When Bank of America virtual assistant Erica was first introduced in 2018, it was lauded by industry analysts as the leading example of artificial intelligence (AI)-powered chatbots in banking. But like most technology, Erica is ever evolving. The $2.2 trillion bank is now deploying Erica as a digital assistant to support live chat agents in call centers as well as employees in commercial banking and wealth investment.

This evolution of Erica was made possible by collecting the data that flows it, said Christian Kitchell, AI solutions executive and head of Erica, Bank of America.

“We have a real-time understanding of what clients are trying to do it … where we might have opportunities to enhance or deliver increased function to them,” Kitchell told Bank Automation News. “That will play hand-in-hand as we have live chat agents, also helping whereas we do more in the chat space, we can also leverage that intelligence to enhance Erica itself.”

Erica isn’t the only evolving chatbot. This year saw chatbots morph into virtual assistants for employee-facing support, particularly in information technology (IT) and human resources where bots can easily handle repetitive questions or tasks to include password resets, said Will McKeon-White, a Forrester Research analyst who specializes in chatbots across all industries.

Pandemic spurs adoption

Chatbots are also no longer solely the domain of large banks — spurred by necessity from COVID-19 as banks closed their branch doors combined with a dire need to improve technology to meet customers’ ever-changing digital needs, adoption has moved down-market to smaller financial institutions, experts told BAN.

“Organizations are realizing that they can do things like automate 20% of their search service desk requests from employees and allow people to do something different than automating a few 100 password reset requests in a week,” McKeon-White said. “So that’s why that’s been a growing trend.”

Chatbots can be cost-effective solutions for help desks of 10 to 15 people, where employees are starting to handle a growing number of tickets — more than a few 100 a week, McKeon-White said.

“That’s around the point where chatbots can be really effective … because you’re freeing up humans to do something else,” he added.

Chatbot vendor Finn AI saw increased chatbot adoption from COVID as smaller banks began implementing the technology, Jake Tyler, Finn AI chief executive officer told BAN. Among Finn AI’s clients are $521.9 billion Truist Bank, $55.8 billion ATB Financial, $3.2 billion United Federal Credit Union and Canadian challenger bank KOHO, which has raised $113.4 million over seven rounds. Finn AI has raised $14 million to date, according to Crunchbase.

“COVID has been five years of adoption happening in a year,” Tyler said. “We’re still responding to that. I think what we’ll see over the next few years is just mass adoption of conversational AI.”

Smaller banks, credit unions add AI technology

The heyday of chatbot adoption began in 2016 as vendors with budgets to invest in the technology began targeting the financial services market, McKeon-White said.

Chatbot vendors told BAN said they’re seeing increased adoption from smaller banks and credit unions.

The Dover Federal Credit Union (DFCU), with more than 43,000 members, was among the financial institutions that adopted an intelligent “virtual assistant” this year. The Delaware-based $465.8 million credit union chose Interface.ai for its call center after the pandemic led to call volumes doubling, said Tyler Kuhn, vice president of marketing & digital experience at DFCU.

“I’ll have technology helping me before I get to a human, if I even need to get to a human,” Kuhn told BAN. “This digital system will be able to help them and reduce the wait time and not have to wait on hold for 20 to 30 minutes, whatever the wait time is that day.”

DFCU plans to add Interface.ai to its website and online/mobile banking services, Kuhn added. Interface.ai markets its “out-of-the-box” solution exclusively to banks and credit unions, counting $4.7 billion Gesa Credit Union, with 250,000 members, among its customers.

Simple economics drove DFCU to adopt the AI-powered chatbot, Kuhn said. While he had not previously priced the technology, the bank’s value proposition had changed, he added.

“While the pricing might not have changed, there’s a very obvious value in this technology for us,” Kuhn said “We were apprehensive of getting into it before, we [were] left no choice. We knew this was the answer. It was just that that’s just a value exchange that happened.”

AI-powered digital assistant provider Kasisto is also seeing more adoption by smaller financial institutions. Kasisto’s partnership with digital banking technology provider NCR has helped, said Stephen Epstein, Kasisto’s chief marketing officer. Among Kasisto’s clients are the $499.3 billion Singaporean multinational banking and financial services firm DBS Bank, Mastercard, $1.9 trillion Wells Fargo and $1.3 trillion TD Bank.

For NCR, Kasisto took its AI models and placed them on a multi-tenant cloud, which enables more banks to run a virtual assistant on a single instance of the software. That has led to an economy of scale that reduces cost, Epstein explained to BAN.

“Price has been a barrier to entry for a lot of the small financial institutions to get access to this robust technology and that’s no longer there,” Epstein said. “So the more banks get on the platform, the cheaper this becomes, the more economical this becomes, everyone’s sharing and the overall costs — big banks, small banks and medium-sized bank — the economic dynamics have changed dramatically in our space.”

Finn AI also changed its pricing to appeal to smaller banks and credit unions, adding three levels of AI-based conversational chatbots in June. The first level is a “quick start” chatbot for small- and mid-sized financial institutions that runs visitors through a public dotcom site for live support to common questions; it’s designed to launch within a month. Level two is for mobile and online banking sites and offers a more concierge approach to help customers. Level three can run across any channel the bank designates for API integration to digital banking platforms that help customers via chat.

Turning chatbots into ‘virtual assistants’

API integration with banking platforms is key to a chatbot functioning more as a virtual assistant than simply a navigation system, Tyler said. In general, integrating the chatbot with customer management relationship (CRM) systems, data and other engines is what leads to a personalized, “smarter” virtual assistant experience, he added.

“That’s something that Erica It has done a very good job of,” Tyler said. “It’s something that you see most larger institutions today doing with their virtual assistant projects.

“For sure, it’s what I think most people will have — not a chatbot, but a virtual assistant, in the future and that’s what the virtual assistant will look like. It’s certainly where our product is going,” he added.

Meanwhile, companies are starting to combine chatbot technology with robotic process automation (RPA) bots, an emerging trend that will continue to help chatbots become even smarter, Forrester’s McKeon-White said.

“I think that’s especially relevant to the financial services market, because many of them have adopted an RPA platform for some internal process automation,” McKeon-White said.

A use case for combining the two might be to query a backend system or mainframe that doesn’t have a common API to connect it.

“Instead of sending this task off to a human to do, [the bot] can then instead replicate the movements on a screen to go and query a specific system that isn’t integrated into the rest of the environment,” McKeon-White said.

Often, people focus on the “chat” in chatbot, but it’s important to remember the “bot” adds the action and automation, McKeon-White explained. While bank functions can be too personalized for chatbots, he does see increased automation on the backend accelerating banking processes while reducing employee workload and paperwork.

“It’s doing the language processing, it’s trying to understand what the heck people are saying to it,” McKeon-White said. “But if you were doing something like implementing this in a support circumstance, if it can’t take action, if it doesn’t have the bot component of it built out as well, it’s just going to kind of stare at you and you’re going to have the same problem as if it wasn’t able to understand you.”