Bank of America's AI Playbook: What CX Leaders Can Learn From Erica's Long Game

Bank of America announced enhancements to EricaAssist on July 21, giving more than 18,000 customer service representatives generative AI-powered guidance that resolves client needs in under three seconds.

This comes a few months after the bank's March 2026 digital engagement report — which put total client interactions at approximately 30 billion for the year, up 14% year-over-year.

The bank's digital experience roadmap goes back long before the debut of ChatGPT. Erica, its AI chatbot, debuted in 2018. By November 2021, when Holly O'Neill, then Bank of America's president of retail banking, spoke with CMSWire's CX Decoded podcast, Erica had already logged over 200 million interactions in its first 18 months and was helping 99% of clients self-serve without calling into a contact center.

O'Neill described a "high-tech, high-touch" strategy built around personalized, intent-based service — the same language and philosophy Bank of America is still using in 2026.

The Numbers Show Compounding, Not Just Growth

The March 2026 release put Erica's 2025 activity at 20.6 million users and nearly 700 million interactions in a single year, pushing the cumulative total past 3.2 billion since launch. In an upcoming video interview with CMSWire TV, Jorge Camargo, Bank of America's Head of Digital Platforms, cited an updated total of more than 3.4 billion interactions since 2018 — a sign the momentum documented in March has continued. Alongside Erica, the bank reported 16.6 billion digital logins and 13.3 billion proactive alerts sent in 2025, both records, with 86% of clients rating their digital experience at least a 9 out of 10.

EricaAssist's ability to cut average call times by nearly a minute per interaction isn't a modest efficiency tweak when it's applied across an 18,000-person workforce handling millions of client conversations. Small per-interaction gains compound quickly at that volume — a lesson for any CX organization evaluating whether a given AI capability is "worth it."

Employee-Facing AI Isn't a Separate Track

In the upcoming video interview with CMSWire TV, Camargo said the technology behind Erica was never built as a single-purpose tool. Camargo described it as enterprise infrastructure that the bank has extended into new use cases over time, with the contact center — and eventually EricaAssist — among the first places that infrastructure was deployed for employees rather than clients.

Bank of America runs its AI strategy for customers and staff, applying the same underlying system, and the same feedback loops, to both sides of the conversation. Camargo pointed to nearly 90,000 post-launch improvements made to Erica as evidence of how continuously the system gets retrained based on what both clients and employees are asking.

Related Article: Holly O'Neill of Bank of America on the Great Pillars of CX

What CX Leaders Can Take From It

For digital CX leaders evaluating their own AI investments, Bank of America's pattern offers a few practical signals:

• Treat AI as infrastructure, not a feature. Erica's expansion into EricaAssist worked because the underlying platform was designed to be extended, not rebuilt for each new use case.
• Measure compounding, not just launch-day metrics. A minute saved per call or a few seconds shaved off a lookup sounds small until it's multiplied across tens of thousands of employees or hundreds of millions of interactions.
• Keep humans in the loop by design. Both the March and July releases emphasize that AI supports employee decision-making rather than replacing it — consistent with the "high-tech, high-touch" philosophy O'Neill described back in 2021.
• Build the feedback loop early. Camargo's point about nearly 90,000 iterative changes suggests the real differentiator is the operational discipline to keep tuning it.

What Matters Here: Is Bank of America's AI strategy actually new, or just newly visible?

The core strategy — proactive, intent-based service delivered through one continuously improving AI platform — has been in place since at least 2018. What's new in 2026 is the scale and the extension of that same platform to employees through EricaAssist.