Why Walmart is developing its own retail-specific AI models
This story was originally published on Modern Retail’s sibling site Digiday.
The world’s long list of large language models now includes a macropod.
On Wednesday, Walmart debuted a new set of retail-specific AI models to help power the company’s “Adaptive Retail” era of personalized shopping and customer service. Called Wallaby, the LLMs were trained on decades of Walmart data with company-specific knowledge about customers, employees, Walmart terminology, its corporate values and other brand-specific data. Walmart also plans to integrate Wallaby with other AI models based on the goals for each application.
Along with Wallaby, other initiatives in the works include a new “content decision platform” that uses AI to understand customers, and a generative AI-powered tool to predict content each shopper would want to see on the Walmart’s website. Walmart also has developed an AR platform called Retina to engage with customers in new online environments using virtual spaces and avatars. Another update includes new immersive commerce APIs, which are in alpha tests with the gaming engine Unity and the virtual world platform Zepeto.
Wallaby follows Walmart’s other generative AI models to build on the retail giant existing foundation for multiple LLMs to assist employee and shoppers. The plan is to release new Wallaby-enabled customer experiences in the U.S. by the end of 2025. Walmart International also will use the platform to power personalized product recommendations in Canada and Mexico. For now, it’s being tested “quite heavily” internally, said Desirée Gosby, Walmart Global Tech’s vp of emerging technology.
“From a consumer perspective, [Wallaby has] a good understanding of our products and what it is that we sell and we serve,” Gosby told Digiday. “An example of that would be having a deep understanding of our brand. We have ‘Great Value’ as a brand within Walmart. And in the generic sense, everybody wants to get a great value. But knowing in the context of Walmart that you’re actually talking about our brand specifically is important to us to make sure we’re communicating correctly with our customers.”
Wallaby joins a long list of AI models named after animals. Others include Llama from Meta, Orca from Microsoft, and Alpaca from Stanford University. Google’s PaLM 2 models also have animal eponyms — Gecko, Otter, Bison, and Unicorns — named in reference to each LLM’s size. Some names are derived from acronyms, but that’s not the case for Walmart.
“We like animal names, so we started with that,” Gosby said. “But obviously we wanted to do something that was relevant also to the Walmart name. There’s not a lot to choose from, but Wallaby was a good name and that we all sort of settled on. So not much more than that.”
The AI updates follow many others from Walmart over the past two years. Last year, it debuted a new in-house AI platform called Converse, which helps with customer service and chat-based shopping. Earlier this year, Walmart introduced its Element AI platform that helps with search optimization, market intelligence and other areas. Walmart also already added generative AI to search earlier this year and adding generative AI features for subsidiaries like Sam’s Club.
Generative AI tools for e-commerce are increasingly an area of exploration for retail giants and startups alike. Walmart rivals including Amazon have been rolling out a range of new tools for advertisers and customers. In fact, also on Wednesday, Amazon announced a new tool powered for shopping guides.
According to a recent report by McKinsey, 71% of CPG leaders reported adopting AI in at least one business function in their organizations in a 2024 survey, up from 42% in 2023. Meanwhile, the report also found generative AI use-cases could increase the economic impact of traditional AI for CPG companies globally and unlock and additional $160 billion to $270 billion in profit annually.
“Retailers typically have data about past promotion performance but may not have the tools to synthesize the data to build a clear picture of how to target promotions,” according to the report released last week. “Digital and analytics tools can leverage historical promotion and stock data to inform the extent and timing of future promotions.”
Other martech companies released their own M&A or funding news with info about retail AI, including Zeta Global’s acquisition of LiveIntent earlier this week and Vizit’s funding news announcement from last week.
Meanwhile, startups like Vantage Discovery are helping smaller online retailers’ overall online search features to improve semantic understanding and personalization. Vantage Discovery is also exploring ways to use multiple AI models to combine generative AI results with a separate model trained on a company’s click data.
“We think that search recommendations and personalization all fit in one platform,” Vantage co-founder Nigel Daley told Digiday last month. “We find retailers right now are using different products for those, but they use the same data. They really should be the same platform, and so what we bring to retailers is a unified platform for search and recommendations.”