When Raj Neervannan, CTO and co-founder of financial search engine company AlphaSense, thinks about search, he thinks about it “as a killer app that is only growing…..People want answers, not noise. They want to ask more intelligent questions and get to the next level of computer-aided intelligence.”
For AlphaSense’s customers – analysts at large investment firms and banks or any other industry, as well as one-person shops – that means search needs to get them out of ferreting through piles of research docs for the nuggets of information they really need. Neervannan knows the pain of trying to interpret a CEO’s commentary to understand what he or she was really saying when making the point that numbers were going down when referring to inventory turns. (Jack Kokko, former analyst at Morgan Stanley, is AlphaSense’s other co-founder.)
“You are essentially digging through sets of documents [using keyword search], finding locations of terms, pulling them in piece by piece and constructing a case as to what the company’s inventory turn was really like – what other companies’ similar information was, how that matches up. You have to do quantitative analysis and benchmarks, and it can take weeks,” he says.
Clearly, avoiding having to dig through multiple documents using keywords (and multiple variations of them) in favor of getting a quick summary of what you are looking for and where it can be found across documents would be a help, taking the analysis process from weeks to minutes, perhaps even seconds – for those in the financial industry or any other sector. AlphaSense’s advanced linguistics search engine leads to those ends.
“It just quickly summarizes that what you are looking for is in the 13th line of the 14th page of the 10k filing from the last quarter,” he says, “and also that the CEO had a discussion about this topic in this conference call, and here is the part in the transcript part where he said this in response to a question. And that his peer [at another company] had a conference call where he said this about that. Then the analyst can look at the two companies’ inventory turns back to back to back in one shot” for speedy comparisons to help with analysis.
Not only that, but it buys them back time to do more in-depth research that should result in higher quality analysis. That, he says, “could make a huge difference in decisions about investing in a particular stock or not.”
How AlphaSense arrives at this starts with curating the kinds of information analysts should even be looking at – SEC filings, press releases, conference call transcripts, for example – and what to ignore to keep from drowning in noise. “You start off with critical sources of primary and secondary information,” he says. “Every source we have we want to add value.” Next comes budgeting to a higher level of resolution of text information – that is, it builds upon a base natural language processing technology to look for the tense and meaning of a word in the context of financial terms.
“People don’t just want keyword guidance but guidance in the form of concepts like revenue, profit and various important terminologies, including those associated with futuristic words like expect,” Neervannan says. “People are looking for what is going to happen because the stock market is a function of future expectations. Essentially the concept of time is involved, so you have to build an information architecture that understands at the word and grammar level what is the time component here and what is the tense, as well as tone and meaning.”
Backlog, for example, means expected revenue in sales language, but an entirely different thing in the context of inventory. “The information architecture has to build these topic semantics,” he says. “Our taxonomy had to be built out and NLP had to be built out to extract tense, tone and other related grammar. All this information, along with financial topics plus curated sources, now gives you a better sense of what to look at and how.” While AlphaSense provides industry-specific and ready-made queries, terms and taxonomies to start users off quickly, users also can build their own curated complex search queries and powerful taxonomies.
As an analyst, “you can’t be satisfied by pulling the balance sheet alone – that’s all done,” says Neervannan. “In the day and age of Big Data so much more is available.”