CloudCrowd: An assembly line for content

New "wide-sourcing" apps lower cost, improve timeliness of rote text jobs like translation and proofreading.

Rafe Needleman Former Editor at Large
Rafe Needleman reviews mobile apps and products for fun, and picks startups apart when he gets bored. He has evaluated thousands of new companies, most of which have since gone out of business.
Rafe Needleman
4 min read

"We're doing to service industries what China did to manufacturing," CloudCrowd CEO Alex Edelstein tells me.

That's not exactly a warm and fuzzy concept for a journalist, as the particular service industries that CloudCrowd affects are in the writing and editing space. CloudCrowd collects tasks from clients--currently translation and proofreading jobs--via its Serv.io site, and then distributes them to waiting workers via a Facebook app.

Unlike the content farm Demand Media, where a computer assigns writing jobs to waiting workers based on popular search engine queries, CloudCrowd creates work when paying customers ask for it, so it's unlikely to be squashed by Google (still a risk for the newly public Demand Media). And unlike Mechanical Turk, which fractures jobs into minuscule Captcha-sized pieces and then simply reassembles the work units for customers, Serv.io jobs go through a fairly rigorous work flow, incorporating several layers of human workers double-checking and smoothing out each others' piecework.

CloudCrowd CEO Alex Edelstein in the company war room, showing how translation tasks bounce around the globe. Rafe Needleman/CNET

CloudCrowd's process breaks moderately complex editorial tasks into small, repeatable pieces that can be done by specialists. Some people make better first-pass translators, for example, than assemblers of the translated text. CloudCrowd may take the fulfillment out of building a finished product, but it also, says Edelstein, makes construction of products faster and leads to a more consistent quality level, as the Model T assembly line did when it was fired up. And, like industrial manufacturing did for consumer goods, CloudCrowd puts significant downward price pressure on the product it creates. Translation tasks go for 6.7 cents a word, less than a third the price traditional translation agencies charge. Turnaround is fast, too, at less than 24 hours. Even faster service will soon be available, Edelstein says.

Enter the 'garble hunter'
The Serv.io translation service begins by first doing machine translation of the job in question. Then, a "garble hunter" (a person) picks out the sentences that need further refinement. These sentences go to two translators in sequence, their work then goes to two editors in turn, and finally the finished product is delivered to the customer. Even with all these hands on the product, the cost to the consumer is low, thanks to efficiency of the work flow system, its capability to quickly and automatically direct work segments to specialists, and its capability to quickly tease out the pieces of the job that need human help compared to those that work fine with the machine translation (generally about 70 percent of any translation job).

As a content worker myself, I had to ask Edelstein: Doesn't spreading jobs out to pieceworkers around the globe lower the actual amount of money a person can make?

"We pay above market rates," he told me. Perhaps true, but his service does level fees globally, and it also brooks no inefficiency: Workers must continue to accept piecework jobs and maintain quality (as judged by other workers in the CloudCrowd work flow system) in each tiny task to earn and hold their places in the job queue. It really does make an editorial job into an assembly-line function. It's also true that a worker in a high-cost location may find himself priced out of the market as he's now competing with people in less affluent regions; remote contractor-hiring services like eLance and oDesk have already had this impact, and in those services you hire actual people instead of paying for piecework.

Serv.io clients get cheap and fast translating and proofreading. More services are coming. Screenshot by Rafe Needleman/CNET

Still room for talent
Edelstein says his services won't impact people who need or provide difficult or high-end services like analysis or creative writing; Porsche still sells $100,000 911s while Ford cranks out Fiestas. Edelstein also says a service like his increases the sizes of markets by making editorial services more affordable. In the multibillion-dollar global translation market, the impact should be good for workers, he maintains, even if it may cause a big upset at the 6,000 or so U.S. translation businesses that currently hire them.

Economically, Serv.io's translation service makes intuitive sense, as does its other current offering, proofreading. The company will also soon turn on its editorial assembly lines for other services, some of which seem a bit less appropriate for the model: Press release writing, resume creation, online research, Web site content creation, and "blog building." That's getting into the Demand Media space, but Edelstein strongly believes his content will be better, thanks to the work flow system CloudCrowd uses.

As much as it feels like I'm digging my own grave writing this story, there is a good economic argument for putting some content creation and editing on an assembly line. With a proper oversight system, you can achieve a level of, if not excellence, then at least reliable consistency. And maybe--just maybe--shunting the less-interesting jobs to the assembly line workers will leave craftsmen with more time to work on the hard, interesting, and genuinely difficult tasks; and leave publishers and advertisers with enough money to pay them to do so.

If not, I will try to land an apprenticeship at a Porsche factory.