Machine learning helps this company deliver a better online shopping experience

Buying products online can feel like a gamble. Wayfair uses machine learning to ensure the odds are in your favor, and the technology boosts internal productivity, too.
Written by Mark Samuels, Contributor
Reviewed by Min Shin
Person making a purchase online
Westend61/Getty Images

When you're buying a bulky item like a sofa on the internet, the last thing you need is for the product to arrive and for it to look nothing like you expected -- that kind of error is costly and frustrating not only for the person buying the sofa, but also for the company selling the item.

So, what if there was a way to use technology to make the buying experience more enjoyable and to reduce the risk of an unpleasant surprise on delivery day?

Also: This retailer is using RFID tags to make in-person clothes shopping less frustrating

That's exactly what Wayfair is doing with the help of machine learning. The e-commerce company, which sells furniture and home goods online, is using a specially designed platform that's been built alongside technology specialist Snorkel AI.

Tulia Plumettaz, director of machine learning at Wayfair, says the platform is helping her company boost the quality of the online search experience it provides to consumers and to help ensure the sofa you receive looks like the sofa you ordered.

"We have these bulky items that are hard to transport," she says. "We want you to get inspired and feel confident that what you're going to get is what you're buying. And we want that to happen without you even touching the product."

Also: Is Temu legit? What to know about this shopping app before placing your first order

Delivering this kind of high-quality online experience is far from straightforward. Wayfair's site includes thousands of products with a huge number of potential variables, including size, color, and texture.

An added complication comes from the fact that the e-commerce company provides a platform for its suppliers to sell goods to customers. Plumettaz says Wayfair sometimes receives a limited amount of information on products from its suppliers, so providing detailed descriptions to customers can be tough.

It's at that point that Snorkel's platform plays a key role in providing enriched product information. 

"We want suppliers to find that it's easy to work with us using our advanced technology. We want them to say, 'I gave Wayfair a picture, some information, and -- with not a lot of effort -- my item just started selling,'" Plumettaz says. 

Plumettaz also says machine learning supports "fast-labeling operations" through a bespoke solution that's been developed through a design partnership.

Also: The best online thrifting apps

Snorkel already had its key product called Snorkel Flow, which is a data-centric AI platform for automated data labeling, integrated model training, and analysis. 

But while Snorkel Flow is focused on text, Wayfair needed a solution that would support the programmatic labeling of images. 

Plumettaz says the solution, which was developed over a twelve-month period by the two companies in combination, provides benefits for both companies: Wayfair gets to shape the technology it's using, and Snorkel gets a route into a new and fast-emerging market.

"We engaged together, and the result is a novel development that brings programmatic labeling into computer vision," says Plumettaz. 

Also: Here's why everything on Temu is so cheap

Now, with the bespoke technology in place, Wayfair's team can label and re-label products quickly and effectively.

Rather than having to rely on humans to label up to 40 million products manually, automation deals with a lot of the heavy lifting before specialists within the business -- such as category managers -- ensure the right images are served to online shoppers, says Plumettaz. "With programmatic image labeling, we can match products in the catalogue to the items that customers are looking for as new trends emerge." 

Machine learning is also a productivity enhancement -- with less time being spent on labeling images, employees can now focus on higher-value activities. "It's making what we do a lot more interesting," she says. "At Wayfair, our employees don't lack activities to do -- think about maintaining such a rich catalogue. So, now we can be more productive. It's helped make our lives easier and our work a lot more cost-effective."

While Wayfair has chosen to work with Snorkel, Plumettaz recognizes there are other technology players who continue to develop their own machine-learning solutions. 

Also: I bought four brand-name tech gadgets on Temu for work. Here's how it went

She says each company has its own stack and, in such a fast-developing market, it's tough to know where machine learning goes next. Plumettaz advises other professionals who are looking at emerging technology to make early inroads and build strong partnerships.

"The field is moving so fast," she says. "Five years ago, it was a lot harder to integrate with a vendor in machine learning. Now, the hurdles to get a vendor approved are disappearing fast." 

While machine learning can provide a big boon to customer and employee experiences, Plumettaz says professionals shouldn't let emerging technology work in isolation. 

Left to its own devices, an automated system might start labeling products wrongly, leading to unhappy customers and what she refers to as "tremendous consequences". 

"You can have an amazing model, but the noise that can come your way through a 1% error rate -- such as when a bulky item gets delivered to your home and it's wrong -- is huge." 

The lesson for all business leaders is to ensure the human stays in the loop in what remains a nascent area of development.

Also: I bought some off-brand geeky stuff from Temu (and wasn't mad about it)

"It's a journey with a lot of these applications," she says. "Let's automate, but let's still have a layer that is checking that the automation is working."

Plumettaz provides more details about how that process works at Wayfair. "When we're not confident, we put the outputs in front of a human and get some feedback," she says. "I call it the path to automation. It's like a toddler; it's not yet an adult who can run. And that's the framework that we've been using for those kinds of applications."

Another lesson for professionals who are thinking about dabbling in machine learning is to focus on cross-organization integration and processes, especially in terms of how the technology is implemented, used, and exploited.

Plumettaz says the takeaway will be a familiar one for professionals who introduce new systems or systems: Don't just implement technology for the sake of it. "Partnering really closely with business owners and product owners is key," she says. "I think the blocker is less around the technology and more around thinking about machine learning as a business-value driver from the get-go."

Editorial standards