Inside the development of Erica, Bank of America’s AI-powered bot

Brands talk a lot about artificial intelligence and its potential to change the way they interact with customers, but few really understand what exactly that potential is or what it takes to make it real.

However, when Bank of America got a sense of the vision for its AI-enabled “digital assistant,” called erica, it didn’t take the bank long to gather the resources necessary to make her real.

“We realized that to do what we wanted, there would have to be a huge investment of time, energy and resources to make it happen,” said Henry Agusti, digital banking executive. the bank’s digital banking executive. “We committed to doing that very early on — about 9 or 10 months ago.”

The bot is now in the beta testing phase.

Since last fall, Bank of America has pooled together a team of more than 100 people dedicated to building erica, a voice- and chat-driven product designed to help customers manage their financial lives using predictive analytics and cognitive messaging. The “vast majority” of the team are people who were working at B of A before it started the idea of erica, Agusti said, since a lot of the functions designed into the digital assistant will be deeply integrated with a lot of the bank’s existing back-end systems.

Erica is really like a personal banker. Only she takes the relationship to a whole new level of personal; she’ll have all of the customer’s data along with thousands of other machine learning inputs that will give her even more information about the customer. In the near term, customers will be able to ask questions, execute transactions and look at balances.

As erica learns from those transactions, she’ll be able to offer more to customers, like insights and advice; and from there, she can start to give advice more proactively instead of merely reciting the information she has.

“In the next 18 to 24 months, the path is about how much functionality we can give erica,” Agusti said. “We really want to make sure she’s able to help clients and customers execute on a lot of functionalities from day to day. That requires a deep level of [technology] integration.”

It’s a far cry from the rest of the industry: While 85 percent of financial institutions surveyed by Celent agree that AI will have a significant impact on banking in the next three to five years, 32 percent indicated they’re making any direct investment into AI technology. Half of the banks surveyed indicated that the expense of integrating AI is a major concern.

Bank of America declined to give details of its financial investment in AI or in erica.

“It looks like an aggressive timeline, but it we still need to see the finished product. Advanced AI or natural language processing skills take time to learn and mature before they are ready to go live,” said Celent analyst Stephen Greer. “Banks using even the most advanced AI like Watson need time to train it.”

Erica will be available to the bank’s associates in a few months, and in November she’ll offer the ability to provide real-time insights and advice, like flagging a dip in a customer’s FICO score or suggesting payment plans based on changes in subscriptions.

“We teach erica how to deal with one customer request but we’re finding customers are asking for other things we never even imagined,” Agusti said.

The work is as much on the front end as the back end, Agusti said. It’s a common criticism of chatbots and other forms of personal financial management services that they don’t do much more than give customers different entry points to their transactional histories — does anyone really need five ways to check their checking account balances? The same can be said of fintech more broadly — that most new offerings are front-end solutions, but it’s the back-end of banking that needs an upgrade.

Banks surveyed by Celent, however, indicated they aren’t expecting more than 30 percent of their front or back office operations to be handled by AI.

“Erica won’t necessarily create a conversation from scratch. We have to help her understand: when customers say this, here’s what she needs to do. Theres a certain amount of back end work we can do to identify what conversation customers want to have with her and if she’s not prepared to deliver, we need to make sure the right conversation would be in those situations.”