Brian Moynihan (브라이언 모이니한), CEO of Bank of America, pointed to accuracy as the top task in banks' artificial intelligence strategy. He said AI should not be seen only as a tool for streamlining or cutting costs, but whether it can provide customers with correct answers is key.

American Banker, a fintech media outlet, reported on June 8 that Moynihan said in a recent interview that the precision of answers is as important for bank AI as response speed.

He cited a case where a customer asks the mobile app AI assistant Erica about check transaction history for a specific account and said the related data must be perfect. In bank services, incorrect answers are directly linked to customer trust, leaving a very narrow tolerance for mistakes and requiring substantial work to build and validate systems.

The view differs from the trend at other companies that describe AI as a means to boost productivity or cut labour costs. Theodora Lau, co-founder of Unconventional Ventures, said most CEOs still talk about AI from an operational efficiency perspective, but in banking accuracy becomes a more important standard.

Matej Jendrolovics, CEO of Intuitech, also said on a podcast hosted by Lau, "If a solution is 80 percent accurate, it has no value in banking." He meant that unlike general office tasks that can be used with a certain level of accuracy, in financial services a single wrong answer can lead to a major problem.

Bank of America invested about $250 million in new AI deployments this year. Its annual technology budget is $13 billion, of which $4 billion is spent on new technologies including AI. At the centre of its AI strategy is Erica, an AI assistant launched in 2018.

Erica began with adding a Google search function inside the mobile app. At the time, when customers asked about their balance, irrelevant results such as photos of scales or yoga classes appeared, and internal developers later worked with Stanford researchers to build Erica as a predictive model connecting customer language with bank language.

Erica is now linked to 110 systems and designed to answer 700 types of questions. It is used about 200 million times a quarter by 20 million customers. Internally, about 200,000 employees also use AI in daily work, and the employee version of Erica handles IT and HR queries, cutting help desk calls by 55 percent.

Bank of America is deliberately limiting the scope of its AI use. Moynihan said humans must be involved in work that requires real judgment and that in an industry where trust is key, answers must be perfect. He said the bank is operating by sufficiently constraining the system so it produces correct answers.

It was also mentioned that a free-form agent-type AI could replace Erica in the future. Moynihan drew a line, saying at this stage it is more important that a proven approach actually works.

The comments show that even as banks expand the adoption of generative AI, accuracy over speed and validation over automation are likely to take priority in customer service and core operations. As competition in financial AI accelerates, data infrastructure and the ability to control models are emerging as core standards for banks' AI strategies.