Blab Tech
Bank of America CEO Prioritizes Accuracy Over Efficiency in AI Assistant Erica
Accuracy, not speed, is Bank of America’s guiding principle for its AI assistant Erica. In a recent interview, chief executive Brian Moynihan underscored that the bank’s priority is precision, even as other institutions chase cost savings and rapid deployment.
Moynihan explained, “When a customer types a question into the bank’s AI assistant, Erica, you can be driving 60 miles an hour up Sixth Avenue, and the infrastructure to get that inquiry to us and back has to be instantaneous, or you’re mad at us, so the data has to be perfect, and you ask about one check transaction of 20 that month in one of your accounts, and we have five, and we have to be exactly accurate.” The CEO’s emphasis on exactness reflects the risk that a single error could distort balances, misallocate funds, or trigger regulatory violations.
Theodora Lau, co‑founder of Unconventional Ventures, noted that Moynihan’s focus “stands out because most CEOs—at banks and other companies—are still talking about AI from the perspective of operational efficiency, productivity gains and cost cutting.” Lau added that an 80 % accurate model would be “zero value for a bank.” These remarks highlight the industry debate over whether speed and volume can compensate for lower accuracy in financial contexts.
Erica debuted in 2018 as a modest language model developed in partnership with Stanford researchers. Today the system spans 110 internal banking systems and is programmed to answer 700 distinct customer intents. Bank of America reports that 20 million customers use Erica, which is invoked 200 million times each quarter. The assistant handles tasks ranging from balance inquiries to initiating wire transfers, while a human remains in the loop for any judgment‑laden decision.
Last year, the bank intensified its investment in the technology. A 2025 press release announced that Erica now resolves 98 % of customer inquiries without further human intervention, and that 60 % of interactions are proactive, driven by the assistant’s own outreach. The company’s AI and machine‑learning patent portfolio has more than doubled since 2022, and it has spent billions expanding its AI capabilities.
Moynihan described the bank’s AI strategy as a “build once” philosophy. The assistant was designed to be a reusable platform that can be adapted across the bank’s product lines. He also highlighted that the data infrastructure and model verification processes are critical to maintaining accuracy, saying, “We keep it constrained enough to give the right answers.” Internal teams conduct continuous updates and validations, and the bank relies on external partners for model testing.
Regulatory bodies have issued guidance on AI model validation in the banking sector. The Federal Reserve’s SR11‑7 and the Office of the Comptroller of the Currency’s OCC11‑12 frameworks require banks to demonstrate that predictive models meet accuracy and fairness standards. Bank of America’s compliance teams report that they use a structured validation pipeline that includes scenario testing, bias assessment, and performance monitoring.
The focus on accuracy also aligns with broader industry concerns about AI risk. Analysts warn that generative models can produce plausible but incorrect information, and financial institutions must guard against such errors. Bank of America’s approach—tight constraints, human oversight, and rigorous validation—represents a cautious path that prioritizes customer trust over rapid deployment.
Looking ahead, Bank of America plans to expand Erica’s capabilities to cover additional banking services, but the company has stated that any new features will undergo the same stringent accuracy checks. The bank’s leadership believes that a reliable AI assistant will improve customer experience while mitigating operational risk.
In summary, Bank of America’s CEO has positioned accuracy as the defining metric for its AI initiatives. Erica, built with Stanford researchers and supported by a robust validation framework, exemplifies this focus. While other banks continue to pursue efficiency gains, Bank of America’s strategy underscores the importance of precision in financial AI applications.
Moynihan explained, “When a customer types a question into the bank’s AI assistant, Erica, you can be driving 60 miles an hour up Sixth Avenue, and the infrastructure to get that inquiry to us and back has to be instantaneous, or you’re mad at us, so the data has to be perfect, and you ask about one check transaction of 20 that month in one of your accounts, and we have five, and we have to be exactly accurate.” The CEO’s emphasis on exactness reflects the risk that a single error could distort balances, misallocate funds, or trigger regulatory violations.
Theodora Lau, co‑founder of Unconventional Ventures, noted that Moynihan’s focus “stands out because most CEOs—at banks and other companies—are still talking about AI from the perspective of operational efficiency, productivity gains and cost cutting.” Lau added that an 80 % accurate model would be “zero value for a bank.” These remarks highlight the industry debate over whether speed and volume can compensate for lower accuracy in financial contexts.
Erica debuted in 2018 as a modest language model developed in partnership with Stanford researchers. Today the system spans 110 internal banking systems and is programmed to answer 700 distinct customer intents. Bank of America reports that 20 million customers use Erica, which is invoked 200 million times each quarter. The assistant handles tasks ranging from balance inquiries to initiating wire transfers, while a human remains in the loop for any judgment‑laden decision.
Last year, the bank intensified its investment in the technology. A 2025 press release announced that Erica now resolves 98 % of customer inquiries without further human intervention, and that 60 % of interactions are proactive, driven by the assistant’s own outreach. The company’s AI and machine‑learning patent portfolio has more than doubled since 2022, and it has spent billions expanding its AI capabilities.
Moynihan described the bank’s AI strategy as a “build once” philosophy. The assistant was designed to be a reusable platform that can be adapted across the bank’s product lines. He also highlighted that the data infrastructure and model verification processes are critical to maintaining accuracy, saying, “We keep it constrained enough to give the right answers.” Internal teams conduct continuous updates and validations, and the bank relies on external partners for model testing.
Regulatory bodies have issued guidance on AI model validation in the banking sector. The Federal Reserve’s SR11‑7 and the Office of the Comptroller of the Currency’s OCC11‑12 frameworks require banks to demonstrate that predictive models meet accuracy and fairness standards. Bank of America’s compliance teams report that they use a structured validation pipeline that includes scenario testing, bias assessment, and performance monitoring.
The focus on accuracy also aligns with broader industry concerns about AI risk. Analysts warn that generative models can produce plausible but incorrect information, and financial institutions must guard against such errors. Bank of America’s approach—tight constraints, human oversight, and rigorous validation—represents a cautious path that prioritizes customer trust over rapid deployment.
Looking ahead, Bank of America plans to expand Erica’s capabilities to cover additional banking services, but the company has stated that any new features will undergo the same stringent accuracy checks. The bank’s leadership believes that a reliable AI assistant will improve customer experience while mitigating operational risk.
In summary, Bank of America’s CEO has positioned accuracy as the defining metric for its AI initiatives. Erica, built with Stanford researchers and supported by a robust validation framework, exemplifies this focus. While other banks continue to pursue efficiency gains, Bank of America’s strategy underscores the importance of precision in financial AI applications.
