The Walmart Version Of Search

In its quest to master ecommerce, WalmartLabs built a search engine from scratch to meet the needs of its unique customers. Here’s a look under the hood.

The Walmart Version Of Search

Walmart online is powering up a new search engine built from scratch at its tech toolshop, WalmartLabs.

The tech overhaul is the biggest fundamental change WalmartLabs has orchestrated since Walmart acquired it as a startup called Kosmix in April last year. But a search engine makes sense. After all, search was Kosmix’s core strength, and WalmartLabs’ founders—Venky Harinarayan and Anand Rajaraman–got their start building a product search engine called Junglee, which was bought up and integrated into Walmart’s nemesis Amazon.

Polaris, as Walmart is calling its search engine, bears clear imprints of its legacy. It was built in ten months by a core group of 15, relying heavily on existing tech that Kosmix was working on. Polaris now fields all queries submitted on the website, and now delivers results on mobile devices as well.

Sri Subramaniam

Fast Company spoke to Sri Subramaniam, currently VP at WalmartLabs, to find out what was different, about Polaris. Subramaniam was a chief of engineering at Kosmix when it was bought by Walmart. Before that, he spent four years as the director of search product development at eBay, and like several members of the WalmartLabs team, has serious experience with product search.

Their goal while building Polaris was to change the experience of search on Walmart’s website, which saw about 45 million unique visitors in July 2012 (compared to Amazon’s 103.5 million uniques), according to comScore. “We want it to be more of a shopping guide,” Subramaniam says. “We encourage [customers] to explore in a more visual manner,” because it’s often difficult to guess from a search term what product someone is really looking for. The discovery process at is different from the one on, say, Google. One of the ways Polaris is designed to augment the search experience is by treating search terms as categories. So, a search for “garden furniture” may not serve up results with the word “garden” in it, but would offer up suggestions and options for hardy and rainproof, garden-friendly furniture.

Over the last three months, WalmartLabs has been collecting search and purchase data of visitors to, to figure out how a search query translates to a purchase. (As with other sites that track traffic, no personal information is stored.)

For example, after about two weeks of watching customers type “House” into the search box and tracking which results they clicked on, Polaris figured out that most of them were looking for the TV show starring Hugh Laurie, and not supplies or hardware. Similarly, people looking for “flats” were really looking for shoes, and Polaris ranks those results higher now. In that way, it does share the same challenge of semantic search, such as the ability to tell Apple from apples, with Google.

A key characteristic of Polaris is that it is influenced by the popularity of product on social media. This “product popularity score” is essentially like a Google page rank, Subramaniam explains. A product’s presence on Facebook—-how many “Likes” it has, for example–is part of what determines at what spot it shows up in a search result queue. A product’s popularity on other social sites like Twitter or Pinterest could also one day feed into Polaris’s radar.

The more immediate next step is taking Polaris global. And the first stop will be Brazil, where Polaris is set to roll out in the coming months, Subramaniam says. And even though its founders have left to take a break, other products are in the pipeline for the holiday season.

[Image: Flickr user Walmart]