Are algorithms the new stylist? Google joins fashion fray with launch of


This article was originally on a blog post platform and may be missing photos, graphics or links. See About archive blog posts.

On Wednesday, search giant Google took the wraps off, its ambitious attempt to make cyber shopping for clothes and accessories a lot more like the curate bricks-and-mortar retail experience.

The site, which has been in development at Google for the last year, is both a nod to the growing importance of fashion e-tail, and an acknowledgement that searching the Internet for the perfect one-shoulder wrap dress is fundamentally different than shopping for the right SLR digital camera.


“Apparel is the second-largest category for online shopping behind consumer electronics and growing quickly” said Sameer Samat, a director of product management at Google. “And we didn’t believe our experience at Google did a good job at solving some of the soft goods shopping problems.”

While time – and ultimately the bag-buying, pants-purchasing cyber-shopper will decide if is, as Samat calls it “the beginning of the reinvention of the way people shop for apparel,” based on a

recent demonstration, the company’s version of personalized-preference product search significantly moves the needle.

Partnering with retailers (who pay either per click-through or per completed purchase), the aim is to serve up potential clothing purchases the same way Pandora suggests music you might enjoy based on your musical preferences. The first step requires users to either create their own shopping profile by taking a visual quiz (“Which outfit is more your style?”) and answering a list of preferences (including likes and dislikes that include choosing designers, colors, patterns and silhouettes), or to choose from among existing profiles by a laundry list of fashion tastemakers that includes celebrities (Sarah Jessica Parker, Kim Kardashian, Diane Kruger), fashion designers (Diane von Furstenberg, Tory Burch, Derek Lam), fashion editors and style bloggers.

While that kind of winnowing down the wardrobe choices is hardly new –- it’s standard operating procedure at sites like Shop It To Me –- the Google folks say what they’ve spent a good deal of time, and resources, on is refining what kind of results get served up.

That would be about five years In Manjal* Shah’s case. Now a director of product management at Google, he’s the former CEO of Like.Com, a site Google recently purchased. Shah said his company has been tweaking the style-shopping algorithm that’s now part of since 2005. “Shopping for clothes is about discovery and not just search,” explained Shah. “You go into a store looking for one thing and you might come out with nine other things too. We needed to find a way to create that ‘bump into’ experience online. Even though you can only search for something you know you want, you can teach the computer to search for something you don’t know you want.” tackles that problem in two ways –- both of which required “teaching” computers how to be more stylish.


The first addressed what Shah calls “genre.” During the profile-building phase, users can specify their sense of style by choosing between seven choices that include classic chic, boho, romantic and edgy.
“What we did was build a machine learning program that maps each item to a genre,” Shah said, “and hired a hundred fashionistas and local fashion students to tag 50,000 different items -– each item by two different people. And then we trained an algorithm on it.”

Shah said this enables the algorithm to tag –- and then suggest new products in customers’ specified style as they become available.

Another way will try to approximate the “bump into” experience is by using algorithms to suggest other pieces or accessories to complete a given look.

“In addition to computer-learning PhDs and scientists, our team had people who went to fashion school and designers and a stylist who literally wrote fashion rules like ‘heavily patterned dresses don’t go with heavily patterned handbags,’ and ‘vertical stripes don’t go with horizontal stripes.’ She wrote hundreds of these rules to go with our ‘complete the look’ algorithm.”

As if that’s not HAL 2000 enough for you, on top of that the site taps into Google Trends data so if a customer is shopping for dresses in the springtime -– when the word “floral” is trending high, for example –- searches will automatically seed more floral offerings into the mix. If boyfriend blazers or jeggings are hot topics, the offerings served up once a day or once a week by e-mail will reflect that.

So it turns out the debate about artificial intelligence is so last season. As Google gives us the ability to accessorize by algorithm, we’re about to find out if computers can learn a sense of style.


And if it doesn’t, you can always just shrug your shoulders and say your motherboard let you leave the house that way.

-- Adam Tschorn

[*UPDATED 11/22/10 10:06 AM: In an earlier version of this post, the first name of Manjal Shah was spelled incorrectly.]