Automated platforms define autonomy and manage labor in startling new ways that may soon crossover and impact how we all lead and work
For many senior human resources executives I know, Uber’s arrival into the labor ecosystem is a remarkable crystal ball in which they see future working models evolving before their eyes. Uber and its fellow platform companies are developing labor models in which the traditional middle-management command and control layer is replaced by algorithms that closely monitor and control what workers can and can’t do. With AI-based systems more and more common across large corporate workforces, researchers are eager to understand the models and techniques platform companies are developing today, since their migration into “mainstream” work settings may only be a matter of time.
If the preceding claim seems exaggerated, keep in mind that algorithm-driven platforms already manage the schedule of most large-scale factory production, most large-scale warehousing activities, most retail worker shifts, airline crew assignments, cargo ship navigation, and port operations, and endless other work settings. Algorithms are also involved in an increasing number of recruiting, hiring and performance evaluation processes, their gateway into the “white-collar” working world. Since the dawn of the industrial age, there have always been control systems in corporations; however, these new AI-based systems have raised new alarms for many reasons, from their mechanistic nature to the opaqueness of their coding and decision-making logic.
The concerns about algorithm managers are spurring a lot of research into the subject, and a new paper from Wharton’s Lindsey Cameron is representative of this line of inquiry. She conducted a four-year study of workers in the largest sector of the platform economy, ride-hailing services. During this time, she observed and interviewed drivers and worked as a driver herself. Her specific focus is what she calls “consent,” which refers to workers’ continuing acceptance of the way in which the platforms define, manage, censure, and compensate their work. Her paper, she notes, “goes beyond the ‘carrots and sticks’ metaphors used to describe how consent is generated in traditional workplace and, instead, considers how platforms, through algorithmic management, have renewed and repurposed notions of workplace consent to create this impression of freedom within a presumably even tighter iron cage.”
As expected, she focused her research on platform companies “where algorithms manage the entire work process by selecting applicants, creating schedules, assigning tasks, and evaluating performance.” How, she wondered, do platforms convince people to work in their hyper-monitored environment? To find out, she spent three years doing field research on platform drivers, which included driver observations, interviews as well as her own participation as both driver and passenger. Her work covered 23 North American cities and multiple driver platforms.
Across all her studies, she considered a seeming contradiction of these new platform jobs. On the one hand, they claim to offer autonomy and freedom to drivers. On the other hand, they micro-manage and micro-monitor work in a way that was unimaginable even a decade ago. Indeed, it’s this tension that animates a lot of the research in this topic.
From her study, the author defines several dimensions through which platform algorithms and mechanics generate consent and manage workers. I add my own management take-away to each dimension below.
Technique 1: Scaffold work
The author’s first major conclusion is that algorithms “scaffold the ride-hailing system by coordinating drivers’ work cycle: 1) matching them to riders; 2) instructing them how to work (e.g., giving directions); 3) adjusting ride prices dynamically during busy times (surge pricing); 4) offering bonuses (e.g., “Do 50 rides in the next 5 days for an extra $50.”); and 5) evaluating performance via customer ratings.” Algorithms also scaffold the system by using various types of rewards, targeted penalties, and internal/cycle timing to control drivers. “Scaffolding,” by the way, is a term from education theory, and it refers to a model in which an instructor (or peers) provides steps that a learner completes in order to master a complex task. In this case, the platform divides the income-generating activities into these five “scaffolds” that the system and driver must complete in order to work in the system. “These control and coordination mechanisms,” notes Cameron, “create a workplace where the algorithms are the focal feature. For example, in a typical shift a driver may only complete a dozen rides but will have more than a hundred unique interactions with the algorithm.”
Key management take-away 1: divide a job in smaller pieces that are easier to code and control.
Dimension 2: Reward good behavior (quickly and visibly)
The pricing algorithms of the platforms — primarily, price surges and bonuses in high demand settings — reward drivers if they match their labor to demand:
Surges can be predictable (e.g., rush hours) or sporadic based on local events. Texts and in-app notes alerts tell drivers that: “Demand is higher than usual in Center City. Take advantage of higher-than-normal fares!”, “1.2 – 1.8x boost – 4.30PM-7PM in downtown DC!” and “Adele is playing at [venue] tonight! The streets will be full of people!!” Heat maps that indicate surge areas pop-up when drivers first sign on and after every ride. Checking heat maps becomes routine, so that one “turn[s] on [the] app and then you see that very orange, bright color” and rush to “try to go where the heat maps are surging” for higher pay.
Weekly bonuses, furthermore, “offer extra money for completing an algorithmically-determined ride quota.” Just how algorithms determine bonuses are both secret and, at times, seemingly arbitrary. For example, notes Cameron, “a driver who meets the quota one week may receive an easier or harder quota the following week.” Bonuses, concludes the author, “encourage organizational commitment by offering incentives for staying on-line for longer periods.”
Cameron does not highlight the fact that these reward systems are both rapid and visible, which should only increase their effectiveness. Indeed, this approach is in sharp contrast to the traditional white-collar annual incentive definition and reward model and much more in line with traditional sales and business development “quota/sales reward” systems.
Key management take-away 2: reward desired behavior as quickly as possible and make sure workers see their rewards in real-time whenever possible.
Dimension 3: Sanction bad behavior
Just as they reward, the matching and evaluation algorithms punish bad behavior. When the algorithm completes a match, notes Cameron, “the driver’s phone buzzes and displays the distance to pick-up spot, rider details (including rating), and surge amount (if any).” Drivers have “fifteen seconds to accept; if they do not accept a certain percentage of rides, they are sanctioned.” Repercussions for declining rides include “warnings, temporary blocks, and permanent deactivation; consequently, most drivers reported accepting all rides.”
Sanctions, of course, “are also driven by customer ratings that serve as a proxy for performance evaluations; high ratings (> 4.6/5) are required to continue driving.” In short, riders can “fire drivers” and while drivers also rate riders, “these ratings do not have the same consequences.”
Key management take-away 3: punish unwanted behavior as quickly as possible and base it on real customer decisions whenever possible.
Dimension 4: Coordinate timing
Once a driver accepts a ride, the drivers typically follow a very structured process:
“After you accept a request, tap ‘Navigate.’ The app guide[s] you to the rider. The rider will see your car approaching on their app and your ETA. When you’re close, we’ll send them a text.”
Once the driver arrives, notes the author, “a timer appears dictating how long drivers must wait (sixty seconds to ten minutes based on ride type) before marking the rider as a no-show and receiving a new match.” Route instructions, furthermore, “are critical in shared rides, wherein multiple riders with different destinations share the same vehicle.” Drivers know they must follow the app-defined route, which is algorithmically defined based on rider destination.
Key management take-away 4: create timing cycles, and cycles within cycles, that convert a day’s work into a series of small tasks that can be micro-benchmarked whenever possible.
Dimension 5: Nudge behavior
Algorithmically informed suggestions also nudge drivers, notes the study. Inactive drivers will get texts such as, “You haven’t driven in three days. Go out there and make some money!” Other notifications “encourage longer hours, such as pop-ups that appear only when logging off: “You’ve only driven 11 hours today!” or “Only $18 to go until you meet yesterday’s pay-out.” Interestingly, drivers acknowledge the message before they can leave work. This type of nudging is only the start:
Telemetrics monitor speed, acceleration, and deceleration, offering encouragements such as: “Good job keeping your breaking smooth!” In sum – through rewarding, sanctioning, and timing work activities the five algorithm types create the structural environment or algorithm workplace within which drivers navigate.
Key management take-away 5: avoid macro-effects and use, instead, small adjustments and incentives to keep workers on the platform.
Taking a step back, Cameron sees a major difference between how consent is created in this new platform and how it’s created in traditional work settings. Rather than consent being a “grand bargain” [my term] between company and worker that is, if all goes well, reconsidered at times of promotion or wage discussions, in platform companies consent is a continuously generated outcome of many daily choices. As Cameron notes:
A common theme…was drivers’ belief that their choices were aligned with their self-interests in that they were engaging with the algorithms in order to do the work ‘better’ – i.e., work faster while earning more and maintaining a good reputation. As a result of these frequent choices, in which workers repeatedly opted into the work process, consent was generated.
All of this brings us to California Proposition 22, a new law that passed last fall in California. Prop 22 classifies platform-based transportation and delivery drivers as independent contractors which means they do not need to be treated as normal workers. The debate on Prop 22 was closely watched by companies outside the platform world for many reasons, and the moment the law went into effect, notes Bloomberg, some of those companies began to act:
In December, Albertsons Cos., the supermarket chain, started informing delivery drivers they’d be replaced by contractors. In California hundreds of Albertsons employees are being swapped for DoorDash Inc. workers, according to the United Food & Commercial Workers union. Albertsons declined to comment on the layoff figures but says that the move is happening in multiple states to “help us create a more efficient operation” and that affected workers are being offered other jobs there. (Some workers dispute that last part.) Startups such as Jyve Corp., which sends contractors to grocery stores to stock shelves in lieu of employees, are seeking similar exemptions.
For many people, these early reactions are a harbinger of things to come. For example, Shawn Carolan, a partner Menlo Ventures (an early Uber investor), thinks that Prop 22’s vision of work could spread “from agriculture to zookeeping,” including to “nursing, executive assistance, tutoring, programming, restaurant work and design.”
Helping to push this model’s migration from platform companies to traditional workplaces is the fact that the people who use these new platforms (both customers and workers) generally like them. As the same Bloomberg article notes:
In the long history of American companies seeking to cleave workers from workplace laws, Uber is an outlier. Despite almost a decade of lousy press and scandal after scandal, the service remains broadly popular. Its simple interface and large network—facilitated by billions of dollars in venture funding and shielded from legal reckoning by forced arbitration—have made it a cherished convenience, a cultural touchstone, and the most ubiquitous new verb since “Googled.” By the time most regulators began to seriously worry about the effects of the Uber model, the company had the public support and the money to defy them.
Bloomberg’s opinion is backed by the research since more than one study has shown that platform workers prefer this version of a job to its non-platform equivalent. As one multi-study analysis of drivers noted:
In Study 1, we find that UberX drivers have higher control perceptions than taxi drivers. In Study 2, we find that limousine drivers who use Uber report greater control, and greater enjoyment, when prompted to answer questions about driving for Uber as opposed to a limousine company.
This outcome exists despite other research showing that drivers earn more in the traditional not-platform setting.
What explains this outcome then? Why is it that workers whose every action is defined and monitored by an algorithm consistently feel as if they have greater levels of control and autonomy in the workplace? Perhaps this one sentence from the second paper cited gives us a clue:
We find that gig work increases perceptions of control.
There is a big difference between control and “perceptions of control,” of course. Perhaps the most innovative feature of this new generation of worker control platforms is that, by offering many decision points throughout a day, workers perceive themselves to be in greater control of their working lives than the traditional employee who accepted a job and then felt bound to do whatever a manager said, sometimes for decades at a time. This outcome is frustrating to labor activists, union organizers (and many researchers, frankly) who expect to find great dissatisfaction among most platform workers but don’t. Of course, many platform workers hate the work, but the most common reaction to the negative aspects of a given platform is not to leave it but to game it. Indeed, as the New York Times noted in a revealing 2019 article:
An app called Surge provides Uber drivers the ability to monitor their current location or fixed locations for surges. They can receive notifications when surges start, change and end. With this information, they can decide when to go online and accept rides. They can also put pins in a map to be notified of surges in those areas. Receiving a surge notification, they sometimes turn off all ride-share apps until they arrive in the area, so they are not called to a non-surge trip on the way.
SherpaShare uses location tracking to chart the mileage driven, and at the end of each trip, the driver indicates its purpose as work or leisure. Drivers can deduct 58 cents per mile driven from their taxes as a business expense, so it’s important to keep track, and the app can create I.R.S.-compliant reports for tax purposes. TripLog and Everlance are other similar apps that some drivers have come to rely on.
In short, platform workers are developing their own algorithms to counter what they see as the negative aspects of the platform on which they earn a living. Work in these settings is a technological arms race between the bosses’ software and the workers’ equivalent.
“Control and autonomy,” notes Cameron early in her paper, “need not be antithetical meaning that the integration of algorithms into the labor process does not necessarily mean a delimitation of choice.” The early evidence suggests she’s correct, and that working for and with an algorithm will generate new, more high-tech labor-management chess matches. In the future, algorithms will define the day’s tasks at a company, offer that work out to a collection of contractors and then continuously scaffold, reward, sanction, and nudge workers until the day’s tasks are complete. All the while workers, armed with a variety of their own apps, will shape their choices, actions, and customer interactions to maximize daily wages, believing they are free to leave at any moment of any day. In this world, work will be like a modern-day stock market, a trading platform where people exchange labor for income, each side armed with the most potent algorithms it can create.
Lindsey Cameron, “(Relative) Freedom in Algorithms: How Digital Platforms Repurpose Workplace Consent”. In Proceedings of the Eighty-first Annual Meeting of the Academy of Management. Online ISSN: 2151-6561, edited by Sonia Taneja, (2021)