Enterprise AI — March 2026
Merrill Lynch Deployed AI
to Every Client Meeting.
Bank of America’s Erica AI technology now powers advisor meeting prep, notetaking, and follow-up across Merrill and Private Bank, delivered through advisor-facing tools called Ask Merrill and Ask PB, integrated with Salesforce CRM and Zoom.
Bank of America’s Merrill Wealth Management and Bank of America Private Bank announced on March 26, 2026 the full-scale rollout of AI-Powered Meeting Journey, an integrated tool that helps advisors and private bankers prepare for, conduct, and follow up on client meetings. The advisor-facing assistants, Ask Merrill and Ask PB, are built on the same underlying AI as Erica, the consumer-facing assistant Bank of America launched in 2018. The rollout covers advisors and private bankers across both Merrill and the Private Bank, a group the company has put at roughly 15,000 people.
How the System Actually Works
AI-Powered Meeting Journey integrates Bank of America’s Erica-derived AI platform, Salesforce CRM, and Zoom’s meeting infrastructure. Before a client meeting, the system pulls the client’s account history, recent transactions, portfolio performance, and prior meeting notes from Salesforce. It generates a briefing document that summarizes the client relationship and highlights items likely to need attention, giving the advisor prep materials ahead of the call.
During the meeting, the system records and transcribes the conversation via Zoom’s AI companion, with client consent. After the meeting, it generates a summary, extracts action items, and stages follow-up tasks in Salesforce CRM. The advisor reviews and approves the outputs before they are saved. The human-in-the-loop approval step is non-negotiable in financial services: regulatory requirements (SEC, FINRA) mandate that client communications and account actions have human oversight.
How the Meeting Intelligence Architecture Works
Pre-meeting: CRM context loading. When an advisor opens a meeting linked to a Salesforce contact, the system loads the client’s portfolio summary, recent transaction history, and any open service cases into a prep briefing the advisor reviews before the call.
During meeting: AI notetaking, with consent. The system’s Zoom-based AI companion transcribes the meeting with the client’s consent. Bank of America’s public materials describe notetaking and prep, not a specific real-time, mid-call suggestion engine; that level of detail has not been disclosed, so it is treated here as unconfirmed rather than asserted as a documented feature.
Post-meeting: automated CRM update. After the call, the system drafts a CRM note summarizing discussed topics and flagged follow-ups. The advisor reviews and approves before it is saved to Salesforce. Bank of America has said AI suggestions are subject to the same compliance and audit requirements as any advisor action.
Why the Compliance Layer Is the Hard Part
FINRA requires that every product recommendation made by a registered representative pass a suitability analysis specific to the client. An AI that suggests a product without a traceable suitability determination is a compliance liability. Bank of America’s implementation is built around logging advisor actions and linking them to the client’s current profile, so that if an advisor acts on an AI-surfaced prompt, the audit trail shows what was surfaced, the client’s profile at that moment, and the advisor’s approval decision.
The system does not make recommendations to clients directly. Every suggestion goes through the advisor, who must exercise independent judgment before acting. The AI is a context and drafting engine, not a decision maker. This is the architecture that passes FINRA review. The system is built for meeting prep, notetaking, and follow-up, not for complex tax planning, estate structuring, or custom portfolio construction, the nuanced analysis that justifies Merrill Lynch’s advisor compensation model.
Why the 8-Year Build Matters
Bank of America launched Erica in 2018, four years before ChatGPT made AI assistants mainstream. Erica started as a simple mobile banking chatbot handling balance inquiries and bill payments. Over eight years, the platform has logged more than 3.2 billion client interactions, building a training corpus of financial conversations, client intent patterns, and regulatory-compliant response templates that no competitor can replicate quickly.
The “build once, deploy many” strategy means Erica’s underlying capabilities now extend from consumer banking (where it started) to wealth management (via Ask Merrill and Ask PB), to commercial banking and internal operations. Each deployment adds training data that improves the underlying model. A competitor starting from scratch in 2026 would need years of interaction data to match the nuance of Erica’s understanding of financial client conversations. The data moat is the real competitive advantage, not the AI technology itself.
Microsoft’s Copilot for Finance offers similar meeting preparation and summarization capabilities as a general-purpose tool. The difference is domain depth: Copilot understands meetings generically. Bank of America’s system understands financial advisory meetings specifically, and is purpose-built around the compliance and suitability requirements that generic assistants are not designed for.
The Deployment Scale
Deploying an AI system across roughly 15,000 advisors and private bankers simultaneously is a scale that most enterprise AI projects never reach. The logistics include training thousands of users on new workflows, integrating with individual Salesforce configurations that vary by client segment, product permissions, and compliance requirements, and maintaining regulatory compliance across all 50 states.
Bank of America’s ability to deploy at this scale in one release, rather than a phased rollout over quarters, reflects the institutional engineering capability that distinguishes large financial institutions from fintech startups. The compliance infrastructure, the change management process, the internal training programs, and the IT support capacity already existed. The AI feature plugged into an operational machine built over decades. This is the enterprise deployment advantage that pure-play AI companies cannot replicate: not the technology, but the organizational infrastructure to deploy technology at scale in a regulated environment.
Sources: Bank of America and Merrill press release, March 26, 2026; American Banker and WealthManagement.com coverage of the launch; FINRA AI guidance. Updated 2026-08-18 to correct the advisor count (standardized to the roughly 15,000 figure Bank of America has used, replacing a conflicting 19,000 figure), correct the lifetime Erica interaction count to the confirmed 3.2 billion, clarify that advisors interact with the system through Ask Merrill and Ask PB rather than Erica directly, and remove specific real-time mid-meeting suggestion details that Bank of America has not publicly confirmed.