
By answering employees' questions, the assistant, Erica for Employees, has reduced calls to the IT help desk by more than half, the bank said. More than 90% of employees use Erica, the bank said.
The company has also quietly rolled out generative AI in areas like Merrill Lynch, private banking and call centers, to help advisors and relationship bankers answer questions and draft client memos and to generate summaries of calls for customer service reps.
While other large banks have been public about their use of AI, especially generative AI, throughout their organizations,
Gopalkrishnan declined to share names of specific companies the bank is working with.
"When we think about build versus buy, we don't want to be building things that are increasingly foundational and are available as a commodity," Gopalkrishnan said. "We want to leverage the heck out of innovation happening in the industry. We want to leverage the heck out of partnerships we have that are investing tons of money in the space that said they have to eventually fit in with our framework."
The company spends $13 billion annually on technology; about $4 billion of that will be directed to new technology initiatives in 2025.
"I do think it is a good idea to use generative AI with employees first, as they are going to use it anyway, and it is a good way to make sure that IT stays in the loop," said consultant Aaron McPherson. "Erica now seems prescient, given the way things have evolved. Other banks are playing catchup with their own chatbots." However, some of the other improvements
Not using generative AI in client-facing Erica
"We just want a predictable response to a client," Gopalkrishnan said. "There's no reason to have a model write an essay when we're just trying to send the client to the right place."
Erica, which was one of the first virtual assistants in the industry in 2018, is based on natural language understanding that interprets what a customer is saying and deciphering the intent of what that person wants. It uses language models to classify content. Today, 20 million customers use Erica at least once a month.
Erica continues to be refined and get better over time, Gopalkrishnan said. "We essentially improve our classification schemes based upon all the knowledge we learn every single day. We have call centers that process calls. We have people that actually talk into Erica every single day. We use all that to glean better classification."
Erica for Employees, the version of Erica given to staff, was widely used during the COVID-19 pandemic by employees seeking technology support for things like mobile device password reset and device activation. In 2023, it was expanded to be able to help with HR topics like where to review health benefits and locate payroll and tax forms.
Gen AI for Merrill Lynch, research and elsewhere
The bank is adopting generative AI in other areas. In the Merrill Lynch unit, advisors can use a generative AI tool called "Ask Merrill" to run queries on behalf of customers, such as how to do estate planning in Delaware. In 2024, there were more than 23 million interactions with Ask Merrill and a similar tool, Ask Private Banking, an increase of 1 million over 2023, according to the bank.
"In doing so, we save hours of time for a relationship banker," Gopalkrishnan said.
Software developers are using a gen AI-based tool to assist with code writing and optimization, through which they have experienced efficiency gains of over 20%, the bank said.
An internally developed generative AI platform enables the global markets sales and trading team to search, summarize, and synthesize internal research and market commentary more quickly and efficiently.
In all these projects, the bank has a framework around responsible AI, he said, through which it looks at things like, bias, fairness and transparency, Gopalkrishnan said.
"It has to fit in with that framework," he said. "And if it does, we absolutely are partnering with all those players. We don't talk a lot about it. We just focus on, what do clients want to do? What are the client-facing solutions we want to deliver, and we talk more about that."