Bank of America said Tuesday that it has expanded EricaAssist with generative AI that supplies contextual guidance to customer-service employees during live conversations. The bank says the tool can summarize why a client is calling, gather relevant information and recommend next steps without taking the conversation away from the employee.
The July 21 Bank of America announcement says more than 18,000 service representatives use EricaAssist. It reports that guidance arrives in under three seconds and that the tool has reduced average call time by nearly one minute. Those performance figures are the bank’s own and have not been independently verified.
Reuters independently reported the rollout as part of a broader push by large banks to embed digital assistants in daily work. A separate Banking Dive report described EricaAssist as a desktop tool for employees, distinct from the customer-facing Erica virtual assistant.
The design choice matters more than the headline
The most relevant feature for credit unions is where Bank of America placed the technology. EricaAssist works inside an employee workflow, where a person can judge the guidance, explain a solution and remain accountable for the result. It is not being presented as an autonomous replacement for the service representative.
That makes employee assist a practical starting point for smaller institutions. A credit union could apply the same pattern to policy retrieval, call-reason summaries, product comparisons or next-step prompts before allowing an AI system to communicate directly with members or change an account. The sequence creates a review layer while teams learn where the tool is accurate, where it hesitates and where it needs better source material.
The approach also connects with earlier work on AI-generated contact summaries. Summarization can reduce after-call work, but real-time recommendations introduce a more consequential question: whether the underlying policy, eligibility rule or account information is current and correctly matched to the member’s situation.
Controls credit unions should test first
Contact-center leaders should not evaluate an assistant on speed alone. A shorter interaction can be a poor outcome if the employee receives outdated guidance, skips required disclosures or resolves the wrong problem. Useful pilots should track accuracy, repeat contacts, escalations, complaints and quality-review findings alongside handle time.
Before a production rollout, credit unions should require the system to identify the approved source behind each recommendation, preserve a record of what the employee saw, and fall back cleanly when data or model services are unavailable. The assistant should not make account changes, send member communications or bypass an established approval step simply because it can generate a plausible answer.
Data boundaries matter as well. Management should know whether call audio, transcripts and account details leave the credit union’s environment; how long prompts and outputs are retained; whether information can be used to train a vendor model; and which subcontractors support the service. Those answers belong in vendor review, privacy analysis and incident-response planning—not only in the contact-center implementation plan.
A useful benchmark, not a transferable result
Bank of America’s scale and technology budget are far beyond those of most credit unions, so its reported timing and efficiency gains should not be treated as a ready-made business case. The transferable lesson is narrower and more useful: put generative AI where an employee can supervise it, constrain the sources it may use and measure service quality before expanding authority.
Credit unions already considering member-facing virtual assistants can use this deployment as a checkpoint. An institution that cannot show how its own employees verify AI guidance is not yet ready to give the same technology more direct control over member interactions.