In a slowly emerging labor trend, full-time employees are being replaced with freelance workers connected via the web.

New York University professor Clay Shirky says this is due to "cognitive surplus" — the ability of the world's population to collaborate on large, sometimes global projects. "The world has over one trillion hours of free time to commit to shared projects," Shirky said in a Ted talk. This is how Wikipedia got built. It's also how cat memes are made, Shirky notes.

While Max Yankelevich was a researcher at MIT, he and his colleagues deemed this trend a major shift in the global workforce.

"In 2008, when we were losing jobs, we were gaining about a million jobs in freelance economy," he says.

Yankelevich has parlayed this idea into a business called CrowdComputing Systems, which Wednesday announced a solution for financial information providers and financial institutions called WorkFusion. The platform combines crowdsourcing and automation tools with machine learning to increase productivity.

"The big problem with leveraging this labor, because these people are so distributed — millions of people available around the world, people in Asia, stay at home moms and retired people in the U.S. — is being able to aggregate the labor and bring it to the enterprise in a manner they can use for their business processes," says Yankelevich, CrowdComputing's CEO. "And being able to distribute this work to that labor and validate that these workers are producing output the company can use."

Yankelevich says his company is disrupting the business process outsourcing market, which is worth $200 billion a year.

"Banks are huge users of BPOs and captive labor," he notes. "They have a lot of human intensive processes like Know Your Customer compliance and transaction reconciliation in the call center." However, Yankelevich points out that the cost of labor in traditional business process outsourcing meccas like India is rising and that outsourcing arrangements often aren't flexible enough to accommodate slow periods and spikes in work.

CrowdComputing uses Amazon Web Services to make its software accessible to its database of 20 million freelancers. About 10% are in the U.S. Some are part of existing large communities like Facebook and Zynga, or from freelance labor pools like Elance, Amazon Mechanical Turk, or oDesk.

The software breaks up work processes, such as vetting a new business customer for KYC purposes, into tiny pieces or "microtasks" that can easily be handled in parallel by several people. One person might research a potential client on a government website; another might read an article about the company.

"Microtasks are doing for knowledge work what Henry Ford did for the car industry," says Yankelevich. Assembly lines make it possible for each person to handle a small component of the work without being an expert on the entire product. "That's how you get scale and efficiency," he says.

The core of this platform is an artificial intelligence engine that finds workers, breaks processes into small tasks that can be done in a short time, evaluates the work, pays and motivates the workers, and aggregates the work in a form that's useful to the corporate customer.

CrowdComputing trains workers and examines their output. If they do well, they get more work. If not, they're trained again. They're also paid for performance — if they do a job well, they're paid immediately. If they mess up, they might get nothing or half the pay of a properly completed task. And workers are only paid for the time they're productive, not during downtime.

"This is attractive to enterprises because they're not spending money on capital expenditures, office space, computers, and down time," says Yankelevich.

The software gauges work quality in several ways. It uses probabilistic analysis, which means it looks at 60-90 data points of current and historical performance, to predict the quality of the work an individual can produce on a specific task with 99% accuracy. It uses plurality, in which it takes an answer from several workers and compares those answers. It uses moderation, in which another person in the crowd will review a piece of work. It uses gold testing, issuing work for which the true answer is known.

Freelancers need to meet age requirements and need a bank account to receive payments. CrowdComputing learns what country they reside in because some processes can't be farmed outside of certain countries. "We know what every worker is good at — we do that with upfront and in-place testing, as well as watching how they perform in work," Yankelevich says.

Sensitive data can be encrypted or obfuscated from workers. Also, the shredding of bigger tasks into small pieces protects against any one person seeing the big picture.

For one customer, CrowdComputing created tear sheets — information that an analyst or investment banker uses to make a decision. Information is pulled from more than 1,000 publicly available sources. "Usually a company would outsource this — one company was using 3,000 people in India to create these tear sheets," says Yankelevich. "We were able to reduce the price per tear sheet from $3.50 to about 22 cents, and quality of output went up from 80% to 95%. Time to market was also shorter."