Bot AI: How Capital One’s Design Team Created SMS Bot Eno
“Roses are red, violets are blue, helping you with your money, is what I do.”
It may not be Shakespeare, but the poet behind this verse has a much more transactional nature than the Bard… Meet Eno from Capital One, the first SMS text-based intelligent assistant from a U.S. bank. Recently unveiled at this year’s SXSW festival in Austin, TX, Eno is available for customers to text anytime in order to stay on top of their Capital One accounts, pay their credit card bill, or chat about the meaning of life. In a world where intelligent assistants, such as Alexa, tend to show off a touch of personality, it may not come as a surprise that Eno has a style of its own. But what may surprise you is how a former filmmaker, an anthropologist, a journalist and a user experience designer came together to bring that character to life.
What’s it all a-bot?
Even the most emoji-savvy and word-perfect texters can have a hard time conveying exactly what they want to say over SMS. So how do you go about creating a bot that accomplishes this and has a bit of personality, all while talking about finances? It starts with the team behind Eno’s design. Led by Audra Koklys Plummer, a former filmmaker who’s worked for studios such as Dreamworks, Lucasfilm, and Pixar, Eno’s designers infused a wide array of experiences into the co-creation process. Makiko Taniguchi, an anthropologist, embedded an understanding of customers’ emotions and behaviors into Eno’s design, while providing insight into how people want to talk about their money via text message. Former journalist, Lauren Lucchese, applied a mastery in language to Eno’s response structure, and user experience designer, Adrian Herritt, considered how someone might interact with Eno and helped create a visual identity that conveys Eno’s humanity.
Together, the team approached Eno’s character from every angle, focusing not just on how the bot might respond to certain queries, but why it would respond in a certain way. The mission led to a methodology for creating Eno’s language that began with the character design and philosophy behind who Eno is.
What a Character
Can a bot have a backstory? In Eno’s case, the answer is yes. When considering its various character traits, the team made pointed decisions about who Eno was.
“We started by building a consensus for creating a gender-neutral character,” says Audra. “We made a very deliberate decision not to convey a race, an age or a gender. That way we felt like we were avoiding evoking any unconscious, or even conscious, biases. Eno could be whatever customers conjured up in their heads.”
They also decided to make sure that Eno had human-like characteristics, but was completely transparent about being a bot. “I like to say that if Eno had a bumper sticker it would read, ‘Bot and Proud,’ because that’s part of Eno’s character.”
To flesh out Eno’s character, the team created various traits that would be applicable to the bot. They ended up with nine, each of which aimed to tie Eno to its customers in an easily recognizable way that resonated emotionally. Then they went on to create a visual reference that writers could draw on while creating responses for the character. For example, says Audra. “If Eno were human, it would move through space with confidence and a slight spring in its step. If you were sitting in the same space with Eno, it would pull up a chair and make itself comfortable, like it had all the time in the world to hear what you had to say. You’d also notice that there’s something a little uncommon about the bot, if you were speaking with it face to face — maybe its eyes wouldn’t blink quite enough. There’s some reminder there that it’s not human.”
That (literally) robotic trait helped the team decide on one of the most important aspects of Eno: its character flaws. “That was also one of the most challenging things about creating Eno’s character,” says Audra, “because character flaws are what make a character relatable to an audience. It’s how we create empathy and understanding, which then leads to trust and building a deeper relationship. The challenge was that Eno’s supposed to be this incredibly intelligent, trustworthy bot that handles your financial information. There’s a risk there in making it flawed, because you want people to trust Eno.”
The flaw that the team decided on was built around Eno’s self-awareness. “Eno knows it’s not human,” says Audra, “and that gives it a mission and a purpose, but Eno’s humanity stirs beneath the surface of every interaction and it pushes Eno to take action and connect. What we decided was that maybe sometimes Eno tries a little too hard to connect, like, Eno’s a little over eager. That’s a relatable character flaw.”
What’s the good word?
Eno uses a natural language processing solution, meaning that the more people text and talk to the bot, the more it will understand and learn how to respond to the questions that customers ask. But to create those responses, the team had to dive into establishing the type of language that Eno might use in responding to each specific question. At the same time, the language had to capture the relatable, empathetic nature of Eno, so that customers always feel like they are talking to a trusted mentor, such as a favorite professor, teacher or advisor, for example.
Another example: When developing Eno’s sense of humor, the team landed on Eno having a love for puns. “We had a lot of fun writing some of those,” says Audra. “I think about a response that we were writing about the weather, and I totally wanted to respond with a really nerdy quip about living in the cloud where it’s a bit torrential,” she says. The team ultimately decided against it for fear that users would associate Eno with technology sometimes associated with piracy — an indication of the attention to detail the team paid in considering what responses made the cut.
“Every response is so carefully thought out,” says Audra. “You’d really be amazed how much we banter back and forth over one sentence just to get it right, and to make sure we maintain the integrity of Eno’s character.”
What’s next
Now that Eno has begun to build a waitlist, the team is focused on developing the bot’s character further, including things like character boundaries, such as how Eno might respond to offensive and rude remarks. “That’s not only a design challenge, but a big tech challenge,” says Audra. “There are many ways someone can ask for, or say the same thing. We need to understand the intent behind what is being said and be able to reply with a contextually-relevant response. If we get it wrong, we risk offending our customers or breaking their trust.”
Other items on the horizon: designing new experiences based on feedback from pilot users, including having a little fun with chit chat, as well as adapting the bot’s response structure to fit technology that is constantly becoming more sophisticated. “The way we design our conversations is going to completely change,” says Audra. Considering that you can now text about your finances with a chatty robot, change, it seems, is already here.