The product and engineering teams at StubHub continually blow my mind. Not only do they keep our platform at the leading edge of live events, but they also push the envelope in terms of new ways consumers can find great things to do.
Their most recent innovation was the StubHub Skill for Cortana. Unveiled this week at Microsoft’s Build Developer Conference, the StubHub Skill lets event-goers discover sports, music and entertainment options through easy voice commands.
StubHub wasn’t the only partner to be announced at the conference. But our key differentiator lies in how StubHub’s engineering team developed the StubHub Skill to create the most conversational experience possible. The team took inputs from a variety of sources that other companies don’t have – our StubHub for Skype chatbot logs, customer service calls, and especially real-life interactions between customers and StubHub’s Last Minute Service (LMS) center managers – and trained Cortana to learn from those scenarios.
For example, we’ve been doing this long enough to know that when fans reach out to our customer service team or visit a StubHub Last Minute Service location, there are a few common questions that generally come up, such as, “What should I do tonight?” In building the StubHub Skill, our team knew they would need to create an elegant solution to this and other common inquiries, training our Skill first on the basics, and then on the more advanced options, to make it easy for our customers to immediately get the information they need.
Much like a concierge, the team knew we had to help consumers get to their options quickly, and bring the same high-quality experience that they would receive from engaging with a service representative to Cortana.
The team led by StubHub project lead Jay Vasudevan and engineering lead Pablo Flores used the terabytes of customer interactions that we have as a baseline for determining the key conversational structures and back-and-forth dialogue that usually take place. Then, by applying natural language processing (NLP) and forms of AI, the team taught Cortana to pick up on these initial models and learn from them.
The trickiest part of this process was understanding the context of the thousands of options inherent in live events.
Think about it – there’s a lot that goes into a conversation about sports or music that we don’t even realize. If you and I are talking about basketball, and I mention the Warriors, you likely know that it’s a professional sports team. If you follow the team, the league, or the sport in general, you may also intuitively know the Warriors play in Oakland (for now), are an NBA team, and are currently making their way to another NBA Finals! But the StubHub team needed to train our Cortana Skill to understand those and other details. This required an immense amount of high-level training and applications of NLP and neural network learning, on top of the baseline A
Keeping the Conversation Human
As advanced as AI has gotten – and as much as there’s some concern about technology taking over the world, even a tongue-in-cheek view of it from us! – there are still limits to artificial intelligence that simply don’t exist in human interactions. For example, when you ask a bot about the weather, the response will be about today’s forecast. If you want to know what to expect tomorrow or next week, you’ll have to ask a new question each time.
But in our own daily interactions, that’s not how human conversations work. We don’t have to start from the beginning because we already understand the context. It’s that magic that our team built into the StubHub Skill for Cortana. When someone asks about an upcoming performance, we’ve taught Cortana to prioritize different options, like event location, when responding. It’s similar to what you’d experience in a real conversation – after all, your friend wouldn’t bring up a show in Atlanta if you both lived in San Francisco. It’s the next-level intelligence that makes the difference.
The small StubHub team that built our Skill also carefully considered less obvious nuances in human conversation, like introductions. Many of these learnings came from testing our StubHub chatbot for Skype. When customers first engaged with our Skype chatbot, it started off by asking a question, and we found that a high percentage of users dropped off right away. Because humans are hardwired to be polite, beginning a conversation without any pleasantries was too jarring. Instead, by adding a simple introduction to the start of the interaction, we were able to retain and engage more people. Now, our StubHub chatbots – for Skype, Cortana and even Facebook Messenger – will start a conversation by asking for a person’s name, which creates a level of comfort and familiarity between (wo)man and machine. It’s a small change, but it mimics how we engage with each other as human beings.
Similarly, the team is teaching our chat AI to recognize gratitude. When a customer has a great experience in real life, they are naturally inclined to say “thank you,” and they expect their appreciation to be acknowledged. Now, when a customer thanks our chatbot for its help, it responds with an emoji. That makes the entire experience more authentic.
As we look ahead to what’s possible for the StubHub Skill on Cortana and other messaging platforms, you can expect to see these subtleties and nuances in human conversation shine through. Artificial intelligence doesn’t have to be fake; the real potential is in creating interactions that are as natural as possible.