Signal AI opens External Intelligence Graph for enterprise use

We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 — 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!

The world is awash in news, and that’s a challenge for any enterprise. Some events are like earthquakes that will shake up the business model and for the company to reinvent itself. Others are inconsequential. Some will hurt competitors and others will help everyone in the same business segment. But how can anyone know which is which? How can anyone spot the moments as they’re unfolding?

This is the challenge that artificial intelligence (AI)-powered, PR and communications company, Signal AI, aim to tackle. Last week, it rolled out its new External Intelligence Graph, a data structure that’s constantly tracking the major and minor events that course through the zeitgeist each day. The system is a constantly evolving natural language model that tracks the way that companies and topics are discussed. 

“You also want to be able to say that, reputationally, your company is doing a lot of good work, but if it’s not actually what external people see, well, that’s probably something that needs to be worked on,” said Clancy Childs, the chief product officer at Signal AI. 

Reimagining PR business as ‘usual’

The company began nine years ago as a media monitoring effort that would gather data from the news sources and social media. It was largely targeting keywords, and it found that there was a ready market for companies that needed to think strategically about their image. 

The new announcement shows some results of the company’s recent $50 million fundraising round last December. At the time, Highland Europe along with Redline Capital, MMC, Hearst and GMG Ventures invested to build better mechanisms for what they were calling, “decision augmentation.” 

The External Intelligence Graph emerged from the company’s efforts to tap into the capabilities of the new emerging algorithms from machine learning (ML). Signal AI’s team wanted to think of text data as more than a stream of characters to be searched, but rather a collection of entities with relationships between them that can be tracked and measured. 

“We’re not going to follow an approach where we make people write massive keyword-based queries to try to disambiguate things.” explained Childs. “We’re actually going to use natural language processing, entity resolution and all these cool toys, effectively to make it easier for people. I don’t want to write a page-long query to explain to you What Apple Computer is. I just want to be able to look for Apple as a trained entity by the AI.”

Signal AI is selling its service both to companies that want to keep tabs on news themselves and investors who wish to help choose potential investments. Some customers are professionals like chief communications officers who aim to track mentions of their own company and their competitors. Others simply want to understand which businesses are succeeding and failing in the world of public opinion, to ensure that their investments are sound. 

These large language models and events are becoming more common. Google reportedly uses its large, internal model of language and the world to guide how it ranks answers for the search engine. Facebook and Twitter essentially sell knowledge of users through the ad market, allowing advertisers to select an audience based on their interests. 

Microsoft and Nvidia recently touted their large model, Megatron-Turing NLG 530B, an immense model of language that has 530 billion parameters arranged in 105 layers. This was the culmination of a research project, but both companies are folding similar results into their products at many levels. 

Some are starting to open up these large systems to customers. Microsoft both helps companies build classifier systems and also bundles prebuilt models into a tool for work like image sorting and classification. Google’s Cloud offers the natural language API that can detect entities and analyze sentiment in raw texts.

Under the hood

The new External Intelligence Graph marries similar algorithms with a large collection of news articles that Signal AI has amassed over the years. Some come from licensed sources like LexisNexis, and others are gathered from the open web through scraping or other techniques. 

Signal AI is selling its service through a web interface and, for some advanced customers, an API. They’re letting companies train basic models of what they want to track, and then it will populate a dashboard with both direct search results, as well as information about how sentiment is changing. 

“Our External Intelligence Graphtakes the world’s burgeoning unstructured content and turns it into actionable insight to augment today’s business decisions, providing organizations with a new kind of in-the-moment critical intelligence.” said Luca Grulla, the CTO of Signal AI, “We are able to provide an entirely new kind of data through our unique External Intelligence Graph, and an exciting new chapter in harnessing unstructured data awaits.” 

While the raw search results can be useful, the more useful insights may come from watching how the External Intelligence Graph evolves. That is, do some companies gain or lose in mentions with positive sentiment. Or do companies grow closer to some topics over time. 

Childs offered an example of the company Tesla. At one time, its name in the graph may be closely connected to electric vehicles. Lately, though, as news about its autonomous guidance algorithms appears, it will grow closer to those entities. 

“These kinds of connections and relationships between these entities and topics make it easier for companies who are interested in managing their own reputation and to identify where they stand relative to their goals,” Childs said. 

The job for company managers has only grown more complicated as some investors and customers have started to ask for better accounting about non-monetary goals like environmental stewardship. Calculating profits is simple. Tracking progress toward building a trustworthy brand, though, is harder. 

“[Many businesses are] no longer just sort of interested in the single bottom line of ‘Are we making enough profit?’” Childs explained. “This gives them quantifiable reputation metrics on things like ESG [environmental, social and governance] which are very helpful for companies that are trying to manage their sort of stakeholder capitalism and ESG responsibilities.”