How Walmart builds artificial intelligence tools

Walmart Global Tech approaches emerging technologies like Legos: reusable blocks that can be used to build both Boba Fett and Baby Yoda. Or, in this case, to build tools for both associate productivity and shopping efficiency.

“We want to build reusable capabilities that can easily be composed and plugged into different experiences by the different market teams in different businesses,” said Cheryl Ainoa, Walmart’s EVP of new business and emerging technologies. “Our goal is to identify those core technologies that are going to change the way consumers interact with each other, with their devices…and, therefore, are going to impact the future of retail.”

One example of Walmart’s “Lego” strategy is the retailer’s approach to conversational AI, Ainoa said.

Early adopters: Walmart has been investing in natural language understanding—the ability to recognize human intent and generate responses—for five years. Ainoa and her team are focused on three major AI programs: retail intent, entity recognition, and contextual understanding.

In other words, they’re training AI to recognize retail specific activities (Are you browsing or looking for a specific product? Do you want to see how a rug looks in your room?) to differentiate between products (knowing Walmart’s immense catalog) and pulling from the context in which words are being used to determine user intent.

“Building out in these ways allows our teams to really quickly adopt, scale quickly, and adapt to their specific use cases,” Ainoa said.

Like the shoe: The company’s internal conversational AI platform, Converse, powers a wide range of applications including voice ordering, text-to-shop, customer care chatbots, and Ask Sam, the voice assistant used by US store associates.

Pre-trained models like those created by OpenAI serve as the foundational structure for Converse, and Walmart’s own algorithms add customization, tailoring for specific business needs.

Translation? Converse speaks retail.

“I’d say even more specifically [Converse] speaks Walmart retail,” Ainoa said.

Speaking “Walmart retail” means that, for example, Converse knows that when a shopper using voice ordering asks for “Great Value Almond Milk,” they’re looking to buy a flavor of milk from Walmart’s in-house brand, not a good deal on almonds and milk, she explained.

Big picture: Ainoa said conversational AI is now pervasive across Walmart’s many business functions, and is also being used outside of the US. (In Chile, for example, customer satisfaction scores increased 20% after Converse was added to the service experience.)

“Any place where a customer or an associate is needing information or interacting with a screen is probably a place where conversational AI can help,” Ainoa said.