Chatbots are getting smarter — and nicer, too
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
Customer service chatbots are becoming kinder, smarter and even more helpful, thanks to huge leaps in artificial intelligence.
Why it matters: Chatbots can be aggravating if their formulaic responses don't provide the help you're searching for — what customer service experts refer to as "the spiral of misery," per the New York Times.
- More capable AI promises to make those encounters less robotic with personas that employ "soft skills" like empathy to read between the lines and defuse tension.
- That could make customers happier and allow companies to spend less time answering questions on the website and more time focused on other things.
What's happening: Chatbots, computer programs intended to replicate human conversation, have been around for decades. Now, however, they are becoming far more ubiquitous — and personable.
- New machine learning techniques have made them much better at carrying on their end of the conversation, via both text and voice.
- Improvements in natural language processing mean bots are better at understanding and producing language.
- They can even interpret the intent of a customer's inquiry and analyze what's transpiring during a chat session, explains Bern Elliot, a technology analyst at Gartner.
- For example: "It sounds like you need to make a deposit, and you need to locate an ATM — is that correct?" he tells Axios.
- Some of the most successful chatbots even have human personas like Nanci (GM Financial), Sydney (health insurance company Anthem) and Erica (Bank of America).
State of play: "This is fundamentally a tech-enabled surge in interest," Elliot says. "That doesn't mean they can't go wrong. They can and they do. Design is really important, and picking the right use cases is really important."
Chatbots serve as a successor to the automated phone trees still widely used by many organizations. ("If you want X, press 1... If you'd like to speak to a customer service representative, press zero.")
- Chatbots interpret users' questions and reply from a library of pre-programmed answers.
Yes, but: Conversations can be stilted, and bots are only as good as the data on the back end.
- Apple's Siri and Amazon's Alexa, for example, can search the entire internet to answer questions. But most chatbots are mining their company's own limited databases for answers to customer inquiries.
- The more data they're trained on, the better they are at providing relevant answers.
What to watch: New user-friendly AI products make it easy for non-techies to create custom chatbots without writing any lines of computer code themselves.
How it works: Recruiters at the University of Illinois used startup Juji's "no-code" software to create a custom AI chatbot to manage mounting inquiries about its fast-growing online MBA program.
- They started by loading commonly asked questions and answers into a spreadsheet that Juji turned into a friendly chatbot.
- "In the beginning, it was a little bit slower," Myranda Crist, the university's assistant director of recruitment and admissions, tells Axios. "But we didn't have as much information on the back end. We couldn't possibly anticipate all the questions. As there were more questions, it's gotten more conversational."
The intrigue: Juji's software uses machine learning to infer a user's unique characteristics — are they an extrovert or an introvert, for example — in order to personalize each interaction, founder Michelle Zhou tells Axios.
- The bots are designed to be empathetic, and maybe even tell a joke if they detect unease, she said. "We want to make the person feel like, 'I've been heard'."
Our thought bubble: An efficient chatbot can be a blessing, but it's never going to be easy for customers to feel truly "heard" by a piece of code, even one shaped by AI. And cracking jokes trips up actual humans all the time.
