Contact center agents at Sterling Bancorp in Pearl River, New York, are not on duty to solve customer queries at all hours, but Sterling’s intelligent assistant Skye is.
In fact, Skye, which communicates with customers through conversational artificial intelligence, is the first voice that customers hear when they call. Unlike traditional interactive voice response systems that either lock customers in voice-mail jail or quickly bounce callers to an agent, Skye can read back a customer’s balance or recent purchases and explain why a debit card transaction was declined.
The conversational AI assistant was built by Amelia, an AI software company in New York. It has been widely deployed for a year by the $30 billion-asset Sterling, which said in April that
In that time, Skye has handled more than 2 million customer requests, a number that Sterling said far exceeded its expectations and otherwise would have taken more than three years for contact center agents to tackle. The bot is doing the work of 100 full-time employees, the company estimates.
Sterling’s interest in a voice assistant to answer calls predates the pandemic. Though Skye officially launched in June 2020, Sterling tested the technology among friends and family for nine months beforehand. But its use over the past year has shown that conversational AI is effective at resolving everyday queries without clogging the phone lines, and at picking up the slack during unusual times when calls flood in — as they did after stimulus checks were released.
“A recurring thought for quite some time was, how do we use emerging technology to best service our clients?” said Luis Massiani, chief operating officer and president of Sterling's banking subsidiary, Sterling National Bank. “It doesn’t get much better than being able to offer 24/7 banking service.”
This method of handling customer communications is becoming increasingly accessible to financial institutions of all sizes.
“You go to any banking conference and a dozen different conversational AI vendors are offering different solutions,” said Chris Ward, principal consultant in digital banking at FBX, part of the London-based consultancy Informa. “The availability of the technology has massively grown, and the cost has come down dramatically, so it's not exclusively for the biggest banks and biggest budgets.”
Amelia lets financial instutions and other companies integrate its conversational AI product, which it calls a “digital employee,” into back-end applications. It can be applied to multiple channels, including phone, text messaging and webchat, and personified by each client with the name and likeness of their choice. Amelia’s Digital Employee Builder lets business process experts train the assistant rather than relying on developers.
Sterling chose Skye by companywide vote.
“Amelia is designed to have a conversation the same way humans would do in her place,” said Chetan Dube, Amelia's CEO.
He says Amelia can authenticate customers, pull up their transaction details, parse multipronged requests and offer solutions. When handing a complicated request off to an agent, Amelia will stay on the line to learn how the human handles it.
When Sterling first launched Skye, it was limited to a few uses such as password resets and balance inquiries. The company had tracked the type of inquiries it received to the call center after hours and on weekends and found most were relatively straightforward to manage without a human.
Sterling has since broadened its uses to 12. The most common are balance and transaction queries, troubleshooting online banking login issues, providing reasons for debit card declines, and support for debit card claims.
The company also reports that in several instances, Skye resolved more than one request in a single, unbroken conversation. It was recently trained to handle scenarios where clients seek information about different accounts in the same conversation. In a live demonstration, a Sterling employee spoke with Skye over the course of three minutes to check her balance, hear a list of her most recent transactions, solve login troubles and take initial steps to file a claim over a suspicious transaction. Skye rerouted her to an agent to handle the claim.
Beyond those run-of-the-mill inquiries, Skye swooped in when the contact center was inundated with queries during the pandemic. When the stimulus funds were paid out, many customers called purely to confirm if their checks had cleared and whether their balances were $1,400 or $2,000 higher.
“One of the biggest benefits you see with this technology is when there are event-driven, peak-volume calls,” Massiani said. “We could have never staffed up quickly enough to absorb the volumes we saw with the Paycheck Protection Program, with the fiscal stimulus checks.”
Today, Skye greets the vast majority of inbound calls. It solves close to 50% of the most common customer requests received by Sterling’s contact center, with the majority relating to balance and transaction inquiries. The goal over the next 12 to 24 months is to bump up that rate to two-thirds.
Even if the digital assistant doesn’t have all the answers, “Skye is smart enough to determine, based on a hierarchy of decisions, whether it can address this client need itself or route to the right place or product expert,” Massiani said.
These are two advantages over an interactive voice response system, which can’t solve problems on its own, and where there is no decision matrix efficiently directing calls to a subject-matter expert.
“The client experience is better because even when Skye can’t handle the calls, it’s substantially faster than old technology in pointing the client in the right direction,” Massiani said. “That has been the biggest positive piece of feedback — clients don’t wait as long.”
Customers may get frustrated if conversational AI fails to understand their questions, especially those worded in casual language or spoken with a strong accent, and delivers the wrong response, asks them to rephrase or drives them back to the same agents the bank was hoping to unburden. In the case of the phone, the customer may be directed to the wrong department.
“There is a huge variety in the quality of natural language processing,” Ward said.
Massiani says the bank is continuously monitoring calls for colloquial terms, slang and phrases that confuse Skye and tweaking the product accordingly so it can understand diversified speech.
“It’s an evolving process that will continue to evolve, likely forever,” Massiani said.
