The way banks deliver services is changing fast, and one of the biggest drivers of that change is conversational intelligence. Erica, Bank of America’s AI-powered virtual assistant, is a standout example.
During their Investor Day presentation, Bank of America revealed that Erica had more than 20 million users in Q3 2025 alone.1
Erica at the Evident AI Symposium
At the third annual Evident AI Symposium in New York City on October 23, 2025, sponsored by Teradata, Hari Gopalkrishnan, Bank of America’s CIO & CTO, shared some eye-opening stats. Erica now handles 3 billion interactions every month and can even authenticate users seamlessly when transferring them to a live agent.
Erica: Call deflection
Erica’s original mission was simple: reduce call volume by solving customer issues through automation instead of phone conversations. Over time, Erica has evolved to address a broader set of customer needs and more complex interactions.
That evolution has been shaped by close attention to feedback from both customers and employees. One key innovation is Erica’s use of small language transformer models fine-tuned for banking terminology. These models can detect multiple customer intents in a single conversation, making interactions smarter and faster.
At the Symposium, Gopalkrishnan announced several new initiatives, including Erica Assist for contact centers, and enhancements to Erica for Employees, the bank’s internal help desk tool.
The emphasis on process design reflects a clear operational mindset. As Gopalkrishnan put it, “every call to the call center is a defect.” To address this, Erica needs to seamlessly understand the nature of each call, convert speech to text in real time, and store transcripts.
Erica: The backstory
Erica’s journey began in 2017 in the consumer banking division and later expanded to Merrill Lynch wealth management. The goal was to deliver a top-notch digital experience.
According to Gopalkrishnan, a central challenge early on was that some startups often had strong capabilities in Natural Language Understanding (NLU), but they lacked knowledge of financial ontology.2 Bank of America tackled this by adopting open-source NLP engines and hiring linguistics PhDs to build models tailored for banking.
The team focused on integrating Erica with existing HR and help desk systems, rather than reinventing the wheel. “We realized we could leverage all the goodness available in NLU and user experience,” Gopalkrishnan said. “There are models that do a great job of identifying service intent.”
Teradata’s role in conversational intelligence
Teradata’s Customer Journey Analytics helps banks analyze interactions across channels, including semi-structured data like chat logs. It also enables interrogation of unstructured data, such as emails and call transcripts, using open-source NLP models that can run in-database.
Meanwhile, Teradata’s Autonomous AI + Knowledge Platform leverages a financial services industry analytics schema, giving large language models the banking context they need to assist agents effectively.
1. “Investor Day 2025,” Bank of America, 2025.
2. Rabab Ahsan, “The story of Erica, Bank of America’s homegrown digital assistant,” Tearsheet.co, 2025, https://tearsheet.co/podcasts/the-story-of-erica-bank-of-americas-homegrown-digital-assistant/.