Share with your CMO
Bank of America has spent seven years building Erica into something more than a chatbot, and the July 2025 EricaAssist expansion, which puts generative AI-powered guidance in front of 18,000 customer service reps with sub-three-second response times, is the clearest proof yet that the strategy is compounding. Erica has logged more than 3.4 billion interactions since 2018, with nearly 700 million in 2025 alone. Jorge Camargo, BofA’s Head of Digital Platforms, describes the underlying AI platform as enterprise infrastructure extended into new use cases, not rebuilt for each one.
What this means for your business
The question BofA’s numbers force is whether your AI CX investments are accumulating or just running. Most enterprise CX programs treat customer-facing and employee-facing AI as separate budget lines with separate vendors and separate roadmaps. BofA’s architecture runs them on one platform with shared feedback loops, and the result is nearly 90,000 iterative model improvements over seven years. If your org can’t point to a comparable retraining cadence, you’re not compounding, you’re just spending.
The per-interaction efficiency math deserves a harder look than it usually gets. Cutting average call handle time by under a minute sounds like a rounding error until you multiply it across 18,000 reps handling millions of conversations annually. That arithmetic is what separates an AI proof-of-concept from a structural cost and quality advantage. The CMO who only measures AI by CSAT lift or deflection rate is looking at one variable in a multi-variable equation, and probably underselling the investment to the CFO in the process.
The falsification condition here is scale. BofA’s model works because volume turns small per-interaction gains into material outcomes, and because a decade of interaction data gives the retraining loop something real to work with. A CX organization with lower transaction volume or fragmented data infrastructure won’t get the same compounding returns from the same architecture. The honest version of this story isn’t “copy BofA,” it’s “decide whether you have the volume and data continuity to make a unified AI platform pay off, or whether point solutions are still the right call for your transaction density.”
Based on reporting from What BOA’s EricaAssist Proves About Measuring AI’s Real CX Impact, originally published 2026-07-27 11:43:00.
