There's been a lot of chatter in banking circles lately about chatbots.
Of course, as with most things in nascent technology, that chatter is a bit divided on whether chatbots are ready for primetime.
The naysayers seem to be in the minority. A survey conducted by Personetics this year found that all but 13% of bankers say they plan to do something with chatbots within the next two years.
For the banks toying with the idea, there is a growing pool of vendors offering prefab chatbots that have been trained with documents, data and conversations about financial products and topics. Some are intended to replicate the experience a customer might have with a human, while others are styled more like workhorses. Here is a look at a few of them.
The conversationalist
Kasisto, a spinoff of the research lab SRI International, like Apple’s Siri, offers a chatbot called Kai.
“Siri knows a lot — it’s very broad but very shallow,” Zor Gorelov, CEO and co-founder of Kasisto. A group of SRI researchers realized a financial services chatbot would have to be narrow and deep and started working on one. Around 2009, the group was approached by BBVA, which was seeking human-like interactions from a chatbot, and the two organizations partnered to create a virtual banking assistant.
The SRI researchers “interviewed people across the banking universe at BBVA, saying how do you transfer money, how do you pay bills, how do you apply for a mortgage?” Gorelov said. “They transcribed tens of thousands of calls.”
They also conducted thousands of Wizard of Oz sessions — having the chatbot answer questions in a way that appears automated, but really has humans behind it. (Barclays and USAA also use human subject matter experts to power and train their chatbot engines, and hope to eventually hand the reins over to the machine.)
Today, Kai is used by digibank, a mobile-only bank launched in India by DBS Bank in Singapore, and it’s being piloted by Royal Bank of Canada.
A staff of full-time writers — called “artificial intelligence interaction designers” — produces dialogues for Kai. They also constantly monitor behavior and user interactions. General banking knowledge has also been embedded.
“You can ask Kai questions about CDs, IRAs, credit scores — it’s the smartest banker you can imagine from a customer onboarding point of view: how do I open an account, what document do I need?” Gorelov said.
In an average session, there are 10 to 11 messages exchanged between Kai and users, Gorelov said.
Through its pilots, the company has learned that people want to get a better handle on their money and find a way to better spend and save money. Also, while this might be a given in human conversations, some people need a push to generate conversations.
“If you send me a notification that says my balance is low, that does not drive engagement,” Gorelov said. “But if you send me a notification that says, ‘You spent $200 in taxis this month, would you like to see your transactions?’ that brings people back and engages them and gives them a better understanding of their money.”
The doer
Personetics’ chatbots are used by Ally Bank and Societe Generale. They’re fueled with unsupervised and supervised machine learning, natural language understanding, logic inference, and associative knowledge.
The chatbots can walk customers through steps, provide predictive messages and behavior insights, and automatically perform tasks like money management. So though they can “chat” with customers, they’re primarily designed to actually do things for them. For instance, Societe Generale uses the software to take care of fund investing tasks.
While other chatbots might aim to simulate a real conversation, Personetics tries to make it clear the customer is
Personetics has built a library of customer insights for its chatbots. For instance, one might automatically inform a customer when a 30-day free subscription offer turns into a paid subscription. Another can forecast cash flow and let a customer know before their funds become dangerously low.
The chatbot technology has been fed financial services information for five years, according to Eran Livneh, vice president of Personetics. In other words, the chatbot is akin to an employee who understands banking and serving customers really well, he said.
Customers generally like things that help them with the day-to-day financials, like suggestions for saving money, Livneh said.
“If this month you spent more on dining out than you usually do, we may point it out,” Livneh said. “That creates trust and engagement. They’re useful for the customer, but don’t require them to do something outside of what they normally do.”
The linguist
North Side, a software company based in Montreal, specializes in giving its chatbot, VerbalAccess, a precise understanding of language, whether spoken or typed, through natural language processing technology.
North Side didn’t start out in banking. Originally it created a video game that lets players communicate with characters to command them to do things and to master the game.
“That’s how we made our natural language understanding pipeline robust,” said Eugene Joseph, North Side's CEO.
“In 2014, we decided we needed to make more money and oriented that toward financial services,” Joseph said.
North Side doesn’t try to glean insights from analyzing customer behavior. Instead, it takes commands and acts on them, such making a payment or displaying a transaction.
If a user asks, “What have I spent on coffee in the last month?” North Side’s chatbot will understand the question and translate it to an API call that will extract the answer automatically. It would make the same calls that might be made by a mobile banking app or online banking site, but with the added ability to translate from the imprecise way a human might request something to language the software can understand.
“We’ve invested $20 million in being able to do things like that,” Joseph said.
Joseph would not name customers, but said U.S. banks are using North Side’s software.
The chatbot is trained to clarify users’ questions. If a customer types “Send $100 to my brother,” the software will enter into a dialogue to find out who the brother is and which account should be drawn from.
“If what is said is incomplete, it will elicit the missing information,” Joseph said. “That’s very important because people speak in an incomplete way. We know what to ask for.”
The teller
Sidharth Garg began working on his chatbot, Teller, during his last semester at Columbia Business School. His thought was to provide general personal finance education as well as personal banking information.
“I wanted to use the recent advances in natural language processing and the opening of messaging platforms to help people learn the basics of personal finance,” Garg said.
He said he wanted it to be able to handle everything from account balances to savings advice.
To feed Teller the information to answer customers’ questions, Garg went through blogs and personal finance books, attempting to answer every general personal finance question he could think of.
“I summarized those answers, tried to condense all that information into easily understood explanations and put together Buzzfeed-style graphics that explain the different types of retirement accounts into something someone could understand on their mobile device,” he said.
A few months after graduating, he decided to focus on the business-to-consumer market, rather than offering a standalone product.
“People are more inclined to chat with a banking assistant that comes from an institution they already trust,” Garg said. “They already have a captive audience of customers and they would also allow for easy integration with their bank accounts.”
Recently, he’s been running a pilot program with Brooklyn Cooperative Federal Credit Union that has provided additional real-world data to train the chatbot. (The credit union did not make anyone available for interviews.)
Brooklyn Cooperative is looking to automate customer service and customer education.
“As a credit union they do a lot of coaching and counseling sessions, so they were also interested in offering this as a tool to people to build their financial skills,” Garg said.
For now the bot is answering only the most general questions for the credit union’s customers.
“Any more complicated thing we route towards a human to handle,” Garg said.
Eventually, the plan is to integrate the chatbot with the credit union’s back-end system, to let it answer questions specific to customers’ accounts.
“We’re still early in the chatbot days, but I think there’s a lot of potential there for a bank to slowly coach someone into being a really engaged customer,” Garg said.
Editor at Large Penny Crosman welcomes feedback at penny.crosman@sourcemedia.com.
