How WaFd embraced Amazon Lex’s conversational AI to improve and accelerate phone banking

Phone banking is starting to undergo a dramatic personality change, thanks in no small part to artificial intelligence (AI) and conversational AI.

The first generation of telephone banking was largely driven by interactive voice response (IVR) technology. That’s the touch tone activated technology that provides the monotone voice that tells you to “press 3 for your bank balance”. IVR is a technology that nobody really liked, but it has been working for many banks around the world for decades, albeit in a sub-optimal way.

A new, more modern approach is now emerging with capabilities such as voice print authorization, which provides a fingerprint-like feature that allows a user’s voice to authenticate into the system. The booming machine voice of IVR and the frustrating experience of pressing buttons to move through multiple menus is also coming to an end with the help of an AI-powered customer contact center platform.

Formerly known as Washington Federal, WaFd is one such financial institution that is now using Amazon Lex’s conversational AI technology to revolutionize the way it does phone banking. WaFd is based in Seattle, Washington, and has more than 200 locations in eight states. The company has built out its own digital support organization, known as Pike Street Labs, to help advance technology initiatives.

For WaFd, moving to conversational AI was also a way to help improve the user experience by helping customers get the information they need faster. Dustin Hubbard, chief technology officer at WaFd Bank and Pike Street Labs, told VentureBeat that it used to take about four and a half minutes for a customer to get to the point where they could press a button to actually withdraw their balance.

“Now when you call in, the system knows if you have voice authentication, which means you can prove your identity,” Hubbard said. “You say ‘My voice is my password’ and the system replies ‘Great, how can I help you?’ and at that moment you immediately have a conversation with the chatbot.”

Instead of the user having to go through the IVR menu to get to the right place to figure out which button to press to get what they want, all the customer has to do is ask for what they want. Hubbard estimated that instead of four and a half minutes for a user to get a bank balance, they can now get it in about 28 seconds.

Several components of the WaFd platform have replaced the old IVR system.

The voice printing authorization capability comes from Talkdesk, which offers a cloud-based call center as a service. Hubbard explained that when a customer calls the WaFd bank number, the call is recorded by the call center system. The voice authentication system verifies a user’s voice and analyzes the phone number and location from which a call is coming in before granting access to an account. The call center system can then log into the WaFd online banking backend system through a series of APIs.

Once logged in, Amazon Lex’s conversational AI technology kicks in. Amazon Lex is the fundamental conversational AI technology behind Amazon’s Alexa voice assistant. Amazon’s Lex service knows that the user has been authenticated and simply asks users to check a balance or make a transfer between accounts.

While WaFd uses two systems, with Talkdesk and Amazon Lex, Hubbard said it’s completely transparent to users who don’t know they’re moving between systems. To enable the seamless integration, especially in terms of the actual voice that users will hear, WaFd uses Amazon Polly technology.

Amazon Polly offers a text-to-speech feature that WaFd has used to record the spoken prompts the user hears into the Talkdesk system when they first call in. WaFd uses the same voice with Amazon Polly as with Amazon Lex. Hubbard said he wanted to make sure users got a consistent experience with the same voice.

Show me the money – how Amazon uses Lex utterances to train conversational AI

Getting Amazon Lex trained for the WaFd banking use case didn’t take much effort.

Hubbard explained that Amazon Lex is trained with an approach known as utterances. For example, giving the system the phrase “check balance” and variations that users might use in a normal conversation instructs the system to check the user’s account balance.

Over time, Amazon Lex gets better at deducing what the user’s intent is. For example, if a user does not say exactly what the trained utterance is, the system can infer the probability that it is close by and will still perform the expected action.

“For example, the conversational AI part with Amazon Lex is better than a lot of other systems I’ve seen, where if it’s not an exact match, the thing just has no idea what you’re talking about,” Hubbard said. “For example, you can say ‘show me the money’ and the system knows you want to check your balance. And in general it gets better at recognizing patterns over time.”