Uber and “The Taking Economy”: The dynamics of two-sided markets and algorithmic exploitation

Uber’s had a rough few weeks.

From #deleteUber, to Travis Kalanick’s caught-on-camera argument with a driver, to Susan Fowler’s account of unchecked sexual harassment at the company, patience for the aggressive Silicon Valley unicorn was already wearing thin.

Then the New York Times dropped another bombshell: for several years Uber has been using a computer program to evade local regulators, flagging law enforcement and investigators, and showing them misleading representations of the app’s dynamic home screen while instructing drivers to cancel any rides those users requested. But as Data & Society’s Alex Rosenblat — who has long researched Uber and its management practices — points out, the program is “part of much longer pattern of manipulating or fudging variables they control.”

Rosenblat and University of Washington law professor Ryan Calo recently released a well-timed working paper that examines the power imbalance intrinsically created when sharing economy platforms mediate transactions between two discrete groups of users (often buyers and sellers or service providers).

If you take nothing else away about Rosenblat and Calo’s paper, take this:

The sharing economy seems poised to do a great deal of taking — extracting more and more value from participants while continuing to enjoy the veneer of a disruptive, socially-minded enterprise.

To fully appreciate the paper’s insights and recommendations, however, it’s worth taking a short detour into the world of economics and business strategy.

Uber is operating a two-sided market, where a strategy to meddle makes economic sense.

The economic structure of online platforms are well-understood by business strategists: platforms must continue to attract users on both sides of the market — both buyers and sellers — since they extract revenue from the economic activity on the platform. Because marginal costs for such platforms are minimal, every new user stands to significantly increase revenue while requiring almost no additional resources (whereas a widget company would have to pay to produce new widgets if they wanted to sell at greater volumes).

This two-sided market structure means that user acquisition and retention are critical; each side of the market needs the other for the market to succeed. But user acquisition costs money. Advertising is pricy, and the $10, $20, $500 signup incentives for riders and drivers add up fast, too. Platforms also need to discourage participants from leaving — either by leveraging network effects such that its nearly impossible for alternate services to seriously compete since everyone and their friends are already in one place (think Facebook), or by raising switching costs to make it more difficult to defect to competitor platforms.

Because two-sided markets need to keep an optimal balance of users in each group or risk losing users on the other side, pricing is a key consideration for platforms operating these markets. If Uber has too many drivers, they’ll be driving around aimlessly without earning anything and will quickly abandon the service. If there are too few drivers available, riders might defect to a competitor.

Uber’s surge pricing represents the company’s strategy to manage real-time supply and demand by enticing more drivers out onto the road during busy periods. But behind the scenes, the company also needs to manage long term driver supply.

To avoid labor regulation, the company vehemently resists establishing contractual employment relationships with drivers; although they are subject to Uber’s terms of service, drivers are free to switch platforms at will. But since the barrier to entry for drivers is higher that for riders — drivers need a car, must wait to pass a rudimentary background check, and need to commit far more time than riders — platforms must do more to attract and retain them.

From a business strategy perspective, then, it’s in the company’s interests to make it as hard or undesirable as possible for drivers to switch. This economic imperative becomes highly relevant when imagining all the ways Uber might be manipulating its position as a two-sided market platform.

Talking about taking.

The economic realities of two-sided markets set the stage for Rosenblat and Calo’s discussion. When digital platforms facilitate transactions between different sides of a marketplace, they are in a position to monitor, analyze, and nudge user behavior to deliver value to their users — and in doing so, to accumulate value for themselves. But as the authors point out, the inherent information asymmetry between companies and users takes on a new dimension when “users” rely on these platforms for income — particularly lower-income users for whom that income is meaningful, if not critical, for their financial well-being.

Uber’s dispatch algorithm is managing workers, and since users have little vision into how the algorithms work, it’s all too easy to make minor tweaks that, while difficult for users to detect, directly affect both those users and larger systems of pricing, labor, and regulation. Uber’s Greyball program, the authors note, lends weight to the suspicion that Uber and similar platforms are manipulating their users in other, more pernicious ways.

Such behavior fits into a broader pattern of deception and misdirection: Uber has admitted that many of the cars users see on their screen are “phantoms,” and has also openly revealed that it knows, for example, that customers with low battery are more willing to pay higher prices.

Rosenblat and Calo describe a number of ways sharing economy companies might be taking more than we think. Beyond familiar issues like regulatory arbitrage, (French regulators and taxi drivers have favored the term “economic terrorism” to describe Uber’s approach to regulation), enabling discrimination, and questionable privacy practices, the paper explores how these companies are manipulating the market at a systemic level, combining their capacity to structure transactions with a surpassing degree of insight into customer behavior. Phantom cars and predatory surge pricing set the stage for other concerning practices like dynamic price discrimination: Researchers have found that similarly situated users are presented different prices, anomalies the company blames on system bugs.

Returning to our MBA lesson earlier, opportunistic price discrimination is a critical strategy in platform economics because marginal costs are minimal; companies benefit by attracting as many customers as possible, charging each just what she is willing to pay. Who is to say a ridesharing company wouldn’t use data collected about passengers to introduce invisible differential pricing, charging more to riders who won’t bat an eyelid at a slightly elevated price? The company only stands to benefit — as long as they don’t overestimate and push riders into the arms of a competitor.

The authors also describe how a company like Uber could take from the “entrepreneurial consumer” side of the market — the drivers. The company unilaterally and rapidly changes policies in ways that confuse drivers, and increasingly hides important details (like destination, passenger information, and precise surge rates) from them. Drivers are forced to accept trips without complete information, and if they cancel too frequently, risk temporary or permanent deactivation from the platform.

Do Uber’s targeted incentives exploit price inelasticity of drivers?

More concerning is Rosenblat and Calo’s discussion about how Uber might manipulate hourly guaranteed pay for drivers. While the platform, like taxis, sets regular rates based on time and distance, the company also offers drivers temporary incentives: an hourly guarantee, so long as the driver meets certain conditions like completing a certain number of rides and accepting a certain proportion of ride requests. But not all drivers are offered these special rates.

How does Uber decide who to invite? Rosenblat has described in the past how wage theft might occur in app-mediated work— for example, where the platform claims a driver failed to accept the minimum 90% of rides required to unlock an hourly guarantee, even if the driver attempted to do so.

Rosenblat has pointed to two competing theories about how these hourly incentives are related to Uber’s algorithmic dispatching:

  1. Uber could be favoring “power user” drivers as a way to discourage them from defecting, and
  2. Uber could be offering hourly guarantees but flooding the market with drivers so that drivers are less likely to be able to complete the minimum requirements.

What might these theories look like in practice, and can they be updated especially in light of the Greyball scandal?

First, Uber could indeed be engaging in dynamic price discrimination — only this time, targeting drivers who are willing to work for less money. If we know that not all drivers are offered incentives, why shouldn’t we assume that rather than proactively offering incentive pay to “power users,” Uber uses hourly guarantees or other types of incentive pay to target drivers they actually predict will soon defect?

Employers of all stripes have already begun using predictive analytics to flag employees with the highest risk of attrition and offer them raises or more responsibility to entice them to stay. Uber lays out hundreds of dollars or more in sign-up bonuses for new drivers and users who refer them; throwing drivers a few shifts with higher pay costs significantly less than recruiting a new driver to replace one who switches to another platform. (Many drivers use multiple platforms at once, but the same logic would apply for drivers who shift their primary loyalty to a different app.)

At the same time, Uber is likely able to see which drivers require frequent financial incentives to remain active and which do not. And it’s very likely that those drivers who rely most on these platforms as their primary source of income are also more price-inflexible — they are willing to work for less, because they need the income more and so are less likely to abandon the platform even if they are not offered incentives (or receive them less frequently). Such a calculation would direct opportunities for higher pay toward part-time drivers who need it least — and who are therefore at greatest risk of leaving. If Uber is employing this sort of price discrimination or exploiting price inelasticity on the driver side of the market, it would be highly discriminatory toward low-income drivers. Taking, indeed.

Second, as Rosenblat has described, Uber’s dispatching algorithm could be actively preventing drivers from hitting the minimum requirements for incentive pay. She has hypothesized that Uber could be flooding the market with drivers as a way to make it more difficult for them to qualify for incentives, and describes how some drivers have reported receiving “phantom pings” for rides that they try to accept, but that disappear too quickly while still counting against the drivers’ acceptance rates.

But what if, in addition to using phantom pings to depress drivers’ ride acceptance rates, the dispatching algorithm actively routed ride requests away from drivers about to hit their minimum number of rides, regardless of how many drivers are on the road and without even giving drivers the illusion of potential passengers?

As we learned before, not all drivers are offered guarantees, and nobody knows precisely how Uber prioritizes ride allocation among its drivers; the practices uncovered in the reporting of the Greyball program have only fueled skepticism that things are more complicated than they appear. By offering rides first to drivers earning the lowest rates before sending rides to drivers hoping to unlock incentives, even if some drivers manage to complete the requisite number of rides and achieve temporarily higher earnings, Uber maximizes its revenue over time. At the same time, the company would be maintaining the illusion of higher earning opportunity for drivers, and so convincing them to stick with Uber for a while longer rather than switching over to another platform.

Unfair and deceptive? Consumer protection could be a more effective regulatory approach.

Calo and Rosenblat posit that consumer protection law is well suited (albeit underdeveloped) to address harms caused by the sort of power and information asymmetry that Uber embodies. Indeed, the FTC recently imposed a $20 million settlement on the company to resolve a complaint that it falsely advertised how much drivers could earn on a yearly and hourly levels. But we ought to recall that current acting FTC commissioner Maureen Ohlhausen dissented to that settlement, arguing that the “remedy is inappropriate for a non-fraudulent enterprise that significantly benefits consumers, including drivers” — emphasis mine. Evidence is stacking up that the platform might be more culpable than Ohlhausen would suggest.

The authors identify strategies to improve consumer protection enforcement, calling for more active investigation by regulators like the FTC into business practices, as well as incentivizing researchers to monitor for cases of unfair, deceptive, and harmful practices. More research is certainly critical: Automated price discrimination at the sharing economy scale could be nearly impossible to detect, likely requiring either an enormous experiment, or collecting ride-by-ride details from a wide swath of drivers and controlling for a significant number of confounding variables like trip distance, time, driver characteristics, rider characteristics, and surge status.

While consumer protection law might indeed be the right approach to regulating gig economy platforms, labor advocates are justifiably concerned about systemic manipulation of labor markets and worker wages, which has become ever more of a risk in algorithm-driven enterprises. Similar to traditional issues like lack of salary transparency making it difficult for workers to organize, drivers may have no idea how much they make compared to other users. And on top of that, actual rates of wage theft and “computer error” in the platforms’ favor are hard to detect when each worker’s rates may be different.

The regulatory strategies proposed by Calo and Rosenblat are thoughtful, comprehensive, and worth a read. But non-regulatory incentives can also play a role in motivating gig economy companies to treat their users better. Following public pressure over evidence of discrimination on Airbnb, CEO Brian Chesky embraced a more proactive approach and the company seems to have prioritized efforts to design products with positive social outcomes in mind, working with civil rights groups to address concerns and identify effective interventions that benefit both the company and the community.

After Uber’s cascade of controversies, 40-year-old CEO Travis Kalanick recognized that he “must fundamentally change as a leader and grow up.” A nice sentiment, but we’ll see what changes actually come to pass. One heartening step would be to see Uber welcome cooperation with labor advocates to begin rebuilding the platform, as a healthy marketplace that also treats drivers with dignity and respect.

Especially as the company scrambles to earn back what little legitimacy it had, reconsidering its approach to the people who rely on the platform for their livelihood would be a good place to start.