The Hidden Challenges of Food Delivery and Ride-Hailing Platforms
Food delivery personnel and drivers operating on various platforms often grapple with the frustrating reality of their earnings and job opportunities fluctuating based on automated decisions made by algorithms that function like a black box. These applications analyse a plethora of data, including customer ratings, accepted orders or rides, cancellations, and the duration of service completion. Such variables feed into calculations that ultimately dictate how work is organised and evaluated.
Those employed through these platforms frequently find themselves in the dark regarding the rationale behind specific decisions made about them. This lack of transparency complicates their ability to identify errors, challenge penalties, or comprehend shifts in their work opportunities and income. This issue is one of the critical gaps highlighted in a study examining platform labour regulations across Mexico, Chile, Colombia, and Uruguay, conducted by Derechos Digitales, a Latin American organisation founded in 2005 dedicated to safeguarding human rights in digital environments.
The Data Conundrum in Platform Work
The crux of the problem lies in the data itself. Information such as location, behaviour, performance metrics, ratings, and, in certain instances, biometric data contribute to the ecosystem that platforms rely on to manage work. Derechos Digitales refers to this phenomenon as the “datification of work”, where these activities are transformed into data that subsequently informs automated systems.
“They take all this information, which enters a set of algorithms, and ultimately use these data to determine wages and tasks, as well as to block workers,” explains Laura Mantilla-León, a Public Policy Analyst at Derechos Digitales, in an interview. She further elaborates that the platforms disclose little about how they utilise this data and how it influences their decisions regarding work.
Understanding the Rules and Their Implications
Recent regulatory developments in Latin America have begun to address this opacity. Various laws assessed by Derechos Digitales acknowledge that workers should receive information about algorithmic criteria that may impact task allocation, income, performance evaluations, or penalties. For instance, in Mexico, the Federal Labour Law mandates that clear and straightforward information be provided regarding factors such as ratings, incentives, penalties, and criteria influencing task assignments.
While Mantilla-León regards this recognition as significant, she believes it remains inadequate. A platform may disclose the variables it employs without clarifying how they result in a specific outcome for an individual. “What is documented tends to be somewhat restricted to general descriptions of criteria and variables, falling short of the crucial element of explainability,” she notes.
The Need for Transparency and Explainability
This distinction becomes evident in everyday scenarios. A platform may inform users that it considers location, acceptance rates, and ratings; however, that does not necessarily elucidate why a specific individual was adversely affected following a particular action. Consequently, Mantilla-León differentiates between transparency and explainability. The former allows for the identification of elements involved in these systems, while the latter ensures that workers can grasp the reasoning behind an automated decision.
This differentiation is vital because knowing only the general rules leaves a significant part of the process opaque. If a worker begins receiving fewer orders, loses an incentive, or faces account suspension, they require more than a list of variables to understand what transpired. The Derechos Digitales report highlights that current obligations often remain confined to general descriptions of criteria and variables.
The Workers’ Struggle to Decipher the System
The repercussions of the lack of clarity surrounding algorithmic operations have been documented in various studies. A research project published in 2024 by the National University of General Sarmiento examined the experiences of delivery workers and drivers in the Buenos Aires Metropolitan Area through approximately 70 interviews and a survey of 750 cases. The study focused on platforms operating during the analysed period, including Uber for transportation and PedidosYa, Glovo, Rappi, and Uber Eats for delivery.
The findings revealed that metrics such as customer ratings and acceptance or cancellation rates could significantly influence the number of orders a person received and their access to higher-paying jobs. The researchers also identified limitations in studying these systems; while they could pinpoint certain variables used by the platforms, they lacked insight into how these variables were combined and the weight each carried in a decision. This absence of information often forced workers to “decrypt the rules of engagement” themselves to enhance their performance or earnings.
Uncovering Hidden Economic Consequences
During the fieldwork, the research reconstructed delivery systems that considered aspects such as completed orders, acceptance rates, cancellations, and customer evaluations. A more advantageous position could correlate with access to specific orders, peak hours, promotions, or better rates. This illustrates one of the economic consequences of opacity: a seemingly minor variable can ultimately influence work opportunities and earnings, even if the individual lacks a clear understanding of how that relationship was established.
This mechanism also demonstrates why variables are not always as neutral as they appear. The study indicated that some women opted to decline rides to certain areas for safety reasons or caregiving responsibilities, yet systems that rewarded high acceptance rates could inadvertently penalise those choices. This finding does not imply that an algorithm was designed with the intent to disadvantage them; rather, it highlights how a rule that appears uniform can perpetuate or exacerbate existing inequalities outside the application.
Racial Bias and Algorithmic Impact
Similar issues have emerged concerning ratings. Research conducted by the Yale School of Management and the National University of Singapore concluded that racially biased evaluations from some clients could be embedded in ratings, affecting subsequent decisions within delivery platforms. The analysis examined over 86,000 services contracted through a home services platform, such as electrical repairs, plumbing, or maintenance. In this market, discrimination increased the proportion of ratings below five stars by 33.9% among workers perceived as belonging to minority groups, resulting in an income gap of 6.5%.
The Necessity for Effective Review Mechanisms
According to Mantilla-León, simply providing information on why an automated decision was made is insufficient; there must also be a mechanism for individuals to review such decisions, especially when they significantly impact their work, such as account suspension. Countries like Mexico, Colombia, and Uruguay recognise various forms of human review for automated decisions. However, Mantilla-León asserts that critical questions remain unanswered, such as the duration of the review process, the independence of the reviewer, and who can externally oversee the effectiveness of the process.
“The review mechanism risks becoming a mere formal procedure lacking the genuine capacity to rectify unjust decisions made by algorithmic systems. This remains one of the challenges that must be addressed,” she adds. For individuals whose income relies on their ability to connect, this distinction is crucial. A blockage that takes days or weeks to review can mean significant time without access to work, making the right to appeal inadequate if the process does not provide a prompt resolution.
The Challenge of Algorithm Oversight
Even if a platform clarifies which variables it utilises, a fundamental question lingers: who verifies that the system operates as the company claims? Mantilla-León identifies this as a primary obstacle in implementing new regulations. Oversight authorities require resources, specialised personnel, and technical expertise to scrutinise these systems. They must also be able to request sufficient information from companies to ensure compliance with legal obligations.
“Having the authority to audit an algorithm does not automatically confer the technical ability to do so,” warns Mantilla-León. Derechos Digitales asserts that it is essential for authorities to be equipped with the necessary tools to supervise compliance effectively. Another unresolved issue in the regulation of algorithmic management is the extent of access to information regarding these systems. The source code, which comprises the instructions that define how a program operates, can be a valuable tool for examining system functionality, although mere access to it does not guarantee that a worker will understand the reasoning behind a decision.
Towards Greater Transparency in Algorithmic Practices
Mantilla-León notes that Derechos Digitales has documented instances in Latin America where platforms have invoked trade secrets and intellectual property rights to reject audits of their algorithms or requests for access to source code. In response to such arguments, the organisation contends that the protection of this information should not be absolute when labour rights are at stake. Furthermore, disclosing an algorithm does not necessarily entail publishing the entire code. Mantilla-León suggests alternative approaches, such as independent external audits and algorithmic impact assessments, which can be employed to review these systems and identify potential errors, biases, or impacts on workers. Thus, transparency should not rest solely on the information a platform chooses to divulge.
Recognising Workers’ Rights in Algorithmic Transparency
New regulations in Latin America have begun to progress towards this goal. Acknowledging the right to receive information is a crucial first step; however, there remains much to resolve regarding what workers can truly know and what actions they can take when a decision adversely affects their income or access to work. Until this is achieved, those operating behind an app may be aware of a rule without fully understanding how it has impacted them, embodying one of the significant risks associated with algorithmic opacity. Yet, Mantilla-León insists that monitoring these systems should not solely depend on what is established by laws, platforms, or authorities: “It is vital to continue amplifying the voices of workers,” she concludes.