Q&A with VP of Engineering at eBay RE: ShopBot

eBay shoptbot is a personal shopper powered by AI

1) How did the conversation about building ShopBot get started?

ShopBot represents a new kind of personal shopping assistant; one we expect will become ubiquitous over the coming years.

Answers by Japjit Tulsi, VP of Engineering at eBay

Consumers today are spending more and more of their time on fewer and fewer apps, with the majority of their time spent in messaging and other social platforms. Looking at the future of platforms in this distributed space, we looked at personal assistants as a unifying theme. eBay Shopbot is one of the first of its kind: a multi modal (text, speech, vision) assistant that shops for and with you. It’s powered through artificial intelligence, which allows us to effectively understand intent and personalize results for the user.

When we launched the beta in October 2016, we were ahead of the curve for retail bot assistants. Going in early gave us the ability to gain insights and continually evolve our ShopBot based on user behavior.

2) What difficulties and challenges did you feel that ShopBot would solve?

When we built eBay ShopBot, we were looking more at the opportunity it could create, rather than difficulties and challenges. Today, human retail associates help to create a much richer and more satisfying shopping experience when shopping in-store. With eBay ShopBot, we wanted to see how we could bring that human interaction into the online world.

Traditional ecommerce and search capabilities are good for finding items and information in catalogues, but they stop short of the additional context needed to determine the true customer intent. This has always been a challenge for e-commerce: the gap between an in-store assistant who understands the inventory and can field your questions, and a solitary experience online where one sifts through inventory without guidance.

That’s the gap we want eBay ShopBot to fill. We invested our AI capabilities to create a true personal shopping assistant that will surface eBay’s best inventory and deals, personalized for each user. Given our breadth of inventory, customer insights, predictive modelling and machine learning capabilities, we thought eBay was uniquely positioned to capture rich context and human intent.

3) Did the decision to build a chatbot mean that eBay had to make some new hires? Was it challenging to find the right people for that task?

In general, we hire engineers who are solid all around; with eBay ShopBot, our standards were no different. As a business in general, we’re increasingly looking for those engineers with a fundamental understanding of AI.

4) Please tell us how you built ShopBot. Did you feed it examples of customer queries to build its intelligence?

We believe in a test-and-iterate methodology. We have built and rebuilt several of components of eBay ShopBot over the course of the last year — what’s more, this learning has been in partnership with our users! Since launching the beta last fall and inviting users to test it out, eBay ShopBot has gotten smarter and smarter overtime. We’ve been looking closely at how consumers are interacting with the bot and continuously improving how our AI is working on the backend.

As we continue to evolve with a steady beat of new features, algorithmic improvements and enhancements, we expect our ShopBot to get smarter and better as we go.

5) Does ShopBot use Machine learning to improve its conversational abilities?

Yes, we are investing in fundamental facets of Artificial Intelligence — deep and machine learning algorithms, knowledge graphs as well as human expertise to drive eBay ShopBot. The bot uses deep learning algorithms paired with natural language understanding (NLU), semantic search and computer vision to help users express their shopping needs in a natural form.

6) Tell us about some of ShopBot’s most impressive conversations and feats of intelligence.

Computer vision is really accurate across brands, categories, color and more. Our personalization features are also impressive: as we learn more about you ShopBot builds personalized recommendations for you that get better and better over time.

7) What is the response that you have had from users to the launch of ShopBot? Do you have any statistics, data which display this?

Our behavioral data has indicated that fashion and electronics are the two top categories users are browsing on eBay ShopBot. We’ve also found that those who use the ShopBot are nearly three times more likely to do so to ask questions on specific products versus browsing eBay’s inventory for inspiration and discovery.

8) How do you plan to develop ShopBot further?

We’re continuing to develop eBay ShopBot and increase the accuracy and consistency of our NLU and computer vision systems every day. A lot of this is effort from our product teams, expanding and improving the bot’s AI and machine learning in the back end, but a fair amount is also attributed to user behavior and engagement. As our users continue to search, browse and purchase on eBay ShopBot, the bot will grow smarter and smarter with each interaction.

Our goal is to develop a relationship with each customer — to create a personalized experience that strengthens the more a user interacts with the bot. This will be achieved through deep learning as well as NLU. In parallel, we’re integrating our latest advancements in computer vision in order to help users find those items they might not be able to articulate in words.

9) What do you think is the future potential for Chatbots in your business, as new technologies emerge?

We want eBay ShopBot to deliver a truly personalized, contextual shopping experience — in its purest form. We want to surface the exact item that you’re looking for, the exact model, condition and price you want it at. With our continuous advancements in AI across NLU, computer vision and semantic search — with the power of massive amounts of consumer behavior and data logs at our disposal, our aim is to achieve that perfection in the backend, so that you don’t have to frontload any of that effort in your search.

It is one of the key endpoints in our quest for the best personal shopping assistant.

10) From your observations, how do you think chatbot technologies will change retail?

Chatbot technologies, when performed perfectly, can bring a marketplace to your fingertips anytime, anywhere, through your mobile device. We think personal shopping assistants will be the standard in the future.

11) Is eBay paying particular attention to AI technologies? If yes/no, why?

Yes, AI is what drives eBay ShopBot and our core marketplace. eBay ShopBot uses deep learning algorithms paired with natural language understanding (NLU), semantic search and computer vision to help users express their shopping needs naturally. As a technology company, we are paying particular attention to AI at its fundamental level. We are investing in building internal expertise and deep knowledge in this domain.

We are also investing in the fundamentals of AI as opposed to just using AI capabilities provided by others. By building our own capabilities, based in eBay’s vast storehouse of data, we have been able to accelerate the use of AI and make it useful to our customers.

12) Do you have any message for retailers who have not launched chatbots?

Every company should be investing in AI, whether that takes the form of a personal assistant or not. Being able to get your customers what they want, when they want it, should be of utmost priority. Repeated accuracy in this vein will encourage more transactions every day and will start to override any traditional pain points of the retail experience today.