Morgan Stanley's wealth division deploys AI agents to automate advisor workflows

Morgan Stanley's multi-year investment in technology infrastructure and data systems is beginning to yield results as the firm expands its deployment of generative and agentic AI across its operations. Within the wealth management division specifically, AI has matured from simple conversational assistants handling isolated functions into more sophisticated systems that coordinate work across multiple enterprise systems.

According to Hina Shamsi, CIO of wealth management technology at Morgan Stanley, the division's AI capabilities rest on a foundation of earlier strategic decisions about how to organize and think about customer data. "We know these AI agents or agentic workflows can only be as good as the context and data we have," Shamsi told CIO Dive.

Roughly eight years ago, the wealth management division fundamentally reconsidered its approach to data architecture. Rather than organizing information around individual accounts, the team shifted to a relationship-centric model that captures the full scope of each customer's interactions with the firm. This reorientation enabled machine learning capabilities such as next-best action recommendations, which guide advisors on the most appropriate steps to take with clients based on comprehensive portfolio information.

Ongoing work to strengthen data platforms, contextual layers and semantic infrastructure to support generative AI and agent systems remains a top priority for technology leadership. Shamsi explained the strategic importance of this layer: "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."

Expanding agent capabilities

Morgan Stanley's wealth management division is advancing its AI maturity by shifting toward AI-orchestrated workflows that can execute tasks on behalf of advisors or support staff across disparate systems, while keeping human judgment and approval in the decision loop.

The division is concentrating on three primary categories of agent deployment: client interaction, service support and advisor assistance.

The firm has introduced a client assist agent that accepts natural language questions and enables customers to complete routine transactions. This initial version of the client virtual assistant is designed to evolve toward more granular, account-level assistance with richer contextual capabilities as the system matures, Shamsi said.

The service assist agent is generating measurable productivity improvements by functioning as a tool for client service associates. The agent retrieves information and executes actions including fund transfers, account creation and beneficiary modifications. Shamsi described the value proposition: "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."

The adviser assist agent streamlines daily workflows by allowing advisors to delegate work without manually executing every step. The system aims to eliminate friction from routine administrative tasks so advisors can concentrate on direct client engagement. Shamsi noted: "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."

As agent capabilities expand, governance and control mechanisms become increasingly critical. Shamsi emphasized that agents must operate within strict parameters rather than having unrestricted autonomy. The agents deployed by her team function within "clearly defined boundaries" that specify which data agents can access, how they access it, what actions they can perform and the conditions requiring human intervention.

Shamsi stressed the necessity of comprehensive safeguards: "I can't stress enough the importance of guardrails, permissions, monitoring and human oversight."