Main content

Wayfair furnishes improved customer search experiences with ML tech

Mark Samuels Profile picture for user Mark Samuels May 30, 2023
In a novel, design-led partnership, the two companies are bringing the benefits of programmatic labelling to computer vision.

(wayfair )

Wayfair is using Machine Learning [ML] technology to ensure the products its customers see online look just the same as the items that arrive in their homes.

Specifically, the company is using a specially designed platform, built with Snorkel AI, to boost the quality of the online search experience and to ensure consumers can find the exact products they need. Tulia Plumettaz, Director of Machine Learning at the e-commerce giant, explains:

Our customers are able to find products easier using their own terms. So, think of a very specific example – looking for a cerulean sofa. You’re able to write ‘cerulean sofa’ and that’s what will come back. That notion of describing the goods you're looking for in the words that you use means the conversion and customer satisfaction rates are much higher.

Defining the business challenge

As a home-furnishing specialist, Wayfair often sends bulky items – such as sofas – to its customers. Plumettaz says the aim of the initiative was to take the information the firm gets from suppliers and to use ML technology to ensure the customer gets a good feel through for the product they’ll be receiving: 

There’s thousands of categories that we are trying to sell and for each of these there’s rich information and lots of characteristics. So, we started thinking about whether we could build models that are based, for example, on the characteristics of a sofa. Does it have a chevron pattern? Is it yellow? What is the texture? Are the arms rounded? All the aesthetic details that a customer is going to use to search in their own terms.

It takes a huge of amount of training data to building these models, but at the same time, Wayfair minimizes how much information it collects from suppliers to make things easier for the businesses it serves. The challenge, therefore, is to find a way to build models quickly, cost-effectively and accurately – and that’s where Snorkel comes in, says Plumettaz:

With Snorkel, we’re enabling fast labelling operations, what we call programmatic labelling, to be able to train those models – we’re looking for accuracy, speed of development and cost efficiencies. And that is what we do with the technology.

Wayfair’s engagement with Snorkel is what the two firms refer to as a “design partnership”. Snorkel already had its key product called Snorkel Flow, which is a data-centric AI platform for automated data labelling, integrated model training and analysis. However, this platform is focused on text and Wayfair wanted technology that could help with the programmatic labelling of images. The partnership has allowed the two companies to work together and fill this gap, argues Plumettaz:

Wayfair sells images at the end of the day – you look at the sofa online and you fall in love with the sofa. For Snorkel, they’re getting into a new market that brings programmatic labelling into computer vision. We get to shape the product roadmap and get something that really works for us. And Snorkel has a new product that’s in a space where there's nothing else like this out there.

I almost see them not as a vendor but as a partner on a journey where our paths coincide. It's important for them; it's important for us – let’s go together. And the result is a very unorthodox partnership between a vendor and a commercial company.

The two companies have been working on their solution for more than a year. Consideration was originally given to developing a platform in-house, it was recognized that developing ML technology wasn’t a core business activity for a furniture retailer, so the firm went to market with a lengthy list of requirements. After struggling to find the right programmatic system, Wayfair’s CTO encountered Snorkel at a conference and conversations began.

Delivering the right product

The technology that’s emerged from the partnership is used to cleanse and enrich Wayfair’s catalog. The Snorkel platform uses a data-centric approach that’s based on ML best practices and state-of-the-art foundation models. Plumettaz says the programmatic-labelling technology means it’s easy for subject matter experts at Wayfair to create a new product category and ensure the right images are served to consumers:

With a couple of clicks, our subject matter experts can get a first stab at what the category might look like. A lot of strategic decisions are about trying a new idea. If it sticks, we can automate it, and then it goes into production. So, I can empower our category managers, which means the user is not just a data scientist. In fact, the user could be any person in the business that’s trying to label images.

Now the programmatic approach is embedded in Wayfair, Plumettaz believes that it will be much easier for people across the business to make changes to the catalog when new trends in home furnishing emerge:

Before Snorkel, we needed humans to go through 40 million products and try to label things. Now, with programmatic, we can say, ‘OK, let's re-label our catalog because there are items that we already have in this new style that people are talking about’. So, that is a true value unlock from a business point of view.

Plumettaz also envisages the technology being applied to other use cases. From providing more detailed information to marketing to reducing the product description requirements for internal teams, Plumettaz wants to use its partnership with Snorkel to create more efficiencies and innovations:

Data is the big disrupter. So, that's why I'm putting all these applications in the same bucket because data is the differentiator. I believe that programmatic will become an essential tool if we want to make a difference to our business at scale.

Plumettaz says it’s important to recognise that the technology – if you find an effective partner – is often the easy part of implementing machine-learning processes. Other business leaders who want to create a successful data-led strategy must find a strong use case:

Machine Learning needs to drive business value – that's the key thing. You see a lot of businesses thinking this technology is cool. But very early on in the conversation, you need to think about how this technology is going to make your business better. Where is the differentiator? That needs to be the cornerstone of the conversation.

A grey colored placeholder image