RavenPack Adopts 'NLP-as-a-Service' Model with Focus on Firms' Internal Data

RavenPack, an alternative data provider specializing in news analysis using natural language processing (NLP), has adopted what it calls an NLP-as-a-Service model, which involves scouring through a customer’s internal data, looking for unused, alpha-yielding datasets, rather than chasing only external, third-party datasets.

Peter Hafez, chief data scientist at RavenPack, says starting to consume and analyze internal datasets—such as investment notes, analyst reports, emails, and instant messages—can sometimes yield investment alpha, operational alpha, or risk management advantages.

“We have all kinds of stuff,” Hafez says, “but what is often the problem is that it’s not easily consumable—it’s not democratized, you don’t have access to it easily, it’s not centralized, [and] it’s not structured such that you can start building analytics on it.”

The sell side has been hiring data scientists and building out their proprietary data capabilities, but it takes time and resources to assemble and train those teams. Hafez says there is no need to reinvent the wheel; banks already have these untapped datasets lying in wait, and RavenPack has 17 years of experience in analyzing unstructured data. The vendor could act as a partner to these banks, aiding them in the early stages of their data science endeavors. Then, as the banks mature in their capabilities or their needs become more advanced, they can add their own proprietary strategies on top of RavenPack’s work.

In that case, the service becomes more of a software play. New and exciting algorithms and services that can make NLP easier to perform come onto the market with increasing frequency, but there are challenges associated with them. Often, they aren’t fine-tuned to the financial world, and there’s heavy lifting involved in maintaining databases with continually changing entities.

“That’s where this sort of NLP-as-a-Service comes in. It’s a platform where we would go in and say [to clients]: ‘You have your internal textual content; why don’t you run that through our engine?’” Hafez says. “And what you get back is annotated content, as well as analytics, that make it immediately possible for you to start aggregating it, searching it, democratizing it, and normalizing all of the content that you have into a well-defined metric.”

A possible advantage to structuring this internal information is the ability to track the origination of ideas, Hafez adds. They can discern who began talking about certain topics first, and perhaps better understand who the biggest contributors within the organization are, or quantify the conversations employees are having. For instance, whose conversations yield analytics that can be indicative of stock prices? Is the organization, as a whole, better at talking about certain sectors versus others? How does that translate to performance, and what would improvement entail?

On the buy side, RavenPack is working with hedge funds, which are often interested in analyzing their own news sources, some of which the company doesn’t cover, or which request analytics on certain regulatory filings.

The service is the latest evolution of the RavenPack Text Analytics offering, which debuted two years ago.