I recently had the honor of speaking at EmTech Digital, MIT Technology Review’s signature conference on artificial intelligence (AI) and business leadership. This isn’t my first time speaking at the conference, but it is my first time doing it in person.
The EmTech team asked me to present in 2020 at their inaugural event when the pandemic forced organizers to “go virtual.” That seems like ages ago, and I was excited to return to a face-to-face dynamic. It’s amazing how quickly we forget the rush of excitement (and nerves) that comes when presenting in front of a group of peers, especially when talking about one of my favorite topics, innovation.
Equally as exciting as innovation is giving people a look into a Wayfair they don’t see in the commercials or when shopping on our site, Wayfair the technology company. At EmTech, I focused on how we’re improving our business processes with AI. More specifically, I touched on some of the investments we’ve made in machine learning (ML), the thought that goes into how, when, and where we use ML, and tied those to examples that the audience could relate to.
When it comes to deploying ML, Wayfair has a framework that helps us determine which problems are best suited for algorithmic automation. In this article, I will introduce you to one of the not-so-obvious lenses we use to analyze ML problems - risk tolerance. With risk tolerance, we apply ML to problems that can tolerate what we like to call inherent uncertainty in predictions. Another way of putting it is those tasks where A) being mostly correct is ok and B) we have a clear path to improvement through feedback loops.
At the event, I gave an example using search. Search is a great candidate for automation because we have a good volume of relevant data that is easily accessible. With search, the risk of faulty predictions is also tolerable (as opposed to the levels of risk in, let’s say, fully autonomous driving where there is zero room for errors).
Here’s an example. A customer is searching for a blue sofa and using ML, and we pick out and rank 200 blue items to display. If the customer selects the 5th ranked sofa, it’s clear our ranking was not 100% perfect. If it was, the customer would have selected choice number one. But, we were still able to automate the search through our vast catalog and present results that were successful enough. In other words, we were still able to quickly connect the customer with what they wanted.
Another area where tolerance is high is marketing. At Wayfair, marketing is fully optimized and automated. One reason for this is that we can tolerate and manage the risk involved. At EmTech, I gave an example around our ad bidding. Wayfair enters tons of auctions every day. Our goal is to submit a bid that propels the ad to place high enough in Google’s search ranking to drive a customer to purchase while also optimizing our ad spend. We win if the purchases derived from those bids exceed a certain threshold ROI. If we fall short, we have levers in place that we can use to control and mitigate. There are many more examples like this.
There are also areas where ML is not fully automated. The example I like to talk about is our product information extraction process, where we have a lower tolerance for faulty predictions. While ML can help enable and augment rich product information, we have additional processes and workflows as well as team members in the loop to mitigate against possible faulty predictions. This helps to ensure customer satisfaction while empowering our suppliers.
As I said, the way we use ML differs from one area to the next, but there is one theme that permeates everything—ML is playing and will continue to play a central role in our long-term success. When used effectively, it helps drive loyalty. Loyalty from customers who know that when returning to Wayfair.com, they will find the products they are looking for and loyalty from our suppliers, who partner with us because we are providing them with a platform that allows them to effectively grow their business.
It was my pleasure being a part of EmTech again this year and finally getting the chance to meet their team and all my fellow presenters face-to-face. I look forward to joining them again and sharing more insights into Wayfair’s technology!
