Retail banks are converging on the same answer to the AI question: keep the human on the call, and put the model behind them instead of in front of the customer. Bank of America has rolled out a generative AI upgrade to EricaAssist, the tool that supports customer service representatives during live client calls. More than 18,000 representatives now use it, and the bank says it generates contextual guidance in under three seconds, summarizing why a client is calling, surfacing account information, and recommending next steps based on the representative’s role, without interrupting the conversation.
Why it matters: the payoff is measurable and modest, an average reduction of about one minute per call, not a dramatic productivity leap. That restraint is the point. Ashley Ross, the bank’s head of consumer client experience, called it augmentation rather than replacement: “By combining human judgment with real time AI guidance, we’re helping employees navigate complex topics more easily.” CIO Tom Ellis tied the capability to oversight: “This technology helps our teammates deliver relevant insights in seconds, while operating with strong governance.” Bank of America backs this with a $14 billion annual technology budget, over $4 billion of it for new initiatives including AI, and plans to extend EricaAssist into more servicing scenarios later in 2026.
The original insight is about sequencing, not scale. Bank of America is proving generative AI’s value inside a fully human-supervised channel, where an employee reviews every suggestion, before it ever touches an unsupervised interaction. That mirrors a pattern elsewhere in banking, where institutions are choosing governable, human-in-the-loop AI over black-box automation to satisfy compliance requirements. A visible, quantified, supervised deployment doubles as a governance case study for regulators, which may matter as much as the minute saved per call.
Source: Bank of America
