eBay ShopBot beta, Under the Hood

RJ Pittman
6 min readOct 18, 2016

“We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.” — Roy Amara

From ride sharing to subscription music and video streaming, to the smart home — it’s easy to get excited about the potential of such breakthroughs when they first arrive. However, the platforms, tools, and simplicity needed for broader societal impact often takes significantly longer than expected to develop. These sorts of disruptive technologies are vital to driving global change. Derivatives of these innovations often live on in future iterations or new use cases that become mainstays in the world at large.

At eBay we understand this all too well. We’ve developed some of the world’s most transformative and disruptive marketplace capabilities and commerce experiences. For the past 21 years, we’ve worked hard to make buying and selling effortless, and as a result of these foundational contributions to e-commerce, today everyone expects an experience nothing short of seamless.

This is the inspiration behind some exciting work we are doing at the intersection of artificial intelligence, cloud computing, and commerce fabric.

Artificial intelligence + commerce = highly personalized shopping assistance for everyone

“I’m going with my wife on a camping trip in Tahoe next month, need a tent.”

Today, there isn’t a search engine or a shopping site that understands the totality of this customer need. Yet people walk into their local retail stores asking questions like this every day. The difference in our online shopping and retail experiences is driven by the differences between humans and software.

Break apart the camping trip example and you’ll find rich implications in the language far beyond needing a tent. You’re going with your wife (2-people), to Tahoe (altitude, scenery), next month (temperature) — all of which informs which tent you want.

We can even factor in other relevant facets from our own knowledge base:

These types of insights have the potential to create richer and more satisfying shopping experiences. It’s why the experts at retail stores are so helpful — their wealth of camping knowledge and ability to infer destination-specific needs is what enables them to give you the best possible trip.

The only problem is, not every need comes with its own expert to help you.

E-commerce and search capabilities have become very effective at finding items and information in catalogs and across the web. But what is too often missing is the additional context needed to determine the customer’s full intent. The user is left to do the work, requiring multiple searches with exact keywords to get the desired result.

And because search engines are especially strong with boolean and simple keyword searches, customers have been trained to strip valuable context from a query until it is barely a few words, like “camping tent”, or just “tent”.

The richness of natural language provides a wealth of context and meaning, and when paired with the science of understanding language, deep learning, and predictive modeling we can begin to harness the power of human intent — a holy grail in the field of information retrieval. Paired with the breadth and depth of eBay’s unique selection, we see an opportunity to create a radically better and more personalized shopping experience for everyone. So we went to work on exactly that.

Meet eBay ShopBot beta

eBay ShopBot is a personal shopping assistant that shops with you and for you. Its aim is to help you find the best deals, whether you are looking for something specific or just browsing for ideas.

Our goal is to bring the best of eBay to your fingertips, highlighting the best of our inventory, with a focus on fixed-price items, fast and free shipping, and deals. You can use text, speech or photos to let it know what you are looking for and eBay ShopBot will return with the items that give you most value for your time and money.

Moving fast and breaking things.

Today we are launching a true beta. This means that things will break. This means that the bot will fail you on more than one occasion. And this is ok. eBay ShopBot is at the very beginning of our vision. It has a lot to learn before it is a master of its inventory and a student of its users.

We are going to move fast and iterate often, learning and growing with our users as they help us smarten our AI and improve how we serve commerce within a messaging framework. The end state of this new shopping experience will change a great deal over the next year.

We believe that the future of commerce is distributed. The days of the “build it and they will come” approach are long gone, with people downloading apps less and spending most of their time on messaging and social platforms. We will go where the users are, integrating our personal shopping assistant into the commerce fabric, bringing one billion items to one billion users.

We are starting with Facebook Messenger, a platform that allows us to offer a compelling shopping experience within a messaging app. We will continue to evolve the experience on Messenger, while also adding more platforms, even beyond messaging — to integrate more deeply into people’s lives.

To support deep learning and predictive modeling at this level, we built a new instance of modern tech stack utilizing a hybrid cloud architecture that enables fast, scalable, secure access to consumers virtually everywhere in the world. We now have billions of listings, current and historical, for eBay ShopBot on a public cloud accessible in real time. With this new infrastructure in place, we can experiment and develop new commerce capabilities for our marketplace in the cloud. More importantly, we are significantly improving our engineering and R&D productivity.

This new platform has greatly accelerated the development and training of our AI engine in a very short period of time. It now understands multiple input modalities — speech, text, and images — alongside all-new natural language, dialog management, search, and identity systems. Trained on more than 15 meta categories, 2500 leaf categories, and 70% of our GMV, our AI engine will find the item you are looking for and make meaningful recommendations based on your interests.

Today’s launch reflects the early beginnings of our bigger vision. Powerful new features that personalize shopping “just for you” will come as we get to know more about our customers, always built around a simple goal: help everyone save time and money.

Artificial intelligence has the potential to dramatically reshape the online shopping experience, with breakthrough merchandising and personalized shopping. Paired with the power and reach of cloud computing, commerce can take place where you want it and when you want it. Combined with eBay’s inventory and knowledge-base we see something really exciting and transformative taking hold. Online shopping will be as easy as talking to a friend.

We are moving fast, with a lot more to share and a lot more to launch. eBay ShopBot beta will receive many updates in the coming weeks, and new features just in time for the holiday shopping season to make finding gifts a little more fun. Watch for more updates and stay tuned, we’re just getting started.

Oh, and don’t forget to say “Hi!” to eBay ShopBot :)