Regions Bank is using IBM’s Watson, the artificial intelligence machine that famously won "Jeopardy!" several years ago, to handle calls to its call center.
The Birmingham, Ala., bank has been testing and using Watson for a year, but is only going public about it on Thursday.
Watson technology includes natural language understanding, search, entity extraction, and machine learning combined to perceive a customer’s question and find an answer. (Kasisto, Clinc, Finn AI and Personetics are among vendors that also offer such technology.)
Watson has been deployed three ways within Regions so far, according to Chris Brasher, head of banking operations, whose job includes overseeing the bank’s centers of excellence for artificial intelligence and automation.
Regions’ first use of Watson is a virtual banker it calls Reggie. It speaks directly with customers about simple matters, such as updating an address or ordering a new card. More complicated requests and queries it doesn’t understand are routed to a human customer service rep. Twenty-two percent of the time, calls are completely handled (or “fully contained” in the industry parlance) by Reggie.
Watson is also implemented as Banker Assist, a virtual assistant that helps call center staff quickly find information they need. The reps type in search phrases of up to three words. Seven hundred bankers are using it.
The third way Regions is using Watson is to listen to calls after the fact, determine what could have been handled better, and train associates. This iteration is called Conversation Analyzer.
Two other banks using Watson in a similar way are Royal Bank of Scotland and Banco Bradesco, according to Shanker Ramamurthy, general manager, strategy and market development at IBM.
RBS’ virtual assistant Cora is trained on more than 1,000 responses to more than 200 customer queries. The bank’s internal assistant, Ask Archie, can handle common human resources queries.
Banco Bradesco built an AI virtual assistant to allow customers and agents to quickly find answers to 283,000 questions a month, with a 95% accuracy rate.
