Bank of America Rolls Out Generative AI Tool for Agents

What happened

Bank of America has extended generative AI into its internal customer service platform, EricaAssist, giving roughly 18,000 contact-centre agents real-time, AI-generated client insights during live calls. Rather than pushing generative AI toward customer-facing chat, the bank has directed the technology inward, using it to support the humans who handle client interactions.

According to Finextra, EricaAssist draws on the same underlying AI capabilities as Erica, the bank's long-running virtual assistant for retail customers, but repurposes them to surface account information, suggested actions and contextual detail for agents as calls unfold — aiming to cut the time agents spend searching systems mid-conversation.

Why it matters

The move signals a broader shift in how large service organisations are deploying generative AI: not as a replacement for human contact, but as a real-time decision-support layer sitting behind it. For CX leaders, this is a meaningful data point in the ongoing debate about where AI delivers value fastest — augmenting frontline staff tends to reduce operational risk compared with fully automating customer-facing conversations, while still improving speed and consistency of service.

From a behavioural-economics standpoint, reducing an agent's cognitive load during a call changes the interaction itself: agents with faster access to relevant information can focus more attention on tone, empathy and problem-solving rather than system navigation, which shapes the customer's perceived effort and satisfaction even though the AI never speaks directly to them.

By the numbers

• 18,000 call centre agents are being equipped with EricaAssist, according to Finextra.

The Renascence take

Most coverage of generative AI in service settings fixates on customer-facing chatbots and self-service deflection rates. Bank of America's approach is a useful counterpoint, and worth reading closely for what it implies about sequencing and risk management in large-scale AI rollouts.

The real story here isn't the technology — it's the choice of where to place it. Agent-facing AI is a lower-risk, higher-control way to test generative AI at scale: errors are caught by a trained human before they reach the customer, and the bank retains the accountability layer that regulated industries can't easily outsource. The behavioural payoff is subtle but real — an agent who isn't mentally juggling four systems mid-call has more bandwidth for the human parts of service that AI still can't do. Operators watching this space should resist the temptation to lead with a customer-facing bot; proving value and trust internally first, with the people who already own the relationship, is the more durable path to automation that customers actually feel good about.