The request from a customer, “I want to meet with a banker,” seems easy enough to understand and schedule. But it’s a demand that comes with thousands of different variables for Hari Gopalkrishnan, managing director of client-facing platforms at Bank of America Merrill Lynch, who develops tools for consumers and wealth management customers. Where do you want to meet, for example? Or what type of banker do you want to meet with? Is it for business or personal reasons? Is it for a loan? What type of loan?
While these wrinkles are hashed out quickly in conversation, Gopalkrishnan doesn’t have that luxury. Instead, he and his team are developing Erica, Bank of America’s first customer-facing chatbot. Expected to hit the market in the fourth quarter of this year in the U.S., Erica will reside on the company’s mobile application and will need to answer such a request without scheduling a customer in New Jersey to a meeting in Arizona.
In order to ensure the virtual assistant can understand the meaning behind the question, “you need a lot of intelligence,” says Gopalkrishnan.
Artificial intelligence (AI) and robotics have become the buzzwords of the technology sector. As the ability to cull and organize large swaths of data grow, so do the possibilities that banks could use AI or robotics to deliver a myriad of functions, including implementing fraud controls, offering employee resources and even providing customer service. The technology is different from automation in the sense that the software or machine actually learns from the mistakes and experiences it has with customers. That’s the end goal of the AI push.
But the technology isn’t impacting bottom lines yet. Consulting firm Celent surveyed banks to determine what technologies are most important to the function of the business, and not surprisingly, mobile technology and omnichannel delivery came in at the top, no matter the size of the bank. Artificial intelligence, on the other hand, came near the bottom with just 7 percent of banks with assets between $1 billion and $50 billion saying it’s of importance for today’s sales. For smaller banks, the issue is practically ignored.
“Outside of the biggest banks, they have so much stuff on their plate,” says Daniel Latimore, senior vice president of Celent’s banking practice. “[Small banks] don’t have the resources…to engage in it.”
Luckily, smaller banks don’t have to become early developers. Instead, they can wait for others to refine the technology and then implement it through third parties, adds Latimore. For larger banks, like Bank of America, AI creates an opportunity to better service customers with tools that work 24 hours—nonstop—and can cut lead time for certain requests from days down to a few minutes. And they have the resources to make it work. But it comes with significant risks as these technologies become customer facing. Microsoft learned this lesson the hard way when it introduced its Twitter chatbot Tay a year ago. Tay could mimic natural language and adapted the more it was used. But Twitter trolls fed it with racist and misogynist comments, which Tay copied. Microsoft had to pull the bot down a day after its public unveiling.
Erica is an example of banks’ early efforts in the artificial intelligence space. As a chatbot, it must learn and adapt to the natural language of customers. The machine must figure out different contexts, like the search for the banker, but also different colloquialisms or norms. For example, it must understand “twelve, eighty-seven” versus “twelve bucks and eighty-seven cents,” if a customer disputes a charge.
When introducing such a tool, though, Bank of America can’t turn off its customers to the technology on first use, even if they’re continuing to test and improve the technology. For customers, it’s not a test. Answers that inadvertently hurt their finances or fail to make sense could sour them on the service so that they will “never engage again,” says Latimore.
Gopalkrishnan views it more optimistically, pointing to how his grandmother adapted to technology. While his grandmother didn’t have much use for a phone, once she saw she could talk to her grandkids via a tablet—and see them clearly—then she became a user. AI, Gopalkrishnan believes, could “help unlock capabilities that people weren’t using” for mobile banking.
Bank of America isn’t alone. Capital One launched a chatbot named Eno in March, while the Royal Bank of Scotland and BBVA both went public with their virtual assistants last year. In the early forms, it’s difficult to tell how much the chatbots are “learning” from what they hear, or if they’re developed with a large script that provides the bots with different ways to respond. But in order to improve the tools, and actually have machine learning take hold, the banks need reams of data in the form of communication with users. These early tests provide that feedback and create backlogs of information for use as the technology improves.
“The more data that you can feed it, the better off you are,” says Latimore. “Larger institutions with greater data flows have an advantage.”
This early stage AI is seen in other parts of the bank as well. BNY Mellon, for example, uses a number of different robotic process automation (RPS) bots. Hardliners in the machine intelligence space would not consider these tools AI, says Deloitte’s director of technology and enterprise Thibault Chollet, since they’re running off an autonomous script, as opposed to learning. But these tools can be adapted as the technology improves, and they’re currently taking over tasks previously done by humans. KPMG estimates that the use of RPS bots at financial service firms could reduce the need for offshore jobs, like outsourcing clerical work, by as much as 75 percent.
One example of BNY’s use of the technology is in clearing trades. Each day, BNY receives around 250,000 transaction orders. While the vast majority goes through automatically without a hitch, about 15,000 orders are held up every day because account information is incorrect or another error arises. Employees used to have to manually clear or approve the order, a tedious job, says Doug Shulman, global head of client service delivery at BNY Mellon.
Now, the bank uses an RPS feature that helps clear the transactions or sends the request into a separate queue so the transaction can be finalized. In the future, if a machine-learning aspect was added, it could theoretically catch the mistake as it’s happening and inform the person submitting a transaction request that there may be an error, reducing the number of incorrect transactions.
Yet, even though BNY Mellon has over 200 such RPS bots in development, Shulman admits, “AI is in its early days.”
The potential, however, has become clear for those early adopting banks.