¿Cómo se Dice

We all have our own languages. There's Spanish and French and English and some 7,000 more. There's the language of Walmart, where we have Rollbacks and mods and Every Day Low Prices (EDLP) and Great Value and the 10-foot-rule.

In every one of these cases, there's something to note: words are grounded in context, history and experience. And beyond that lies their meaning. For the people worldwide who shop Walmart online or make use of a host of Walmart apps, this was once a problem. It's quickly disappearing.

The Walmart Translation Platform (WTP) got its start with Spanish search in 2022 and since then, has been systematically naturalizing search for non-English-users across our entire catalog. It's a gargantuan task. But led by people, and powered by AI, it's changing the game one word at a time.

Whether a shopper is in Bentonville, Mexico City, Montreal or Santiago, WTP ensures their search intent is understood. Operating at enterprise scale - across millions of products, reviews, and customers - it blends the precision of AI with the cultural fluency of human expertise.

So, how does it work?

There's more than one way to say t-shirt

To understand what's happening behind the scenes of the WTP, some examples might be instructive. At its core, the platform is going beyond literal translation to understand intent in context. Imagine this:

"A customer types in the search bar a product or an item they want to find, but they type it in Spanish. Sometimes it can be a Spanish dialect from Colombia, a Spanish dialect from Mexico, or Puerto Rico or Chile - different variations for different nationalities," said Arpitha Shetty, director of localization, content and accessibility at Walmart International. "So what we do is we adapt that local Spanish, that particular dialect, to the catalog in English so a user gets the best possible experience. We translate their intent, to get them what they want."

Now let's drill down, taking a look at some real problems and their solutions.

Sometimes, things that sound simple actually aren't. Such is the case in translating the humble t-shirt. In Spanish, you might say camiseta, and expect it to be widely understood. But regional variations change how people describe the item. It's the soda is pop is Coke spectrum of middle America: In Mexico, people are likely to say playera, while in Chile you're more likely to hear polera. Each a different word for the same thing.

The WTP understands every one of these nuances and maps the variations back to the same entry, so customers find what they're looking for - no matter where they're searching from.

In Canada, a similar example exists. But this time, it's in French.

Imagine a customer in Quebec is searching for a particular brand of yogurt. They may type yogourt liberte, in search of Liberte Yogurt - which is a brand, and not freedom yogurt, as the translation would suggest.

But with the translation platform in place, linguists, coders and agentic AI join forces to identify, contextualize and map millions of these instances to the products customers are actually trying to find.

Other such examples abound.

A t-shirt bearing the likeness of Lola Bunny and the words "Space Jam" was translated to Spanish as Space Marmalade. Hot Wheels were translated to ruedas calientes, which means something like very warm wheels or even hot-as-in-good-looking-wheels. A pair of sweat shorts, which must exist in opposition to sweatpants, were translated as shorts made of sweat. You can see the need.

Every one of these instances was flagged by an AI agent, raised to a linguist and re-translated properly. Then, the linguist worked with data science and computer engineering teams to instruct Walmart's in-house AI how to stop it from happening again.

Ensuring context and making our items shoppable for everyone is essential. The team responsible says this is the only beginning - a future state could expand to images, audio and video. The goal? Creating a truly inclusive experience at scale.

And all this progress makes a business case too.

"The Walmart Translation Platform is an incredible piece of technology, and it's also got the benefit of being highly efficient," said Tim Simmons, senior vice president and chief product officer at Walmart International. "Beyond the efficiencies we've gained employing agentic AI, we're creating a better overall experience with the focus on cultural adaptation throughout the whole tech-stack. Running the new Walmart Translation Platform costs around 1% of the previous system. We're saving more than $20 million, every year."

A team effort

There are multiple things that make this platform unique, but a couple stand out.

1. Domain adaptations: We've pointed to this already, but it deserves a specific mention: WTP delivers domain-adapted translations. That means its AI is trained on Walmart-specific problems to create Walmart-specific solutions. But don't take it from us. Leo Lezcano leads the data science team behind the WTP and explains it well:
   
   "In addition to standard machine translation benchmarks, WTP uses Walmart-specific ecommerce data," Leo said. "Generic systems don't understand that 'ropa vieja' is a Cuban beef dish - not 'old clothes.' Or that 'kitchen island' should be translated as 'îlot' in French, not 'lot,' which means 'batch,' and points customers to the wrong products. We measure success not just by accuracy, but by whether customers find what they need. That's the power of domain adaptation - reducing friction and improving discoverability at scale."
   
   
2. Localization experts: The linguists working alongside the program are known as 'localization experts.' Their job is to ensure the AI translations meet the highest standards of quality and cultural accuracy. They're an essential part of the feedback loop that helps to consistently improve the quality of translations.
   
   "We really complement each other: The linguists train the model to be more human, while data scientists make it fast, scalable and technically efficient. We're teaching AI to think like a human translator, where tone, grammar, context and intent can flip a sentence from technically correct to truly right for the customer," said Anna Lavinia Dambrosio, who leads the team of localization experts. "It's the partnership between human insight and machine learning that makes the platform unique.
   
   
3. AI scale and speed: Customer experience isn't just about accuracy - it's about performance at Walmart scale. WTP enables millions of products, reviews, and customer interactions to benefit from recent AI breakthroughs, at a cost structure that is 100× lower than raw LLM usage. That keeps Walmart's business sustainable while keeping prices low for customers.
   
   It also operates with real-time speed: in cases like product search or live associate-to-customer conversations, WTP delivers translations in under 50 milliseconds. By running directly on Walmart's infrastructure - alongside the systems powering shopping and search - the platform avoids costly network delays, and can pack linguist feedback into smaller, task-specific models that deliver faster, more reliable results.

In these aspects of the platform, Walmart's use of technology to optimize the role of its people may never have been clearer. The impact of the work landed the team responsible one of the company's highest honors.

"The Walmart Translation Platform represents the best of what happens when technology and people work together," said Vinod Bidarkoppa, executive vice president and chief technology officer at Walmart International. "Winning the President's Innovation Award earlier this year is a reflection of how this team has taken an incredibly complex challenge and delivered real, meaningful improvements for our customers. It's a powerful reminder that when we innovate with purpose, we can make shopping easier and more inclusive for everyone."

In every market where Walmart operates, we believe it should be easy and natural for our customers to save money and live better. The WTP is living up to that promise, making the language of our customers and the language of commerce one and the same.