Bank of America's own count puts Erica past 3.6 billion chats

The conversation

Bank of America launched Erica, a virtual financial assistant built into its mobile app, in 2018. The bank does not run one standing 'Erica' page; instead it periodically restates usage figures across its own materials. Its Fast Facts page, last updated 14 July 2026 as retrieved, and a separate press release dated 21 July 2026 describe two related but distinct products: Erica, used directly by clients, and EricaAssist, a separate tool that gives the bank's own employees AI-generated guidance during client calls.

What the documents show

Fast Facts states Erica 'helps more than 24.6 million clients manage their financial lives' and counts '3.6 billion Erica interactions since' its 2018 launch, 'including 1.9 billion proactive, personalized insights delivered to clients.' It separately states that '97% of employees have adopted Erica for Employees since 2020.' The EricaAssist release states more than 18,000 customer service representatives now use that tool, that it delivers guidance within three seconds, and that it cuts average call time by about one minute per interaction. Each figure is Bank of America's own, tied to its own named product and period; the bank does not combine the client-facing Erica count with the employee-facing EricaAssist count.

The system boundary

The two products mark the boundary themselves. Erica answers clients directly inside the app. EricaAssist, by contrast, is explicitly built to keep a person on the call: it 'summarizes why clients are calling' and 'surfaces relevant guidance in real-time,' but the employee, not the assistant, speaks to the client. Ashley Ross, the bank's Head of Consumer Client Experience, is quoted describing the design as 'combining human judgment with real time AI guidance,' which the release frames as augmenting representatives rather than replacing them.

Where it fails

Neither document explains how an 'interaction' is counted, whether a single client conversation spanning several messages counts once or many times, or what share of Erica's 3.6 billion interactions were fully self-service versus escalated. Because the bank publishes these milestones periodically rather than as one continuous dataset, a reader comparing two dated snapshots cannot verify the interaction rate between them independently.

• How does the bank define a single 'interaction' for Erica's billions-count?
• What share of Erica conversations are resolved without any employee involvement?
• Are EricaAssist's call-time and speed figures measured against a control group, or only reported as an average?

Both documents are self-reported and undated in the sense that neither carries an independent audit; they are useful as a record of what Bank of America chose to publish and when, not as a verified performance study.