Amazon.com Advocates Crowdsourcing, Mechanical Turk in NYC

NEW YORK – A million strangers could benefit a company more than a handful of highly trained employees, at least if you subscribe to the concept of crowdsourcing. In the case of Amazon.com and its Mechanical Turk, the online retailer’s vice president Sharon Chiarella told the audience at the Global Sourcing Forum and Expo here, crowdsourcing drove down costs while increasing efficiency—a model nearly organic in its adaptability to the day’s business conditions.

Amazon.com developed its own crowd to "make merchandise more discoverable" and "improve the customer’s experience," Chiarella said. The crowd can clean data, categorize data for better searching, provide metadata, and even scrub user-generated product reviews of any inappropriate content.  

Based on that experience, Amazon.com "decided to expose the crowd to third parties. We started with third-party partners, and then built a platform: Amazon Mechanical Turk," which bills itself as "A Marketplace for Work."

Mechanical Turk currently utilizes 400,000 workers in 100+ countries—that number fluctuating according to the available work on-hand—and providing labor within a model that provides real-time economic feedback; if a third-party partner asks too little for a particular task, the full power of the crowd will not be applied to the same degree as when the price is right. For businesses using the marketplace, Chiarella said, "The crowd offers the ability to scale very rapidly."

Chiarella cited several other examples of successful crowdsourcing, including the development of open-source software platforms such as Apache. The Netflix Prize, where the online-rental company offered a million-dollar bounty for whomever could improve its movie-recommendation engine by 10 percent, was another case where a crowd helped with an issue that an organization could not solve internally. 

"What Netflix did was, they let the information outside of the Netflix corporation and onto the Internet," Chiarella said. "They exposed internal private data to help develop this algorithm."  

Chiarella summarized the advantages of crowdsourcing as follows:

• No Contract Negotiation: The company can make it clear up-front that workers will only be paid on satisfactory completion of the task at hand.

• Variable Cost Staffing: With crowdsourcing, Chiarella said, “If there’s no work in the system, you’re not paying workers; if you have work, you can suddenly scale up to 900 workers.”

• No Recruiting: Crowdsourcing drives down the need for recruiting overhead.

• Pay for Performance Model: A simple pay structure based on a set amount of payment for a set task.

• No Facilities Management: The crowd working off personal PCs and networks will translate into no overhead for facilities.

• No Training Lead Time: Amazon.com found that crowdsourcing freelancers had a tendency to train each other. When its Kindle e-reader device was first released, freelancers started a Wiki of helpful material to order to help the support community solve customer issues.

• Geo-Political Diversity: Having workers dispersed around the world also allows work to be transferred fluidly within the system, should a bank holiday or natural disaster in one part of the world effectively shut down a country.

• Scale Up/Down Instantly: Some days, Amazon.com needs the crowd to “scrub” a flood of user reviews and other tasks; some days, the number of tasks is low. In either case, by utilizing the crowd, the company can meet the daily work demand without having either too many or too few actual staff.

• Speed: The crowd performs task in parallel, reducing the amount of time necessary to reach a goal. By way of example, Chiarella cited the case of NASA, which released an application through its Website for counting meteor strikes in 88,000 images of Mars. That computation would have taken a NASA scientist two years, but the crowd completed it in a month, with no loss in accuracy.

For tasks such as that NASA project, the crowd gravitates toward the task out of passion for the issue; but Amazon.com’s Mechanical Turk platform motivates the crowd primarily by offering compensation. In theory, the pay-for-performance model keeps the crowd producing quality results; Amazon.com has found that workers trend toward specializing at very specific tasks, becoming more skilled and producing better results.

Wikipedia is another example of an organization whose crowdsourcing has begun to specialize; certain groups updating the online encyclopedia are editors, some are writers, while others try to ensure that entries are free of bias.

Ultimately, Chiarella said, crowdsourcing is an "improvement-based culture" that essentially self-polices. As a service, however, Mechanical Turk still has a small audience—at least when you consider the sheer size of the Internet: on Nov. 11, some 58,939 HITs (Human Intelligence Tasks) were available; back in the summer of 2008, that number was roughly 12,000, according to an earlier eWEEK examination.

The kinks in crowdsourcing may be ironed out by the wisdom of the very clouds driving its processes; the question, however, may be how much a service can truly catch on, and what added factors may be necessary to make it explode.