Many companies are exploring generative AI, but it’s still early days for the technology, which only became widely known in November 2022. Generative AI shows great promise, but will the hype translate to ROI?
For Walmart, it already has. During its last earnings call, the giant retailer reported 4.8% revenue growth, bolstered by 21% growth in its e-commerce function. Walmart executives credited e-commerce growth to several factors, including improvements in deliveries, but one stood out: generative AI.
Walmart was able to roll out these AI tools relatively quickly because of its prior investments in technology, Anshu Bhardwaj, SVP and COO of Walmart Global Technology and Walmart Commerce Technologies, told CFO Brew.
For the past several years, Walmart has focused on “platformizing” its technology, she said, bringing it onto its machine learning platform, called Element. The platform provides a “base level of capabilities and functionality” with governance, compliance, security, and ethical safeguards built in, Bhardwaj said.
Checkin’ out that catalog: Walmart’s product catalog is a vital piece of its operations—it’s what customers use to find products and how the company tracks inventory and delivery orders. Over the past few years, the catalog has expanded as the retailer increased the number of sellers and items for sale, especially online. Populating and cleaning up the catalog is a huge task, but recently Walmart’s gotten an assist from generative AI. It used “multiple large language models” to “create or improve over 850 million pieces of data,” Walmart CEO Doug McMillon said during an earnings call, a task that would have taken “nearly 100 times the current headcount to complete in the same amount of time.”
The revised catalog has made it easier to match products to customers’ “intent” when shopping, Walmart US president and CEO John Furner said during an earnings call, “because the detail of each item and the product display pages have gotten so much better.” Coupled with data from AI-enhanced search capabilities, it’s given Walmart more insight into its customers, he said: “Our ability to show these items in a very high quality way,” he noted, has helped the retailer “understand better what customers are looking for by season, by session, by time of week.”
AI search and Walmart’s AI shopping assistant are also driving impulse sales, McMillon said, because they make “cross-category search” more effective and show customers more higher-margin general merchandise items.
News built for finance pros
CFO Brew helps finance pros navigate their roles with insights into risk management, compliance, and strategy through our newsletter, virtual events, and digital guides.
Setting the stage for gen AI: When generative AI emerged, Walmart was ready for it. “We were among the earliest companies to build generative AI” into their infrastructure, Bhardwaj said. The company trained existing large language models to be “specific to the retail industry and certainly custom to Walmart’s needs,” she said. It made these models available in Element, which became a safe “playground” where developers and engineers could start “experimenting with their specific use cases” for it, Bhardwaj said.
The ”experimenting” takes place within a broader strategy. (“We don’t want to just throw spaghetti at the wall,” Bhardwaj said.) Developers, she said, aim to meet one of five objectives: enhancing customer experience; associate productivity’ developer productivity; improving operations; or generating content.
Ensuring accuracy and utility: Walmart continuously evaluates AI applications for ROI, Bhardwaj said, with a version of the proprietary calculation the rest of the company uses. Teams identify metrics they intend to reach at certain “checkpoints,” she said, and course-correct if projects aren’t going as planned.
Walmart also monitors its AI models to be sure they’re producing accurate results, Bhardwaj said. Her staff regularly tests the models to keep tabs on “drift”—or the phenomenon whereby generative AIs become less accurate over time—comparing answers the AIs currently give with ones they produced when they were first launched, she said. The company has other controls over its AI as well, including A/B tests of its outputs, and human feedback such as upvotes, opinion polls, customer complaints, and employee reports.
Future capabilities: The retail giant has other generative AI applications in the works, Bhardwaj said. She added the company is exploring “ultrapersonalized shopping assistants” that could, for instance, sense which member of a family was using a Walmart+ account while shopping online. And it’s working on a new assistant that would summarize answers to sellers’ questions, saving them from having to sift through articles for the information they’re looking for, McMillon said.
The company will certainly continue to invest in AI, Bhardwaj said, adding that she prefers to invest in technologies that provide a solution to a specific problem.
Machine learning and generative AI are “very, very powerful technologies,” she added. “We’re still just scratching the surface.”