Over the weekend, I had the opportunity to participate in a panel discussion on conversational AI with some insightful colleagues from Google, Wells Fargo and Personetics at the Money 20/20 conference in Las Vegas.
Dan Van Dyke, Managing Analyst at Business Insider, moderated the panel and posed a great question: what myths exist related to intelligent assistants? Dan’s question got me thinking, and below are five common misconceptions that jump to mind.
Myth #1: If you build a chatbot, they will come
An intelligent assistant understands context and reaches out to solve problems in the customer’s exact moment of need (proactive). In contrast, a basic chatbot waits for the customer to ask a question (reactive).
Eno, Capital One’s assistant, began its journey as reactive in its early days and has since become increasingly proactive. Eno looks out for Capital One customers, helping them to manage their money through alerts about potentially fraudulent transactions, suspected double charges and important payment due reminders, to name a few.
Myth #2: Diversity is a “nice to have”
Diversity is a must have. Among the many bright people working to build and train Eno are a former Pixar filmmaker, an anthropologist, a journalist and one of the original engineers for IBM’s Watson.
In the early, “pre-release” days of training Eno’s natural language processing (NLP) solution, we had a single person mapping customer utterances to a defined intent (e.g., “Eno, how much cheese do I have?” might get mapped to Eno’s account balance intent). She and others astutely recognized the potential risks associated with training Eno to understand customers solely as perceived by one person. Eno is a product for everyone, and so Eno needs to be able to understand the wide variety of ways people talk about money. We now have people of different backgrounds, genders, ages, and ethnicities building and training Eno’s NLP.
Myth #3: Intelligent assistants don’t need natural language processing
For too long, the financial services industry has required customers to learn “bank speak.” Technology has advanced far enough that we can now design products that teach the machine how to interact with humans, rather than teaching humans how to interact with the machine. A key benefit of the natural language, conversational interface is discoverability. Conversational interfaces flatten out the hierarchy of design, making everything equidistant from the blinking cursor. This enables them to overcome the “bloat” challenges of graphical user interfaces (GUIs) (e.g., Web sites), which must necessarily bury some features several levels deep in a menu. Scripted conversations that force the customer to choose only among provided responses merely move the shortcomings of GUIs into conversations.
The natural language processing technology behind Eno allows Eno to understand customers however they like to communicate without having to conform their questions into bankspeak.
Myth #4: The sole purpose of an intelligent assistant is to reduce the expense of live chat agents
It is reasonable to forecast the potential for intelligent assistants to reduce call center costs. But this shortchanges the opportunity to evolve the way banks interact with their customers. The increasing availability of data and advances in machine learning enable assistants to thread through the entire customer journey, anticipating when customers may “need help” and then reaching out with actionable solutions. Human live chat agents rarely, if ever, reach out to customers because that model doesn’t scale. But intelligent assistants can enable a highly personalized, useful and contextual customer experience at scale.
Myth #5: An intelligent assistant is tethered to a channel
One of my fellow panelists, Sofia Altuna, Global Product Partnerships, Google Assistant, raised a great point during our discussion about how good intelligent assistants operate agnostic of device or channel. We’ve all heard the popular buzz phrase “go where the people are,” and we hear from our customers that they want their assistants to seamlessly integrate into their lives.
We’ve acted on that customer feedback and begun making Eno available to Capital One customers in all our customer touch points: SMS, our mobile apps, logged in on the Capital One website, through email, push notifications, and even at checkout online through a slick browser extension that helps customers shop safely online.
We’re still in early days
At the end of the day, the best intelligent assistants are proactive, anticipating what the user wants, and threaded throughout a customer’s journey. They allow humans to communicate in natural language and are built by diverse teams. We’re still in the early days of development, with many myths yet to emerge and be dispelled as we continue to test and learn our way to truly intelligent assistants. Do you agree with these myths? Did I miss any? Would love to hear your thoughts.