Morgan Stanley’s investment research chatbot, AskResearch, gets smarter every day.

Developed over the last two-and-a-half years in partnership with Symphony Communications, AskResearch is an AI-powered bot that has been helping the bank’s analyst and sales teams query the near-50,000 reports the researchers generate each year.

Up until now, the chatbot had been an internal research assistant, but the bank is just beginning to embed it within client-facing applications. AskResearch is a text-oriented chatbot at this point—meaning users ask it questions by typing them out—but the bank sees a future in which it will add voice activation in the style of popular consumer voice assistants like Amazon’s Alexa or Apple’s Siri, said Eden Kidner, CTO of equity research at Morgan Stanley, who was speaking at this year’s Symphony Innovate conference, which was held virtually.

“We’ve really now got to the point where we can personalize it—it understands who you are,” Kidner said. “It understands your preferences, and is slowly learning from your behaviors and actions in a way that can make it even more powerful.”

In creating the bot, a primary goal was to make the bank’s research archives omnipresent—“in other words, it would follow you around,” and be looking over your shoulder, Kidner said. Should an analyst or salesperson have a question while just writing an email or sending a chat in Symphony, AskResearch would be there there to answer it. 

During the presentation, Jennifer Casas, vice president of research technology management, demonstrated how the bot understands similar queries phrased in different ways.  So if you ask the chatbot, “Who covers Apple?” it will say, “Katy Huberty”. If you ask, “What does Katy (no Huberty) cover?” you will get all the Morgan Stanley analysts named Katy and their entire coverage areas in a list. But, because you might’ve previously conducted a Google search for “Katy Huberty,” it knows that you likely mean “Katy Huberty” so she will appear first on the list.

Additionally, a company like Apple can have more than 50 fundamental metrics that have been forecasted by bank analysts. So asking, “What are the quarterly estimates for Apple?” will return a precise answer, or, “What is Apple’s valuation?” will provide data that is broken down to explain why Apple is valued as it is.

Beyond equity information, the bot can draw up metrics on the macro side, such as country GDP, for which users can enter the acronym or “gross domestic product” and get the same results. This is the result of a tool the research team created that allows them to create synonyms and acronyms on the fly, which are then stored for future instances.

AskResearch also understands more open-ended questions, like “What should I read today in North America?” which highlights general reports filed that morning that are most relevant to the specific user. And beyond how the bot can retrieve report-specific data points, it also stores information related to events that are important to Morgan Stanley, and can answer a question like, “When is the US announcing policy rates?”

The main challenge in building and maintaining the bot is ensuring that the data—of which there is a vast amount—is tagged in a way that returns specific, concise answers to all the questions asked. In a 40-page report, someone might only need one paragraph or one chart to answer a question, and Kidner says it’s an ongoing piece of work to make the bot find and return only those sections, rather than entire reports, if that’s all the user seeks.

“What the chatbot is allowing us to do is to break that down into small, digestible chunks and get to the deeper information that you want, just in the same way you would ask a person the same question, who would be able to extract the answer,” Kidner said. “It’s ongoing work; it’s complex. But that’s probably the biggest challenge.”