Morgan Stanley’s gen AI launch is about global analysis

When Morgan Stanley announced its new generative AI support tools for financial advisors last week, it talked about gaining efficiencies from its notetaking abilities. But those who track the $54 billion financial firm said that its goal is much larger.

The announced intent of AI @ Morgan Stanley Debrief, one of a suite of generative AI tools the company is developing for its financial advisors, is to record, transcribe and then summarize key points from the more than 1 million conference calls that Morgan Stanley people hold every year.

Powered by OpenAI, the Debrief tool will also create an email for the financial advisor to edit and send at their discretion, and record a note about the call in Morgan Stanley’s Salesforce system.

But what makes the move potentially powerful is not the capturing and summarizing via AI, as that has been done by businesses for years through Apple and Android devices, conference calls from Zoom, Microsoft, and Google, and a wide range of independent apps. The real benefit will come from every Morgan Stanley employee and contractor using the exact same package for those summaries, which means that the data will all be in the same format and can therefore be analyzed comprehensively.

It will mean, in theory, that Morgan Stanley management can see analysis of every call made across the enterprise — often within a few minutes of that call’s completion. Are people saying what corporate wants them to say? What are clients emphasizing — or ignoring? What are customers asking and has that sentiment changed from last month?

Deep analytics

Aaron Cirksena, founder and CEO of MDRN Capital, said the real value for Morgan Stanley in this move is getting into deep global analytics insights, building on the hoped-for data consistency through one centralized offering. 

“It is going to make their data analysis far better. It’s big brother basically in that they are going to have an eye on every one of their conversations,” Cirksena said. “Did our people actually talk about the things that they were told to talk about?”

Jonathan Murray, the chief strategy officer at marketing firm Mod Op, agreed that the Morgan Stanley AI move has significant analytics potential. “The interactions most companies have with their clients are typically an untapped source of insight. Morgan Stanley is pointing the way to how new large language models now enable these interactions to become a rich vein of insight that every organization will want to take advantage of,” Murray said. 

Murray’s colleague at Mod Op, executive-in-residence Monica Richter, added that the potential value will be found in how deeply Morgan Stanley executives probe their newfound data. “Will they use AI to aggregate what is heard across clients to generate ideas for research or buy/sell orders? If a client spoke about starting to set up a retirement account, will AI assist in not just noting that request but automatically responding and emailing out key articles on IRAs, Roths, etc that match the client’s portfolio needs?” Richter asked. 

Discovery

Even if it works precisely as planned, some question whether this analysis could have a downside for Morgan Stanley. Any data that the company can access can also be demanded via legal discovery in litigation, pointed out Rebecca George, the managing director of AI consulting firm Slalom.

What if, George asked, someone wanted to see if Morgan Stanley was giving different advice to clients based on gender, ethnicity, or race? “They are going to be open to potential bias in guidance” litigation, she said. “This is where culture and ethics start to collide. It is all now going to be discoverable. This brings in a whole new level of exposure. Do they have the appetite for that?”

Several observers also noted a strange omission in the Morgan Stanley statement: No reference to how sensitive client data will be protected given the nature of OpenAI systems. 

“Clients are going to ask ‘How are you protecting us?’ Morgan Stanley needs to have a strategy in place,” and to discuss it publicly, said HP Newquist, the executive director of AI consulting firm The Relayer Group.

The biggest questions around data protection or data leakage are around how Morgan Stanley is hosting the OpenAI code and whether Morgan Stanley is interacting with APIs on OpenAI servers.

When asked, individuals familiar with Morgan Stanley thinking on this project said those details are confidential and would offer no assurances that this application is not interacting with OpenAI servers.

Peter Gaugenti, president of Tabnine, expressed serious data protection concerns. 

“From our perspective, sending data to another company’s API for processing and consumption is no longer private,” Gaugenti said. “However, disregard our perspective and ask what a customer would think. Would the average customer be comfortable knowing that their call was recorded, sent over the internet to a third party, and then was processed by an AI platform to document a summary, and that data was then sent back to their financial advisor? Would the typical customer feel that was private?”

Gaugenti said Morgan Stanley should be transparent about how this new application will protect client data. “The conversations you have with your financial advisor are some of the most intimate you can have, and the information covered could do you tremendous harm if exposed. Why would I approve having a complete recording of that conversation shared with a third-party big tech brand, particularly when how that information is processed and used is a complete black box?”

Steady roll-out

The Morgan Stanley statement stressed how integrated this system is designed to be. 

“What sets this technology apart is in how it’s the first to be seamlessly embedded right into Advisor workflows, so there’s no toggling between screens — a common Advisor pain point. All client interactions, with their consent, are captured and logged in our CRM with ease,” the company said. “After the meeting, it summarizes key points, creates an email for an Advisor to edit and send at their discretion, and saves a note into Salesforce.”

Morgan Stanley has been experimenting with this generative AI effort since September 2023, an effort that began with 50 financial advisors (FAs) and ultimately worked with 300 financial advisors reviewing “hundreds” of client meetings, according to a source working with Morgan Stanley on the rollout. 

That Morgan Stanley source was hesitant when asked about the global analytics goal. “That’s not the goal, at least not today,” the source said, adding that initial efforts will not necessarily be reviewed by corporate. “We won’t (initially) have access to an FA’s book. If they want to send it to their branch manager they can, but we are not reviewing the tech output,” the source said. 

Before the new AI app was launched, some financial advisors would take an hour after a call to clean up notes. And some wouldn’t even take notes and would instead have a support professional come into the meeting, solely for the purpose of taking notes and then creating a summary, said Koren Picariello, head of generative AI & execution, Morgan Stanley Wealth Management, who managed the project.

This AI effort “will give time back to that support professional, time that they can spend in a higher-value way,” Picariello said in an interview with CIO.com. 

Other potential risks

Others in the space speculated about potential technical hiccups that may need to be resolved. 

“What happens when someone in the meeting says ‘forget it’ or ‘ignore’? That could effectively lead to a prompt injection risk. But hallucination is the most prominent risk that comes up when summarizing a conversation and action items because LLMs make stuff up, which is a liability. It could lead to mistaken conclusions and wrong decisions,” said Yossi Altevet, the CTO at DeepKeep. “Performance is another important risk because an LLM that isn’t summarizing correctly — like, for example, omitting information or misplacing an emphasis — will eventually lead to a wrong conclusion, customer dissatisfaction and churn. Altevet also pointed to risks arising from the processing of personal identifying information, such as leaks or “drawing conclusions based on personal information leading to biases and breaches in compliance.”

Another concern is how the generated summaries will be verified. In the beginning, it is likely that Morgan Stanley people will be very meticulous in verifying what the app delivers, particularly making sure that nothing important was missed. Over time, though, Cirksena said, people may start to trust the app too much and pull back on time-consuming verification efforts.

“Will they take their foot off of the gas pedal?” Cirksena asked. “It is human nature that they are going to start to get slack over time. Morgan Stanley will need mandates that reviews cannot change in the future.”

Cirksena also raised the possibility of deploying yet more AI to watch the AI. For example, staffers might also run note-taking from Zoom or other means — and then use yet another AI to compare the two results, flagging any differences. “How likely is it that both AI systems will get the same thing wrong?”