Walmart's CTO places bigger bets on generative AI as customer shopping habits evolve

Walmart’s CTO places bigger bets on generative AI as customer shopping habits evolve

Posted by Admin October 16, 2024 Shopping

Over a few decades, consumers have moved from shopping exclusively in brick-and-mortar stores, to online, and more recently to spending billions annually through mobile apps.

Now, consumers are changing their shopping habits again. Walmart calls the new reality “adaptive retail,” which is more about personalized shopping, an expectation that retailers will anticipate what consumers want before they decide themselves, and shoppers buying specific products based on what they see influencers pitching on social media.

This evolution of consumer thinking has led Walmart to invest in newer technologies, including generative artificial intelligence. “Shopping is not just a blend of core retail and e-commerce,” says Hari Vasudev, executive vice president and chief technology officer of Walmart U.S. “It’s really a much more fluid experience.”

As Walmart sees it, online shopping has been narrowly focused. Consumers would enter a few search terms to find the exact television or jeans they were looking for. Online retailers were tasked with selling goods at a desirable price and offering efficient and speedy delivery.

But shoppers are getting more loquacious. They’re asking more detailed questions about products, and they want to use new tools, including 3D that can visualize furniture in their home, for example. According to Walmart, they also increasingly want to shop with the help of virtual reality to see what clothes they want to buy look like. Meanwhile, social channels like TikTok and live streaming services including Twitch provide inspiration for shopping. “The search bar is no longer the way to shop,” says Vasudev.

That’s led Walmart to pivot by announcing this month that it’s starting to use predictive models and generative AI to offer a personalized version of walmart.com for every American shopper. The feature will fully roll out by the end of next year. “The way we’re thinking about applying AI and generative AI is to offer highly curated experiences and content,” says Vasudev.

The goal is to create a personalized experience for each shopper based on what they’ve bought previously or searched for online, matched with what’s been a strong seller for Walmart or trending on social. Location and time of year would also factor in. A shopper who’s previously bought goods for their dog will be shown pet costumes for Halloween. Customers in Minnesota and Arkansas will be marketed new sweaters at different points in the season, based on when winter hits.

Walmart’s generative AI efforts have thus far focused on five areas: the customer shopping experience, improving how in-store associates do their work, operational areas like supply chain and merchandising, content generation, and making software developers more productive. AI-focused tools Walmart has launched in the last few years to help employees save time have included “Ask Sam,” which lets associates ask questions like where to find a specific item in a store, and MyAssistant, a tool for corporate employees that can summarize large documents and write drafts.

In most instances, Walmart is favoring “build” over “buy” when it comes to generative AI. The company centralized machine learning capabilities in a proprietary platform, called Element, so tech staff could experiment with AI models without having to worry if they would fit into a specific vendor or cloud provider. Walmart is using a blend of in-house large language models, proprietary offerings from OpenAI’s ChatGPT and Google’s Gemini, and open-source models.

This month, Walmart also unveiled a generative AI-powered customer support assistant that relies on retail-specific large language models. The generative AI, Walmart says, will enable the chatbot to anticipate customer questions and offer more proactive solutions for handling returns, exchanges, and other problems that may arise. And in some cases, the new AI agents are proving to be twice as fast as the humans, according to Vasudev. Throughout the development process, Walmart says finding the right conversational tone was critical.

“When we have something like a chat experience, you want to make sure that the quality of the conversation feels empathetic and feels very genuine,” says Vasudev.

John Kell

Send thoughts or suggestions to CIO Intelligence here.

NEWS PACKETS

Adobe gets more creative on AI pricing. Adobe says the company’s new video-generating AI product will be priced differently than other AI tools, with CEO Shantanu Narayen acknowledging that the cost to produce video is “a little bit more expensive.” It’s the latest indication that tech firms are tinkering with their approach to pricing while also racing to develop and launch new AI products. Bloomberg reports that until now, Adobe hasn’t charged for the use of the software company’s AI features beyond standard subscription fees. And while users are allotted certain credits for using AI, Adobe isn’t enforcing those limits for most plans.

Amazon, Databricks ink AI chips deal. Amazon.com and Databricks have agreed to a five-year deal that will let the startup use Amazon’s Trainium AI chips to power a service that would let companies customize an AI model, or even build their own. The Wall Street Journal reports that the deal will let Amazon continue to position itself as a neutral provider of AI technology, while Databricks makes money from cloud-based AI software and other services from companies building their technology tools. Databricks and Amazon already work together, as Amazon Web Services is one of Databricks’ cloud partners, and Databricks also rents Nvidia’s graphics processing units through AWS.

TikTok, Boeing announce layoffs for very different reasons. ByteDance, the parent company of TikTok, said the social media platform is laying off 500 employees, most of whom moderated content on the service. TikTok says that 80% of content that violates the company’s guidelines is already removed by automated technologies, but the latest cut is an even greater shift to using AI for content policing. Separately, Boeing is cutting 10% of its workforce, or 17,000 employees. The aerospace company also announced it would end production in 2027 of the 767 freighter, which has faced setbacks for years related to questions about the company’s quality control and safety precautions.

Tesla’s robotics and AI event disappoints; CIO departs. Tesla’s Cybercab event, aimed at persuading investors that the electric vehicle maker is also a leader in AI and robotics, failed to impress Wall Street. Much of the criticism fell on CEO Elon Musk for failing to share any specifics about the timeline or operational rollout of the self-driving Cybercab. Meanwhile, Bloomberg and other outlets reported that humans were remotely controlling some capabilities of the Optimus robot prototypes that had initially made a splash online when they were unveiled last week. Ahead of the event, Bloomberg also reported that Nagesh Saldi, Tesla’s chief information officer, is leaving the company. Saldi had reported directly to Musk and had been involved in building out new data centers in Texas and New York.

ADOPTION CURVE

Increased efficiency is a top genAI aspiration. Executives surveyed by MIT Technology Review Insights, along with business technology provider Snowflake, found that 72% were prioritizing increased efficiency and productivity from their generative AI efforts, far exceeding their hope for increased revenue (30%) or reducing costs (24%). Increased market competitiveness ranked second at 55% in the survey, and product and services innovations and improved customer satisfaction were also seen as more valuable than boosting the top line or trimming expenses.

The poll of around 275 business leaders also found that 53% say their data foundations were “somewhat ready” for gen AI, meaning their data has been properly gathered, aggregated, stored, and is accessible to support generative AI applications. Twenty-two percent were more confident, saying their data foundations are “very ready.” The remaining 25% felt either “somewhat” or “very” unready.