A deeper look into what makes a successful chatbot with Bank of America’s Erica

Shehzil Zahid contributed to this story.

During the pandemic, customers migrated online and banks shuttered their branches — quite a few closed permanently. Simultaneously, banks enhanced and expanded their digital and remote services to better serve their customers. Most in the industry expect the digital migration to continue into the future. An Accenture study found that 72% of banks expect human-chat interactions to increase in the next few years.

Banks are looking to bolster existing digital channels like chat as the move to digital continues. While the primary purpose of a chatbot is to help customers meet their banking needs, chatbot personality can inspire trust and faith in banks. In lieu of human connections, chatbots can offer the next best thing.

Bank of America’s chatbot, Erica, suggests that AI chatbots can offer more than resolving simple bank tasks. Launched in late 2018, Erica had 6.3 million users and 16.5 million interactions by early 2019. When the pandemic hit the U.S., BofA updated Erica to process more than 60,000 phrases and questions related to COVID-19, according to voicebot.ai. One year later, Erica reported 19.5 million users and over 100 million interactions and recorded 90% efficacy for useful answers from users. 

Hari Gopalkrishnan, managing director for client facing platforms technology for Bank of America, says the key to developing Erica was to understand how machines understand banking. Once BofA established the chatbot’s basic ability to function as a digital channel that can answer customer questions, the next step was to make Erica personable in a way that the chatbot guides customers through a conversation, like a teller or a call center agent would. Gopalkrishnan says that these two factors help Erica be more than just a transaction center.

“We want our customers to feel that we are there for them through their lifetime, financially there for them, watching their finances, and helping them navigate through all the things that life has to offer,” says Gopalkrishnan.

In addition to fulfilling basic banking, Erica gains insights through customers’ transactional behavior with their consent. For instance, an anomaly in a customer’s spending pattern can offer insight and prompt Erica to inform the user about their change in spending. If a customer is spending more on takeout in the last two weeks than she ordinarily would based on her transaction history, Erica will bring the discrepancy to the customer’s attention. 

“[These insights] are a huge differentiator,” says Gopalkrishnan. “You don’t get that at a transaction center.”

SPONSORED

Erica also offers a life plan feature which allows customers to share their long term plans with Erica, who can then offer advice about money management in line with the users’ life goals. Gopalkrishnan says offering advice through insight analytics and helping customers meet their financial goals is crucial in instilling trust and loyalty. This feature helps ensure that customers feel that their finances are being looked after. 

Despite all these features, Gopalkrishnan believes Erica’s personability takes the chatbot’s functionality to another level. Erica’s conversations are more natural — the chatbot makes small talk and navigates conversations proactively.

“It’s really about just having this sense of ‘it’s a conversation’,” says Gopalkrishnan. “It’s got a personality. It’s not just giving you canned responses.”

While functionality and personality are both important to Erica’s success, Gopalkrishnan concedes that if it was a choice between the two, functionality beats personality any day.

“Having something that’s really personable, and does a lot of funny things and offers sentiment analysis, but at the end of the day…doesn’t meet the functional needs — you’re done,” says Gopalkrishnan. “You can’t shortcut on functionality.”

Dan Faggella, founder at Emerj, an AI research and advisory company, fully agrees. He says that so many chatbots have failed because firms don’t understand the technological fundamentals to set up a fully functioning chatbot. For example, Facebook’s chatbot M failed to handle most of its own responses, 70% of which were crafted and taken care of by human responders instead of the chatbot. Similarly, Microsoft’s Tay folded within 24 hours of its launch in 2016 while Google’s Allo bit the dust in March 2019, around two and a half years after its initial launch in September 2016. 

“It is a very hard thing to understand, satisfy and respond properly to a user’s intent,” says Fagella. “Just dialing in intent is absolutely challenging and even doing the rudimentary stuff well is a challenge.”

Fagella also believes that chatbot personality and its benefits are being oversold. He doesn’t believe it adds anything competitive between banks, or that it actually inspires trust and loyalty among customers. 

“Emojis and smiley faces are not winning the game here. The personality of the chatbot is very low on the totem pole factor,” says Fagella.