Here Lies the CEO: Israel's Amenity Analytics Uses AI to Uncover Hidden Truths

The startup’s tools scan news articles, financial reports, transcripts of conference calls and other texts to find things human financial analysts miss

The team of Israeli-U.S. startup Amenity Analytics
עופר וקנין

Investment houses and hedge funds employ armies of analysts to sift through tweets, media reports, quarterly earning statements and transcripts of investor conference calls. If they succeed, some seemingly obscure nugget of information could give them an edge over other investors.

The problem is the tedious and time-consuming work. Human analysts are prone to making mistakes and overlooking important details, if only due to the sheer quantity of material to digest and analyze.

Israeli-U.S. startup Amenity Analytics has a solution for speeding it all up and getting better results by employing artificial intelligence. Founded three years ago, Amenity has developed a dedicated set of data mining tools for financial documents.

“Teams of analysts follow scores of companies and read all the material they produce, and that way they can look for clues and new trends,” said Roy Penn, Amenity’s vice president for engineering.

“For example, Nvidia spoke over the course of several quarters about the gaming market and then suddenly stopped. The analysts might not notice that the company had stopped talking about it, but our system could, by analyzing previous quarters and generating insights about something suspicious.”

Penn said Amenity can even measure the reliability of CEOs, for example by whether they answer questions directly in conference calls or try to avoid them.

“We have identified approximately 50 different ways a CEO can avoid answering a question. One of our customers calls it the ‘bullshit detector,’” Penn said with a smile.

>> Debunking the 'Startup Nation' myth: How Israel's economy cut itself off from the world | Analysis

“For instance, if a fairly honest executive, who only avoids answers 10% of the time, is suddenly lying four times as much, I would want to look at the questions that he didn’t answer. That’s critical.”

The company was formed after the two future founders were introduced to each other — CEO Nathaniel Storch, an American whose background is in finance and markets, and Prof. Ronen Feldman, an Israeli expert on data mining who is now Amenity’s chief scientist. They were joined by Feldman’s wife, Hedva, who is responsible for Amenity’s strategic relationships.

Amenity raised its first capital in August 2017 when it raised $7.6 million from a group led by Israel’s State of Mind Ventures and Intel Capital, the investment arm of the giant U.S. semiconductor maker.

Today it employs 60 people, two-thirds of them in Israel. The company doesn’t reveal its revenue figures and says that all profits are plowed back into the business to help it grow.

“Intel decided to invest in us because of Pokemon,” said Chief Product Officer Mati Cohen. “They asked us to do a project on augmented reality and how this trend was talked about in the news. Our system found that it was Nintendo that started the conversation in some article about augmented reality about a year and a half before they launched the mobile game Pokemon Go.”

>> Netanyahu shares responsibility for Israel's growing deficit | Editorial

Cohen stressed that artificial intelligence couldn’t predict when a game like Pokemon Go was due to be launched but it showed Intel what Amenity’s technology could learn text-rich material.

The company itself is sometimes called the “8200 of hedge funds,” in other words providing the same kind of high-tech intelligence the Israel Defense Force’s storied unit does.

Its customers include the Nasdaq and the credit rating agency Moody’s. “These are companies that operate like countries. They have intelligence units and big data staffs, but even they sometimes reach a dead end. That’s where we enter the picture,” said Penn.

Amenity’s system is based on natural language processing, which is concerned with the interactions between computers and human languages, including how to analyze large amounts of natural language data.

NLP engines analyze each sentence as part of a paragraph and each paragraph as part of a document, to learn the full context. At the same time, each sentence is broken into its components and analyzed syntactically and grammatically.

“In a sentence like ‘Toyota declares a recall of its most successful and best-selling care,’ most NLP engines will analyze it positively,” said Cohen. “It includes the words ‘successful’ and ‘best-selling.’ It’s a complicated sentence, but we would be able to understand that it’s a negative event for investors because it’s talking about a recall.”

Can AI understand sarcasm and metaphors? Penn said it can identify metaphors and imagery and even catch the tone of speech based on the length of an answer to a question. “Sarcasm is more difficult for it to identify, but you need to remember that we’re talking about the American financial sector — we don’t encounter almost any cynicism,” Penn said.

There are other ways a close reading of events can be put to work to the advantage of the user. For instance, if a big company fires its CEO, that would normally be taken as a negative for investors, a sign the company is failing and needs new leadership. But if the user is a financial services company, it could be good news, if it turns out the ex-CEO received a $40 million golden parachute, for example, and will need someone to manage his or her money.

The company’s next big challenge is to develop systems that can be used by smaller hedge funds and so-called family offices that manage the assets of wealthy individuals. It’s also looking at the insurance and automotive industries.