The Waiting Game: How Gig Platforms Turn Your Uncertainty Into Underpay

Based on the research ofStoica, Mendler-Duenner, and Hardt, "Stochastic wage suppression on gig platforms and how to organize against it," arXiv preprint, 2026

The Waiting Game

Ana-Andreea Stoica, Celestine Mendler-Dünner, and Moritz Hardt build a formal model of a very ordinary-looking market: a platform wants M tasks done, so it posts a price and lets workers arrive one after another, each accepting if the price beats her own privately estimated cost. Nothing about that setup sounds adversarial. It is how ride-hailing, food delivery, and crowd-work like data annotation already operate.

The result is not ordinary. Under natural assumptions about how workers estimate their own costs, the paper shows there is a simple pricing strategy that lets the platform cover all M tasks while paying only an O(log(M)/M) fraction of what the labor is actually worth. In plain terms: as the number of completed tasks grows, the platform's total wage bill grows only logarithmically, not in proportion to the work being done. A platform running ten times the volume does not have to pay anywhere near ten times as much. The cost to the platform of this trick is patience, formally a wait time of O(M), it has to be willing to sit on a low price until someone accepts it.

That "someone" is the whole mechanism. In a large enough pool, there is reliably a worker whose estimate of her own cost sits below the posted price, not because her actual cost is low, but because she has not yet learned to price in the unpaid time platforms are notorious for omitting: searching for the next task, driving to a pickup, waiting between rides. The researchers, based at the Max Planck Institute for Intelligent Systems, say the project began from "the large empirical literature showing that many digital workers earn very little, once unpaid time and other hidden costs are taken into account," and that they wanted "to understand one structural mechanism that can sustain these outcomes." As Stoica put it in describing the finding, "one reason why platforms pay low wages is algorithmic." This is theory, not a measurement of any specific company's books, the paper reports a formal bound on a stylized posted-price model, not empirical wage data pulled from a platform. But the mechanism it isolates, a price that never has to rise because someone new is always arriving underinformed, describes something real-world reporting has been circling for years.

What the Algorithm Is Betting On

The premise that workers routinely misjudge what a gig actually pays isn't new to labor scholars. Law professor Veena Dubal's "On Algorithmic Wage Discrimination," published in the Columbia Law Review, is built on a multiyear ethnographic study of ride-hail drivers and documents how granular, opaque, personalized pay calculations leave workers unable to predict their own earnings, drivers in her interviews describe the pay structure itself in terms of "gambling and trickery." Dubal's argument is about a different axis, price discrimination imported from consumers into labor, but it rests on the same load-bearing fact the new model formalizes: workers do not know, in advance, what a task will actually cost them in time.

That gap matters most for newcomers. A driver two weeks into the job has not yet built an internal model of how much dead time, positioning, and app-navigation friction eats into an advertised rate. She is exactly the worker a posted-price strategy like the one in the paper is built to find. The platform does not need to fool everyone. It needs the pool to be large enough, and to keep replenishing, so that someone underpriced always shows up before the platform would ever have to raise the number.

Why a Bigger Union Doesn't Fix This

The intuitive response to a wage floor problem is to organize broadly: get as many workers as possible to refuse the low price together. The paper's second result is a direct challenge to that instinct. A "horizontal" coalition, a randomly assembled group of workers who pledge not to accept below some price, turns out to be largely ineffective. The reason is mechanical rather than moral: the platform can simply keep waiting for a worker outside the coalition to accept, and in a large market there is almost always one. To meaningfully bind the platform, a horizontal pledge would need something close to full participation, which is exactly the coordination problem that makes labor organizing hard in the first place.

You can see the shape of this failure in the largest coordinated action rideshare and delivery drivers have staged. On February 14, 2024, the Justice For App Workers coalition, representing roughly 130,000 drivers and delivery workers, organized strikes and airport pickets across ten U.S. cities, the first such call since Uber and Lyft went public in 2019. "A year into algorithmic pricing, drivers have seen incredible decrease of our pay," Nicole Moore, president of Rideshare Drivers United, told Reuters at the time. Uber's response was that "driver earnings remain strong," citing roughly $33 per utilized hour in the fourth quarter of 2023 and describing most drivers as satisfied. Independent data told a rougher story: Gridwise, which tracks gig-mobility earnings, found Uber drivers' average monthly gross earnings fell 17.1 percent over 2023 even as Lyft drivers' rose 2.5 percent. Whatever the true trend, a single day of broad, cross-market pressure did not change the underlying pay formula, which is precisely what the model predicts a horizontal action, however large, would fail to do, because it never removes the specific supply of underpriced labor the platform is routing around.

The Right Few Workers

The paper's more optimistic result is about targeting, not size. A small "vertical" coalition, workers drawn specifically from the low-cost segment the platform's pricing strategy depends on, who commit to a shared price floor, forces the platform's total spending to scale linearly in M instead of logarithmically. That is the difference between a wage bill that barely moves as the business grows and one that tracks the business honestly. The mechanism is the mirror image of the exploit: if the specific workers a platform is counting on to underprice refuse to, the platform runs out of the "always someone" its strategy needs, and it has to pay a real price to keep the tasks covered.

Turkopticon is a rough, decades-old sketch of what closing that gap can look like in practice. Founded by researcher Lilly Irani and worker-organizer Six Silberman as a way for Amazon Mechanical Turk workers to share information about which requesters paid fairly and which didn't, it now describes itself as a worker-led nonprofit that fights issues like mass account suspensions and helps newcomers avoid underpriced or exploitative tasks before they accept them. It isn't a picket line. It's an information intervention aimed at exactly the population the theoretical model says matters most: workers who haven't yet learned what their labor is worth on this particular platform.

The platform does not need every worker to underprice its labor. It only needs to never run out of one who will.

What a Floor Is Actually For

It is worth noting what an externally imposed floor looks like when it does work as intended. New York City's Taxi and Limousine Commission adopted a minimum pay standard in December 2018, effective February 2019, the first policy of its kind in the country to set a minimum trip payment for drivers classified as independent contractors. Rather than setting the passenger fare, the rule fixes what the largest for-hire platforms must pay drivers per trip, built from three components, time, distance, and driver utilization, so the payment tracks the actual work performed rather than whatever a platform's pricing algorithm would otherwise settle on. It is, in effect, a regulator doing by mandate what the paper's targeted coalition does by collective commitment: refusing to let the price be set by whichever side of the market is most willing to accept less.

The throughline across the theory and both real-world cases is the same. A wage floor is not a number; it's a removal of the platform's ability to wait for a cheaper yes. Whether that removal comes from a regulator's formula or from a small group of the exact workers a pricing algorithm is counting on, the effect is structurally identical, and whether it comes from ten thousand people picketing for a day is, on this model's own math, closer to noise. The uncomfortable lesson for organizers is that solidarity, to bite, has to be aimed. The lesson for anyone designing these markets is blunter still: a posted price that never has to rise isn't evidence that the work is worth that little. It's evidence that somebody hasn't found out yet what it's worth, and the platform is in no hurry to tell her.