For anyone who has ever spent hours Control-Fing through research reports and earnings transcripts, Bloomberg is offering an alternative.
The company rolled out AI-Powered Document Insights on Monday. It’s meant to help financial professionals quickly find information from company documents by asking questions in plain English. It works alongside Bloomberg’s AI-generated earnings call summaries, news summaries and analytical tools for users of its terminal.
The announcement is the latest of several rollouts of generative AI models for investment and deal research. Factset, Brightwave and Octus offer help with searching for information and creating deal and pitch books.
“There is tremendous demand for generative AI tools across Wall Street that enhance productivity with clear, measurable value,” said Sumeet Chabria, CEO of Thoughtlinks, an advisory firm that works with banks. “Wall Street is challenged with margin pressures, economic headwinds, global dislocations and large operational teams, so cost and efficiency remain a top priority.”
Gathering information from company documents is a laborious process, Chabria said.
“Analysts spend hundreds of hours dissecting earnings transcripts, annual reports and news feeds,” he said. “Internal search engines, like external ones, are still not fully optimized for information discovery and lack the ability to extract insights from data. I get asked a lot, ‘Where is the real value in generative AI?’ This is one area where it can significantly enhance productivity, and the value is measurable. It’s high-paid knowledge work.”
Suzanne Szur, research and companies product manager at Bloomberg, said analysts spend at least 50% to 60% of their day reading and looking for information, “and there’s a huge opportunity for us to make that much more efficient for them.”
“Ultimately, you’re trying to either form your investment conclusion, your thesis, or if you have a thesis already, you’re constantly looking for ways to either validate it or challenge it,” Szur said.
The transcript summarization feature Bloomberg released a little over a year ago was the first effort toward that efficiency, she said.
The new document search features help users find information in those summaries and the underlying source documents simultaneously.
Bloomberg displays summaries of earnings calls and other transcripts side by side with the original transcript, so users can see where each point of the summary is in the transcript. The audio of the call is aligned to the text, so users can hear if there was hesitation in the way management responded to a question or the tone of voice used.
An ask-a-question feature lets users query documents in natural language. Users can ask factual or document-based queries, not subjective questions like, “Should I buy this stock?”
“If that question was asked, our safety model would say that’s not a question that we answer, we answer fact-based questions or or questions to where we can give somebody else’s perspective,” said Wayne Barlow, global head of core product at Bloomberg.
Beta testers reported that the tools help expand coverage beyond their core portfolios and save hours per week, according to Barlow.
Bloomberg’s document search “provides a great summary of a company’s comments when asked a direct question,” said Magdalena Richardson, credit trading strategist at NatWest Markets, in a statement. “For example, within an earnings call transcript for an original equipment manufacturer, the solution can be asked about the impact of tariffs. I think this solution is a godsend on a day when you have multiple earnings calls and cannot attend them all or need to revisit a topic.”
Bloomberg uses a combination of commercial, open-source and in-house models to generate summaries and search results. An AI safety team works to check that outputs are appropriate, trustworthy and fact-based.
The ask-a-question feature currently works with transcripts and will soon expand to filings, presentations, sell-side research and proprietary internal research. Multi-document analysis and alternative data integrations are also upcoming.
“We see AI as being tremendously important for our customers, for our industry,” Barlow said. “We are working hand in hand with our customers to make sure that we’re evolving the terminal to meet their expectations.”
A year ago, Bloomberg was processing more than 300 billion market data points of structured data every day. Today, this has grown to over 400 billion data points that are distributed to clients in 170 countries. During periods of market volatility this can jump to nearly 500 billion ticks per day.
Bloomberg’s primary competitor, FactSet, began offering generative AI for research analysts and junior bankers in December 2023.
In January, FactSet launched Pitch Creator, a tool that reduces the number of clicks and keystrokes required for pitchbook curation by roughly 80% or 10 banker procurement hour savings per week, according to Kendra Brown, senior director of banking and sell-side research.
For all such uses of generative AI on Wall Street, “a key challenge remains whether a human being is able to verify that the information sourced and summarized is 100% accurate and not rubber-stamped,” Chabria said. “Accuracy and precision are key in financial services.”
AI-powered financial research will completely disrupt the research and analyst industry as we know it today, Chabria said. “This is tedious, high-paid knowledge work,” he said. “Market data providers have a huge advantage: their own data. If they do not allow easier, faster, more insightful access, they are ripe for disruption.”
