “the original Silicon Valley startup,” which may be an apt description. The company applies AI to various parts of the business, including demand forecasting, marketing and other areas. He called it “a huge focus for us, as a key component of our digital transformation.”
The newest features have been more than two years in the making, he added, and they may mark a new phase in the company’s AI push in the e-commerce experience. But it’s not limited to shopping, especially when it comes to computer vision. DiCesari wouldn’t disclose precisely what Levi’s is working on next, but he hinted as much.
“Computer vision can apply to almost every component of what any fashion apparel company does, even down to the design of products,” he said. “So this is an area where the team is definitely dreaming big.”