Banks' long road to retail-ready voice assistants

Artificial intelligence and voice assistants are changing how financial transactions are done. But bank executives say an all-virtual future is still far off, as conversational programs still aren’t capturing the nuance of speech and chatbots have disappointed many customers.

“We are at the mid-'90s of developing websites in terms of maturity here,” Ken Dodelin, vice president of conversational AI products at Capital One, told the audience at the Finovate conference this week in New York.

A low barrier to entry in creating chatbots is one reason they aren’t meeting customer expectations, Dodelin said. “There is a lot of proliferation of ones not well designed," he said. "This creates a stigma. We've got a maturity process going.”

A recent survey found that among the companies developing voice assistants, 88% reported speech error rates above 6%, while 35% of firms reported an error rate above 20%.

“Everyone is having a bad experience with chatbots,” said Jason Mars, co-founder and CEO of Clinc, a conversational AI developer.

Because of those bad experiences, there is a trust barrier that needs to be overcome when conversational banking finally reaches its full potential, said Tyler McIntyre, founder of the startup Bank Novo.

“The biggest issue we have today is accuracy,” McIntyre said. “When you think about a financial institution, one of the things you have is trust. As soon as your data is bad or not clean enough, all that stuff lowers accuracy of what the user says, but also what the data says.”

(From l.) Ken Dodelin, vice president of conversational AI products at Capital One; Chris Zahner, senior vice president and head of U.S. digital channels for Citi’s consumer bank; Jason Mars, co-founder and CEO of Clinc, an A.I. startup; and John Kelly, a financial services client partner at LivePerson,

Bank Novo is developing a voice component in its app, McIntyre said. Its testing is focused on improving accuracy.

A rather practical approach is a hybrid system, said Dodelin. The client questions a bot, and its reply is displayed on a screen, rather than a spoken response.

Capital One is still bullish about new voice technologies. It was the first bank to have an Alexa skill. Last year, it deployed Eno, touted as the first natural language SMS chatbot from an American bank.

On the other hand, Citigroup has been relatively cautious in approaching AI and conversational banking, said Chris Zahner, senior vice president and head of U.S. digital channels for Citi’s consumer bank.

In the iOS version of its app, Citi recently rolled out voice navigation via Siri. It also provided connections to live agents via messaging on the Citi app. Zahner said Citi officials want to see what customers are asking about and then figure out the bank's next move in conversational banking.

John Kelly, a financial services client partner at LivePerson, a business that facilitates conversational commerce for retail and financial services clients, said it is probably wise to take a cautious approach.

“We are dealing with the lifeblood of the company,” Kelly said. “All these conversations have to be managed or we may lose a customer or create a negative word-of-mouth. We can’t screw up with the experimental stuff. You are dealing with people’s money. You cannot mess that up.”

Given the risks, how can a company ensure its conversational banking ability is developing correctly?

Mars suggested that firms build a conversational banking app from scratch. A company cannot just bolt its chatbot to a voice interface, he said.

A great use case for AI and conversational banking, he said, would be to marry an augmented-reality experience and a voice service for customers who are buying a car — an actual project that USAA is developing.

When it comes to AI, banks have a unique advantage over fintech startups, McIntyre said. They have huge amounts of data on customers, going back years. The challenge for banks is how to best harness that data and provide a superior customer experience, via AI-driven voice assistants or chatbots.

“Good AI is built on data,” he said. “The more data you have, the more accurate it is, the more you can train it. Banks are the perfect ground for AI because they have data on everything.”

For reprint and licensing requests for this article, click here.
Artificial intelligence Machine learning Digital banking Online banking Data quality Data management Fintech Citigroup USAA
MORE FROM AMERICAN BANKER