deeply integrated into banking.” And a key aspect of banking or any other business field is that when people are trying to do something complicated, they find themselves in conversation. “We kind of have to figure it out as we are talking to the person on the other side,” says Mark. “So you had to have a system that could actually have a conversation with the user in order to get something done.”
That’s where a lot of the heavy lifting in the Lola project occurred, according to Mark and Bartolomé. Researchers at SRI had to come up with a way to quickly diagnose a user’s intent and remember that intent through multiple conversational exchanges, while remaining ready to detect a different intent and switch to it at any moment. Here’s the example Wolverton played out for me in a live demo:
Wolverton: “What is the next payment amount on my loan?”
Lola: “Please tell me the account.”
Wolverton: “My loan.”
Lola: “The next payment amount for your installment loan is $436.”
Wolverton: “Sorry, my mortgage.”
Lola: “The next payment amount for your mortgage is $1,982.28.”
Wolverton: “And what is the payment due date?”
Lola: “The payment due date for your mortgage account is July 25, 2012.”
Notice the trick here: Wolverton switches intents halfway through the conversation, asking about his mortgage rather than his installment loan. When he asks a followup question, Lola hasn’t forgotten the change in context, and she very reasonably assumes that the question is about his mortgage loan rather than his installment loan, so she gives him the correct due date.
“This happens so naturally in human dialogue that we don’t even think about it,” says Mark. But making it work in software is tricky, and requires a thorough understanding in advance of all the directions in which a banking conversation might flow, and what actions to trigger in each case.
“The speech recognition and natural language is still a pretty hard thing to get good at, and it’s something we’re still working on,” Mark says. “But the ability to follow dialogue and keep the context straight is another hard thing. And then to actually get the work flows right is a third hard thing.”
Work flows are a big deal at BBVA: as Bartolomé noted, the bank has built its whole brand around providing a personalized human touch in all interactions with customers, even automated ones. Bartolomé says BBVA spends more on technology than any other European bank, which makes it more efficient, but that more than half of its technology investment goes to improving customer relationships. “What we have discovered is that as important as the functional needs of our customers is satisfying their emotional needs,” she says. “If you are not happy with our customer service, you are not happy trusting your money to us.”
The more tasks banks can offload to virtual assistants, certainly, the more money they’ll be able to save on human staff. But that’s not the only benefit of virtual personal assistant technology. If Lola were to catch on the way Siri did, it could help boost BBVA’s brand as an innovator. Eventually, BBVA says, Lola could even grow beyond her role as a kind of glorified bank teller and evolve into a financial advisor, giving customers personalized answers to questions about which savings or investment options are best for them. (Don’t expect that from Siri anytime soon.)
BBVA is the second largest bank in Spain and one of the 15 largest banks in the world. It entered the U.S. market in the mid-2000s by buying a series of banks in Texas, Alabama, and other Sun Belt states, forming BBVA Compass, which is based in Birmingham, AL.
