NERD Alert: S&P Taps Kensho to Add IDs, Data Links to Mentions in Historical Transcripts

S&P Global Market Intelligence will use technology from Kensho, the artificial intelligence startup it acquired in 2018, to apply tags and identifiers to its history of transcripts of earnings calls and corporate events, to make it easier for investors to find information on companies, executives, and competitors hidden within those datasets.

S&P’s Transcripts product already provides IDs for the company hosting the event being transcribed, the event type, and individual speakers, assigned by S&P’s analysts, for all transcribed events dating back to 2004. Kensho’s entity recognition tool, dubbed NERD (Named Entity Recognition and Disambiguation), will allow S&P to comb through all transcripts dating back to 2004 and identify—and link to data about—specific companies mentioned during an earnings call, using S&P’s existing Global Entity ID company identifier codes.

“We are going to incorporate Kensho NERD into our transcription offering over the next six months. It will run through all our historical transcripts and identify all entities mentioned during the calls,” David Coluccio, managing director of data management solutions at S&P, tells WatersTechnology. “For example, we will assign the Amazon company ID for their conference call, but if [Amazon CEO] Jeff Bezos mentions Microsoft during the call, NERD will identify Microsoft and tag it with our company ID. This will allow a user who is interested in knowing when another company mentions Microsoft during their call to see the actual transcripts and the exact location that the company was mentioned.”

Once this project is complete, S&P will then focus on tagging individuals and places mentioned on—but not part of—earnings calls. NERD would identify specific individuals mentioned by name, such as company management, board members or advisors, or key executives at competing firms.

This could be important because an executive joining or leaving a company—or a well-known industry figure endorsing or criticizing a company by name—can have a material effect on its fortunes. Alternatively, a user may not have a direct interest in the company holding the earnings call, but may be searching for mentions by other companies of the stocks in their portfolios—for example, if a company discusses the user’s investment as a partner or as a competitor.

This will also make it easier for users to search hundreds of thousands of transcripts to find information on the companies they are interested in. For example, it even allows users to understand who asks a question, and how management answers it. “Maybe it was a very negative question from a sell-side analyst that had a buy recommendation prior to the call. Will the analyst downgrade the stock after the call? Or maybe it was the way management answered a question about their employee work environment that provides additional insight into the company’s corporate governance practices. Clients can back-test these theories on our [historical data] to determine if it works,” Coluccio says.

Kensho will go through all of S&P’s current and historical calls that they have saved to identify and tag all entities that were mentioned, leveraging S&P’s company IDs and also its database of short names and aliases so that NERD can understand who or what company is being discussed, even if the speakers don’t use a person or company’s full name, he says. This will allow users to identify the exact section that an entity was mentioned during a call.

“Since S&P Global Market Intelligence has a team that maintains a professionals database, we know the different names individuals may use while they work at different companies (John Doe, J. Doe, Jonathan Doe, etc.,). Since we capture the names over time, we will accurately identify and tag them to our speaker ID. … Because these variations in names are linked and structured, it’s easy for computers to understand who a person is, and where they work,” he says, adding that this kind of deployment demonstrates the importance and value of the S&P-Kensho deal to both parties. “Kensho had the technology and AI, but not the data. S&P had the data, and so now Kensho has the data to train its models on.”

Coluccio says S&P will also run NERD on its Machine Readable Filings product of SEC filings in machine-readable format, which the vendor launched earlier this year in partnership with Social Market Analytics, and on any other textual dataset that S&P brings to market in the future.

In addition, the vendor plans to offer NERD as a service so clients can run it on their own text data. “We have been utilizing the Kensho services with S&P Global for a few years, and now are looking to offer it to our clients through the S&P Global Marketplace,” Coluccio says.