Kensho, a cloud-based intelligent computer system is rolling out a market research assistant called Warren, capable of answering highly complex financial questions posed in simple English.
Along the lines of Apple's Siri, or IBM's Watson, Kensho's Warren is designed to answer any factual question we can ask of it in our native tongue. For example, according the recent press release announcing its success in securing $10 million in seed funding, Warren can currently answer a million distinct types of natural language questions about the impact of global events on asset prices, such as, "What happens to the share prices of energy companies when oil trades above $100 a barrel and political unrest has recently occurred in the Middle East?"
Detailed statistical relation questions along this vein may have taken hours, even days, for a researcher to answer. Warren needs only a few seconds, if that. Once answered the researcher and traders can move on to the next phase of their investment process, which may include asking more questions about oil.
The name "Warren" just came to Daniel Nadler, founder and CEO of Kensho who says he wanted a name that evoked an avuncular figure, one that is wise and can give answers in friendly, applicable and approachable way. Nadler adds that the name was not literally or directly taken from the famed value investor Warren Buffet, but he welcomes that natural association.
Learning of the Future
"Once you get the systems going, it becomes self learning," explains Nadler. "They just take off exponentially." The path of development is completely dislocated from how fast the human brain has to learn things. "You and I can read a page in a minute, but Warren can read hundreds of thousands of pages in seconds."
For example: Imagine you asked Warren what happens to homebuilder stocks following a category four hurricane touchdown. Maybe Home Depot stock jumps. Then you may ask, what about earthquakes measuring a seven on the Richter scale? Initially, Warren may not understand the Richter magnitude scale so it goes out and crawls the web, learning all it can about earthquakes and statistical correlation, so the next time anyone asks that question it's going to be able to answer with confidence.
After four months of beta testing by several hundred users, with over fifty thousand distinctly new questions added every week, the system continues to be smarter. "That's the difference between cloud-based systems and classic hardware on desktops - the smartest it will ever be is the day you bought it. Cloud-based intelligence systems get smarter as users use it and ask questions."
And how does Kensho feed data to such an intelligent machine? The research and analytics platform was built on NASDAQ OMX FinQloud, a cloud computing platform designed exclusively for the financial services sector.
Applications like Siri use the cloud, not the hardware, to answer questions. "As a user of Siri you don't care about where data of Abraham Lincoln's height is stored. Siri answers 6'4 and that's okay. But financial services professionals want to know how defense stocks react when North Korea tests missiles, and all that data needs to be stored in a secure way."
"FinQloud solved a critical piece of the puzzle," notes Nadler, who says one of the earliest obstacles was building a distributed cloud-based information network that stored financial service data, market data, asset prices and proprietary data around events. He adds that the leveraging of the building blocks like cloud hadn't been done before in a way that was secured and complaint enough for financial service professionals.
"Building a virtual market research assistant who you can talk investment ideas over with, and to whom you can express complicated, multi-conditional questions that draw on both structured and unstructured data (from the history of all asset prices to encyclopedias of global events) constitutes one of the most significant engineering challenges in the history of financial technology" Nadler said. "It will require that an intelligent computer system read millions of pages of natural language documents and have total recall of petabytes of financial data, analyzing the cross-implications in a matter of seconds. Computing these massive datasets in near real-time, and performing split-second investment analysis by searching for correlations between unstructured and structured data is extremely computationally intensive, and will require sophisticated distributed computing environments. In overcoming this colossal technical challenge we are pleased to continue our deep relationship with NASDAQ OMX FinQloud, which augments our cloud-computing infrastructure with the additional financial technology necessary to meet the very specific security and regulatory obligations of financial services."
Perhaps the greatest benefit for companies using Kensho is that a Finqloud client can essentially plug-in their own proprietary data into the system's ability to interact with assets and pricing data to ask more detailed questions and gain deeper insight in their areas of interest. "Yes, people use the base data, but all sorts of companies will have spent time gathering proprietary data, and instead of taking forever to figure out how it works with asset classes to trade better, we feed that into the intelligent assistant."
According to the release the system was built by several veteran software engineers from Google and Apple, including one of the original engineers on the first iPhone team. Although Kensho's Warren project started about a year ago, it took off in the summer when the cofounders left their roles to dedicate themselves to the project.
Beta testers have been a vital part of its development and Nadler says there has been no shortage of applicants. The lucky ones selected to participate, he says, are extremely engaged in terms of the goals of the project and eager to ask the questions that expand Warren's base knowledge.
Kensho continues to invite applicants from all parts of the financial service industry to the beta program, but the rollout of Warren has been gradual, with no fixed date for an open release. Those interested in helping to beta test Warren in the workflow of their organization can apply on their webpage.