How Bank of America is using AI to improve customer experience

Bank of America has spent the past decade turning AI into the backbone of its digital customer experience, with its Erica virtual assistant now central to how tens of millions of customers interact with the bank every month. The strategy goes well beyond a chatbot, blending AI-driven self‑service, proactive financial insights, and real‑time feedback loops that reshape both the digital and human-assisted experience.

Erica as the front door to the bank

Launched in 2018, Erica has become one of the most widely adopted AI-powered financial assistants in the industry, with nearly 50 million users and more than 3 billion client interactions to date. It now averages over 58 million interactions per month, making it the de facto front door for a growing share of Bank of America’s retail and wealth clients.

At its core, Erica uses natural language processing and machine learning to interpret customer queries and select appropriate responses from a curated library of more than 700 answers, which has been iteratively refined through some 75,000 updates since launch. Bank of America positions Erica not as a separate channel but as an embedded layer across mobile, online, Merrill, CashPro and benefits platforms, so customers can ask for help wherever they already are in their financial journey.

From problem resolution to proactive insight

Erica’s value proposition has steadily shifted from simple Q&A into proactive, personalized guidance that aims to prevent issues before they become service problems. The assistant pushes more than 1.7 billion proactive insights to clients, including alerts about upcoming cash‑flow shortfalls, eligibility for Preferred Rewards, and tailored BankAmeriDeals offers based on spending patterns.

Examples of this proactive model include:

• Notifying customers when their balances are likely to trend low in the next seven days.

• Highlighting relevant cash‑back deals that fit past card usage.

• Flagging when a client qualifies for relationship‑based benefits, then guiding them into enrollment.

• Providing guidance on nearly 50 investment topics for Merrill relationships.
  Each of these use cases is designed to reduce friction, cut avoidable fees, and deepen engagement, which in turn lifts digital sales and rewards adoption.

Containment, call‑center relief and “human in the loop”

By Bank of America’s own metrics, Erica resolves the overwhelming majority of interactions without needing to hand off to a human – more than 98% of users find what they need through the assistant. That containment has translated into significantly lower call volumes, with claims that calls into the IT service desk for employees alone have been cut by half after deployment of Erica for Employees.

Crucially, the bank frames AI as a way to re-route effort, not simply remove headcount: freed from handling balance checks, password issues and basic “how do I” questions, human bankers and advisors can focus on complex cases requiring empathy, judgment and tailored structuring. The digital assistant also supports seamless escalation by scheduling appointments and handing customers to high‑touch channels when it detects needs beyond its remit.

Data, feedback and the “Voices” CX engine

Underpinning the AI experience is a deliberate feedback architecture that continuously tunes both digital and human channels. The bank’s “Voices” program pushes millions of surveys across all touchpoints—branch, contact center, digital and Erica—using a consistent 1–10 scale so experience scores can be compared across journeys and segments.

Data from these surveys feeds real‑time dashboards visible from the front line to senior management, highlighting both pain points and positive outliers in customer interactions. That insight is then looped back into Erica’s training data, script refinements, and user‑experience changes, while also shaping coaching and process fixes for human bankers. The net effect is a closed‑loop CX system where AI interactions and human interactions are measured together, rather than as competing experiences.

Scaling AI: from customer tool to enterprise platform

While Erica began as a consumer‑facing assistant, Bank of America has deliberately reused its AI investments across the enterprise. The same underlying NLP and orchestration capabilities now power tools like ask MERRILL and ask PRIVATE BANK, which help wealth teams surface relevant research, documents and opportunities more quickly for clients.

Erica for Employees, used by more than 90% of staff, automates routine queries—from IT and HR issues to benefits and tax‑form lookups—so frontline and back‑office employees spend more time on high‑value client work. Bank of America executives describe a portfolio approach where early Erica projects “pay the freight” for data pipelines, governed feature stores and reusable APIs, making each subsequent AI use case cheaper and faster to stand up. This reuse strategy is now extending into generative AI proofs of concept, many of which are aimed at deepening personalization and further reducing servicing friction.

Guardrails, regulation and lessons for credit & collections

For credit and collections professionals, several elements of Bank of America’s AI playbook are instructive. First, the bank stresses that “AI is not a strategy”—it is a tool deployed against clearly defined client needs, with careful measurement around both satisfaction and operational impact. Second, much of the value comes from proactive, data‑driven outreach—alerts, insights and offers—rather than simply deflecting inbound calls, a pattern that maps directly onto delinquency‑prevention and hardship‑management use cases.

Third, the governance model matters: Erica relies on constrained response libraries, constant human review, and feedback from the Voices system to mitigate the risk of biased or inaccurate outputs—an approach that will be essential as collections shops experiment with LLM‑driven virtual agents under tightening regulatory scrutiny. Finally, the bank’s reuse mantra—building shared data and model infrastructure that supports multiple journeys—offers a blueprint for collections organizations seeking to scale AI across early‑stage outreach, vulnerability detection and repayment‑plan servicing without fragmenting their tech stack.