Fired by an Algorithm: How Platform Giants Are Terminating Workers While Claiming No One Pulled the Trigger

Marcus had been delivering packages for Amazon Flex for eleven months without incident. Then, one Tuesday morning, he opened the app and found his account deactivated. No phone call. No email explaining what had gone wrong. No supervisor to speak with. Just a brief, automated message informing him that he was no longer eligible to make deliveries — and a link to a dispute process that, over the following six weeks, would yield nothing but automated responses.

"I asked them who made this decision," Marcus recounted in an interview conducted through a gig worker advocacy group. "They kept telling me the system flagged my account. I asked what system. They couldn't tell me. I asked what I did wrong. They couldn't tell me that either. I just stopped existing to them."

Marcus's experience is not an anomaly. It is, according to labor researchers, worker advocates, and a growing body of litigation, a defining feature of how America's largest platform companies manage their workforces — or, more precisely, how they avoid managing them at all.

The Architecture of Avoidance

The gig economy's labor model was constructed, from its earliest days, around a central legal fiction: that the people performing the work are not employees but independent contractors. This classification carries enormous consequences. Independent contractors are not entitled to minimum wage protections, overtime pay, unemployment insurance, workers' compensation, or the right to organize under the National Labor Relations Act. By reclassifying millions of workers as contractors, companies like Uber, Lyft, DoorDash, and Amazon's third-party delivery network have shed legal obligations that traditional employers have carried for decades.

Algorithmic management is, in many respects, the operational infrastructure that makes this fiction sustainable. If a human resources department were systematically reviewing worker performance, issuing warnings, and making termination decisions, it would be difficult to argue that those workers were not, in some meaningful sense, employees. The introduction of automated systems allows platforms to exercise the same degree of control — monitoring performance metrics in real time, adjusting pay rates, assigning and withdrawing work opportunities, and ultimately deactivating accounts — while maintaining that no employment relationship exists because no human being is directing the work.

The practical effect is a workforce subject to near-total surveillance and instant dismissal, with none of the procedural protections that employment law has historically provided.

What Algorithmic Deactivation Looks Like in Practice

Each major platform operates its own variant of algorithmic termination, but the underlying mechanics share common features.

Uber and Lyft track driver ratings, trip acceptance rates, cancellation rates, and a range of behavioral signals. When a driver's metrics fall below platform-defined thresholds — thresholds that are not always disclosed to drivers in advance — the account is flagged for deactivation. The process is automated. A driver who has completed thousands of rides may find themselves locked out of the app with no more explanation than a form email.

DoorDash employs a "completion rate" metric that tracks the percentage of accepted orders a dasher successfully delivers. Falling below 80 percent can trigger deactivation. What the algorithm does not reliably distinguish between is a dasher who abandoned deliveries without cause and one who was unable to complete an order because the restaurant had closed, the address was incorrect, or the customer was unreachable. From the system's perspective, an incomplete delivery is an incomplete delivery.

Amazon's Flex platform is particularly opaque. The company's algorithm monitors what it calls "delivery quality" signals — GPS data, delivery confirmation photographs, customer complaints — and uses these inputs to generate a score that determines a driver's continued eligibility. Workers have reported being deactivated for deliveries that customers later confirmed receiving, for GPS discrepancies caused by poor signal in rural areas, and for package thefts that occurred after confirmed delivery.

In each of these cases, the worker's recourse is a dispute process that, in practice, functions as a formality. Appeals are reviewed — if they are reviewed at all — by contractors working from scripts rather than by individuals with authority to exercise judgment.

Workers Speak Out

The human cost of these systems is not abstract. TechToDown gathered accounts from gig workers across multiple platforms through advocacy organizations including the Gig Workers Collective and the National Domestic Workers Alliance.

A former DoorDash driver in Atlanta described losing her account after a string of restaurant delays — circumstances entirely outside her control — caused her completion rate to drop below the threshold. "I had been doing this for two years," she said. "I knew the app, I knew my market. And then one bad week, and I'm gone. No conversation, no context. Just gone."

A former Uber driver in Chicago reported being deactivated following a customer complaint he believes was retaliatory — filed after he declined to take a route he considered unsafe late at night. "There's no way to prove what happened. There's no one to prove it to. The algorithm decided, and that's the end of the story."

For workers who depend on platform income as their primary source of livelihood — a population that research suggests is larger than platforms typically acknowledge — deactivation without warning or meaningful appeal is not an inconvenience. It is a financial emergency.

The Discrimination Problem

Beyond the procedural injustice of algorithmic termination lies a deeper concern: the potential for these systems to perpetuate and amplify discriminatory outcomes.

Dr. Safiya Umoja Noble, a researcher whose work on algorithmic bias has shaped the field, has argued that automated systems trained on historical data inevitably encode the biases present in that data. In the context of gig work, this means that if certain demographic groups have historically received lower customer ratings — due to customer bias rather than service quality — an algorithm that uses ratings as a deactivation trigger will disproportionately terminate workers from those groups.

Research has provided empirical support for this concern. A 2021 study published in the journal Science found that rideshare passengers were significantly more likely to cancel rides when matched with drivers whose names suggested African American identity. If cancellation rates feed into a platform's deactivation algorithm — as they do on both Uber and Lyft — then customer bias becomes, through the algorithm, a mechanism of discriminatory termination.

Platforms have largely declined to release the data that would allow independent researchers to assess whether deactivation rates vary systematically by race, gender, or geography. The opacity that characterizes these systems makes discrimination simultaneously more likely and harder to prove.

States Push Back — Unevenly

In the absence of meaningful federal action, a handful of states have begun to address algorithmic management and gig worker protections through their own legislative processes.

California's Proposition 22, passed in 2020 with heavy financial backing from Uber, Lyft, and DoorDash, preserved the contractor classification for app-based workers while establishing a limited set of earnings guarantees and benefits. Notably, it did not address algorithmic deactivation or establish any meaningful appeal rights. A subsequent court ruling found portions of the measure unconstitutional, and the legal status of its provisions remains contested.

New York City has taken a more direct approach. The city's Local Law 144, which took effect in 2023, requires employers using automated employment decision tools to conduct and publish bias audits. While this law applies primarily to traditional employment contexts, advocates have argued it should be extended to cover platform deactivation systems.

Minnesota enacted legislation in 2023 establishing minimum pay standards for rideshare drivers, and the state has been among the more active in considering broader gig worker protections. Illinois and Washington have introduced bills addressing various aspects of algorithmic management, though none have yet passed comprehensive legislation.

At the federal level, the PRO Act — which would substantially narrow the definition of independent contractor and extend collective bargaining rights to gig workers — has passed the House twice but has not advanced through the Senate. The Department of Labor's 2024 rule tightening the criteria for independent contractor classification represents a meaningful step, but it does not directly address algorithmic termination practices.

Accountability Requires Transparency

The core problem with algorithmic management, as labor attorneys and policy advocates consistently emphasize, is not merely that machines are making decisions. It is that the use of machines has been deliberately structured to eliminate accountability for those decisions.

When no human being is identified as the decision-maker, there is no one to question, no one to appeal to, and no one to hold legally responsible. Platforms have exploited this structure to insulate themselves from the due process obligations that employment law was designed to impose.

Meaningful reform would require, at minimum, three things: mandatory disclosure of the criteria and thresholds used in deactivation algorithms; a genuine human review process for contested deactivations, with defined timelines and binding outcomes; and the extension of basic procedural protections — notice, explanation, and appeal — to all workers whose livelihoods are controlled by platform systems, regardless of their classification.

The technology to build fair, transparent, and accountable management systems exists. What is absent is the regulatory pressure that would compel companies to build them. Until that pressure materializes, millions of American workers will remain subject to the judgment of systems they cannot see, appealing to processes that do not listen, and losing livelihoods to decisions that, officially, no one made.