The average financial institution exchanges anywhere from a million to three million e-mails a year. Buried within all the missives about meetings and lunches might be a few damning indicators of a brewing fraud.

A company called Digital Reasoning hopes to help banks find this critically important information with machine-learning software that raises red flags found in messy, or “unstructured,” text data, including e-mails, tweets, and document files. The software uses statistical models to break down sentences and infer their meaning. This is important because finding warning signs may not be as simple as matching a string of text.

The company aims to help banks benefit from the exploding volume of data available to them that is, at best, underutilized today. More and more, businesses view data as a raw material that can be turned into a valuable resource—if only they could figure out how to use it. 

Digital Reasoning, founded 12 years ago, has already worked with intelligence agencies. The U.S. Army began using its software in Afghanistan in 2010 so that officers could weave together insights from disparate intelligence documents, says CEO Tim Estes. After searching for an insurgents’ name, for example, the program could turn up the person’s known aliases or his connections to other people or groups. And the output could be organized on a map of the country. In-Q-Tel, a venture capital firm backed by the CIA, and Silver Lake Partners have both invested in the company.

Now Estes hopes financial institutions will buy the software. After a few months of meeting with technology executives at some of the world’s biggest banks, he believes there is a “pressing need” for the service.

Financial institutions could save billions by spotting fraud or insider trading by employees, or catching financial advisors giving unethical or illegal advice. U.S. Senate hearings later revealed that in 2007, before the financial meltdown, Goldman Sachs employees wrote e-mails bragging of selling blatantly terrible investments to clients. Estes says his software could have helped compliance departments catch such activity.

Digital Reasoning recently took part in the FinTech Innovation Lab demo day, an annual event dedicated to financial technology and organized by the New York City Investment Fund and Accenture.

Cris Conde, the longtime former CEO of the Fortune 500 software company SunGard, who advised all of the companies in the Fintech program, says Digital Reasoning’s services are hugely needed in the financial sector. Surprisingly, he says, banks are far behind the technology curve in analyzing the exploding terabytes of information that don’t fall neatly into spreadsheets or involve directly “optimizing” a transaction.

Digital Reasoning’s software uses training algorithms to read and assemble masses of e-mails and other text data, and then sorts the words and sentences into organized relationships that can be compared in context. This data can be searched, or alternately, be used to trigger preset alarm bells automatically.

HP, Microsoft, IBM, and Oracle have all developed similar text analytics software. But these programs mostly analyze whole documents based on preset rules rather than using algorithms to break down words and phrases in context. And they generally require a human to read actual text once groups of documents are clustered, says Estes. In contrast, Digital Reasoning’s software can directly display the relevant text data and linkages—as if the user already took notes—and might deliver crucial information in minutes, Estes says.