Morgan Stanley spent years upgrading its technology and data foundation in support of AI capabilities. The investments are starting to pay off as executives expand generative and agentic AI use cases across the business.
In the company’s wealth management division, AI use has evolved from conversational tools assisting with discrete tasks to more orchestrated systems capable of managing workflows across enterprise platforms, according to Hina Shamsi, CIO of wealth management technology at Morgan Stanley. A large part of those expanded capabilities are thanks to initial foundational investments, she said.
“We know these AI agents or agentic workflows can only be as good as the context and data we have,” she told CIO Dive.
The division made a significant pivot when it came to how they viewed data around eight years ago, Shamsi said. Initially, executives thought about data in an account-centric manner. But customers instead wanted to know about the entirety of the relationship — not just specific accounts.
The shift in focus led Morgan Stanley to meaningful machine learning capabilities including next-best action, Shamsi said, giving advisers guidance on appropriate next steps to take with clients based on the broader portfolio of data.
Continued investments in data platforms, context and the semantic layer to support generative AI and agents remain a critical focus among tech leaders, Shamsi said.
“What that has allowed us to do is really start to focus on some of the other things, building that semantic layer that GenAI and agent workflows understand,” she said.
Expanding agent capabilities
The bank’s wealth management division is maturing on its AI journey, shifting the focus to AI-orchestrated workflows, Shamsi said.
The goal is to implement agentic capabilities that can execute work on behalf of an adviser or support professional across different systems, while keeping human oversight and decision-making in the middle, she said.
The three big focus areas for wealth management include client interaction, service and adviser assist agents.
The firm rolled out a client assist agent that lets clients ask questions in plain language and complete common tasks. It’s the first iteration of the client virtual assistant and, as it matures, it will be geared toward account-specific assistance with more contextual support capabilities, she said.
Shamsi said the division is also seeing productivity gains with their service assist agent, which she described as a companion to the client service associates. The agent can retrieve information and take actions such as money movement, opening accounts and changing of beneficiaries.
“While they sound like simple operations, they require going to multiple applications and platforms, but our agentic workflow is designed to automate routine tasks so our support professionals can spend more time on higher value-added tasks,” she said.
The adviser assist agent helps with day-to-day activities, enabling advisers to direct work without having to complete all the manual functions, Shamsi said.
“The idea here is we’re trying to remove friction from things that they have to do and focus on what’s the value-add activity, which is spending more time in the client engagement itself,” she said.
As agentic capabilities increase, controls become even more important, Shamsi said. Agents should not be given unlimited autonomy, she added.
The agents her team builds operate within “clearly defined boundaries,” including what information agents are allowed to access, how they’re allowed to access it, what actions they’re allowed to take and when a human needs to step in, she said.
“I can’t stress enough the importance of guardrails, permissions, monitoring and human oversight,” she said.
