BOLD Idea No. 1: How to have a million people work for you on demand
Imagine the kind of workforce that could build the Empire State Building in the time Americans spend in a single weekend watching commercials, and you have a glimpse of what Lukas Biewald and Chris Van Pelt have created in CrowdFlower. This startup outsources microwork to some 20 million global workers, who spend bits of free time in their day on useful yet potentially monotonous work. Forget recruiting, training and hiring -- this is how to get a big project off the ground.
"Something that's fun to do for five minutes is horrible to do for five hours," Lukas told me. That's where the power of crowdsourcing comes in: A million workers who each spend 10 minutes on a task can complete it quickly and effectively, no matter how repetitive or tedious it might be. Although CrowdFlower's clients include Fortune 500 companies like eBay, Microsoft and Twitter, smaller companies that need big data and economies of scale are also crowdsourcing microwork to build momentum on a budget.
What's in it for them?
"You see all motivations," says Lukas. "Some workers want to learn English better. That's not an uncommon motivation for someone outside the U.S. Some others really love the flexible form of employment, such as stay-at-home moms. You also see a lot of college students, and even a security guard who told me he was being paid twice because at one job -- the security one -- he wasn't doing anything and could do something for us."
CrowdFlower incentivizes its workers with an algorithm that predicts and ensures accuracy. Users each have accuracy scores based on their prior work that they must keep over a certain level to be eligible for project work. This algorithm is essential for CrowdFlower's success, Lukas explained. "When you post 1,000 jobs, and everyone does 10 minutes of work, and 1,000 people show up to do the work, checking that the results are high-quality is a tricky problem."
The power of the crowd
CrowdFlower client projects have formed some compelling case studies for the power of crowdsourcing microwork. Consider the following:
The Harvard Tuberculosis Project: Workers analyzed slides to identify tuberculosis cells that had survived drug treatment by analyzing slides. CrowdFlower's army was able to analyze the slides much faster than on-site grad students could without compromising accuracy. The project proved, as Lukas told me, "the democratic potential of crowd-sourcing when applied to science. We can break down the barriers between scientific research and those who bear the economic and health problems of TB."
Haiti earthquake relief: CrowdFlower helped translate thousands of text messages after the 2010 Haiti earthquake, helping aid workers connect with those who needed help most. "They didn't have translators on staff. They used our crowd to translate the messages into English from Creole," explained Lukas. "Companies can never staff up for the volume they need, or for every disaster."
Policing porn: Skout, a social networking site that uses GPS to help users find friends close to them, uses CrowdFlower to scan photos and analyze text to remove inappropriate images and messages. Initially, Skout had CrowdFlower analyze under 10 uploads a day, but now, things have changed. "They're actually a large customer now, using us to scan millions of images every day for their content."
Lukas says that CrowdFlower is most appropriate for tasks that are objective and measurable: categorizing data, reviewing search engine results for relevance, or scraping data from receipts and websites.
"I did not start CrowdFlower with the idea of doing biological cell labeling," Lukas admits. "But what I wanted to do was develop a tool for people to use. There's a lot of work available, and people with extra bits of time could be doing something useful."
Readers, I'd love to hear your thoughts. What did you find most useful about this idea? What did you disagree with? What would you like to know more about with regards to crowdsourcing and/or microwork? Do you have any stories, anecdotes or recommendations on crowdsourcing? -PHD