Working with Algorithms: Labor, Technology, and the Rise of a Billion-Dollar Startup

Recent years have brought a resurgence of interest in the implications of technological innovation for the future of work. If automation in the factory displaced millions of blue-collar workers during the 20^th century, software algorithms seem poised supplant white-collar knowledge workers in the 21^st. Many observers argue that software algorithms are already beginning to render human labor obsolete (Brynjolfsson and McAfee 2014).

Prior studies of work and technological change point to two reasons for researchers to be skeptical of such alarmist accounts. First, scholars who closely examine how technological innovations are incorporated into production processes often discover that machinery presented as eliminating the need for human intervention often fails to do so in practice (Samuel 1977; Nichols and Beynon 1977; Noble 1999). The high cost, limited capacities, fragility, or rigidity of automated equipment can lead employers to search for human helpers. Second, the shape of technological innovation and the effects of technological change on work are not predetermined by scientific progress or rational considerations of efficiency alone, but are instead a product of social relations (MacKenzie and Wajcman 1999). The effects of new machinery on labor are contingent upon the specific characteristics of the technology introduced, power relations between and among workers and managers, the structure of organizations and the organization of production processes, and the processes through which innovations are implemented.

These studies suggest that those wishing to understand how the digital revolution is changing work should not limit their view to how technology enhances the value of highly skilled labor while eliminating routinized jobs. We should also look for the intertwining of computer software and humans in socio-technical systems. Researchers currently know remarkably little about how organizations create and implement the innovative software algorithms that seem poised to reshape the American economy. Where do we find human labor emerging in and around the software algorithms that many predict will come to define our era?

From 2012-2013, I spent 19 months conducting participant-observation research at a San Francisco-based startup firm called AllDone. The company ran an online market connecting buyers and sellers of local services across the U.S. ranging from house cleaning to wedding photography to tutoring. By late 2015, the firm was valued at over one billion dollars. I found that, at AllDone, software algorithms and routinized human labor were frequently co-constitutive rather than mutually exclusive. This paper shows how the software algorithms produced by the firm’s small San Francisco workforce relied upon the continual intervention of a large, online, low-wage workforce distributed throughout the Philippines and Las Vegas.

AllDone’s remote workforce solved problems arising from two shortcomings of software algorithms. First, AllDone tasked workers in the Philippines with addressing software’s practical and technical limitations. Sometimes solving problems with software algorithms was technically but not practically feasible due to the cost and time-intensiveness of their research and development and the organization’s resource constraints. Additionally, certain tasks required cultural fluencies beyond the capacities of software algorithms—though algorithmic infrastructure could be developed to enhance human performance of such tasks. Workers in the Philippines performed highly routinized knowledge work to adjust or complete the work of software algorithms

Second, AllDone deployed workers in the Las Vegas area to address software’s social limitations. Users do not passively accept technologists’ structuring of their activities; instead, they often develop interpretations of and adaptive responses to technology that can undermine designers’ intentions (Zuboff 1988; Yang and Newman 2013). Some AllDone users were confused or upset by their experiences with the service and with the opacity of its dynamic rules and user interface. In these cases, human workers performed interactive emotional labor, speaking with dissatisfied customers over the telephone to manage how users felt about AllDone.

I also demonstrate how the composition of labor supporting software shifted over time as the organization’s priorities and resources changed. These strategic changes were driven by the cycle of venture capital funding. During the first six months of my research, top managers hoped to take advantage of AllDone’s first round of venture capital funding by prioritizing the growth of the company’s user base. AllDone accordingly expanded its team of online workers in the Philippines whose activities on a digital assembly line drew more users to the website. During the second phase of my research, the company redirected its focus to generating revenue. When top managers decided to subject the company’s most valuable users to a sudden, dramatic increase in fees to achieve that goal, they discovered that there was no solely technological solution to mitigating users’ anger. They chose to supplement software algorithms with online workers in Las Vegas whose interactive emotional labor helped to prevent users from giving up on the product.

During the final six months of my research, the company obtained a second round of venture capital funding and began to rationalize operations as it prepared for another cycle of growth, this time expecting an exponential expansion of the user base. Even as the company’s engineering corps grew over the following months and years, its remote workforces continued to swell to support the software. The team in the Philippines has grown fourfold since my departure from the company, though its scope of activity has narrowed—because some tasks have been automated, most remaining workers are assigned to tasks that cannot be fully programmed. And the small team in Las Vegas has been replaced by a large and more highly trained workforce operating out of a Salt Lake City office and specializing in managing relationships with users.

If AllDone’s articulation of labor has changed over time to suit the company’s shifting strategic imperatives, one thing has remained constant: the intimate intertwining of computer code and human labor. This study suggests that researchers must examine the organizational contexts in which algorithms are developed and deployed to understand the dynamics of work and technology in the digital age.