In the beginning, there were large language models. And they were limited. But as each wave of artificial intelligence innovation arrived, the capabilities became more robust.
Now, agentic AI can take on multiple layers of complexity, not just to perform a single task, but to reach a desired goal like deepening client engagement.
During her keynote presentation at the ADVISE AI conference last week, Amy Young, formerly of Microsoft, discussed how these new capabilities can not only give advisors more time with clients, but use the new freedom in their schedules to deliver greater value.
READ MORE: This is the biggest cybersecurity threat for wealth firms
In her experience, Young said the No. 1 reason firms want to invest in AI is to build client relationships. However, she said, there is remarkably little conversation about how to help advisors spend better — rather than simply more — time with clients.
“And to be fair, the early wave of generative AI tools couldn’t do much to help us spend better time with clients,” she said. “That’s mostly because LLMs are unpredictable, at least on their own.”
READ MORE: When AI wastes more time than it saves for advisors
In the beginning, there were large language models. And they were limited. But as each wave of artificial intelligence innovation arrived, the capabilities became more robust.
In November 2022, when ChatGPT was launched, it was just an LLM and a user interface powered by previous internet search data. What it lacked, Young said, was context.
Cloud providers started to create model marketplaces that made it easier for advisors and developers to find the right model and test it, said Young. By shifting the data sources to be more focused and current, innovations sprung up in its wake.
By early 2024, several AI startups came online, including note-takers Jump and Zocks. Developers then started to experiment by making LLM tools available within software applications and on third-party platforms, said Young.
Over the past couple of years, Young said the quality of LLM outputs has improved substantially. That’s partly because of the LLMs themselves, she said, but also because firms are starting to use the data that they own in-house, making them far more useful for their own purposes.
“That’s important, because while LLMs are a black box, the applications that use them don’t need to be,” she said. “We’re starting to see … the true potential of AI come into focus.”
Now, agentic AI can take on multiple layers of complexity, not just to perform a single task, but to reach a desired goal like deepening client engagement.
From assistance to autonomous agents
Now that AI tools have moved beyond solving for separate tasks, advisors can use the exponential effects of tools to improve client connections.
Reasoning models — which can complete more complex processes because they go through intermediate reasoning before they spit out their final answer — will fuel the next wave of innovations in AI, said Young.
“Memory is how AI systems keep track of their tasks and evolve,” she said. “LLMs on their own have no memory, and so this capability helps them do a much better job of responding to feedback from the user about how they want to see better, different answers in the future.”
What these AI agents can automate depends on the tools plugged into them, said Young. An example would be using a web search and calendar tool to find webinars about important product announcements. Another would be combining optical character recognition, which converts scanned documents into text, with a tool that can analyze client complaints about problems with their statement reporting and automatically create a ticket to send to the back office.
To illustrate how advisors could use these advancements, Young went through the steps of building a meeting preparation agent in Microsoft’s Copilot studio.
An advisor could spell out a series of instructions that could allow a series of sub-agents to perform different tasks under one umbrella.
To prepare for the client’s annual review, an advisor could tell the AI agent to look at their Microsoft Outlook calendar to figure out when the meeting is happening.
Once that’s done, the advisor could have a subordinate agent, called a correspondence review agent, to review all the recent emails and meeting transcripts with the client. This sub-agent would be directed to create a detailed summary of questions and concerns the client previously raised. This information could then be used to highlight any tasks that were supposed to be completed by the advisor or the client.
Other sub-agents would then be triggered by the workflow, as outlined by the advisor. A financial plan agent could chart the client’s progress to their retirement goals. A portfolio review agent would summarize portfolio holdings and performance to generate a narrative. A marketing agent would identify recent newsletters and upcoming events and webinars to highlight.
All this would then be compiled into a briefing memo and delivered to the advisor before the meeting.
“It refines the output from each agent so that the complete narrative has each agent using the context from the other agents, so that you get a much more integrated and holistic answer,” she said.
Young said to make the most of these tools to deepen the engagement with their clients, advisors should ask themselves who their target clients are and how they are differentiating themselves. Once that’s done, she said they should take a hard look at how they can use their existing data sources to build AI agents that could help them show how they stand out.
“Think about how you can bridge the gap between IT and the business,” she said. “They are great at building tools, but they rarely understand the big picture.”
