Introduction
An investigation by Mada Masr in 2022 revealed that a worker at “Talabat” used to receive 15 EGP per order if they maintained a good rating, while pay dropped to 13 or 12 EGP as the rating declined. Thus, a worker’s position within the rating system became a factor in determining their pay, alongside the nature of the task itself. The rating was influenced by factors such as adherence to shift schedules and order-acceptance rules, as well as circumstances beyond the worker’s full control, such as restaurant delays or communication issues with the customer.
This case does not prove that “Talabat” set a unique rate for each worker based on their economic needs or estimated the minimum amount they would accept for a job. Still, it does demonstrate that a job platform in Egypt used worker-related data to categorize workers into different pay tiers. As platforms expand their ability to collect data on acceptance and rejection rates, activity patterns, ratings, and past behavior, the question becomes broader than simply pay variation based on task; it also concerns the extent to which accumulated knowledge about the worker can be used to determine the price offered.
In another context, female workers interviewed by the Roosevelt Institute in the United States reported that different shifts or pay rates could appear for different workers even when the location, timing, and nature of the work were almost the same, while some shifts might not appear to them at all. These reports do not provide a precise comparison between two screens at the same moment, but they reveal the problem that each worker sees their offer individually, without knowing the range of pay available for the same work or the criteria that explain the differences. The Roosevelt report documented reports of varying shift availability and pay without revealing the formulas that determine these differences.
These cases should not be treated as a general condition applicable to all platform workers. They do, however, illustrate a broader shift that algorithmic management and pricing systems can bring about in the relationship between workers and their pay. The platform may adjust compensation in line with prevailing market conditions, raising it when demand increases or lowering it as the number of available workers grows.
However, this personalization takes a different form when the worker’s own characteristics and behavioral history are factored into the pricing process: the platform uses data such as acceptance and rejection rates and activity patterns to estimate each individual’s acceptance rate. In this case, the rate is no longer determined solely by the task’s characteristics and conditions or by published professional standards; it may also be influenced by the platform’s assessment of the worker’s ability to reject the offer.
Researcher Veena Dubal coined the term “algorithmic wage discrimination” to describe a shift in pay-setting in which workers’ pay can be individualized through variable formulas that draw on detailed data about each worker, including their behavior and location, alongside supply-and-demand data. As a result, workers may receive different pay for substantially similar work. Dubal’s concept extends beyond task pricing to encompass algorithmic decisions about how work is allocated and who receives particular opportunities, because these decisions also affect workers’ actual earnings and their access to work.
Dubal links this practice to the transfer of “price discrimination” from consumer markets to labor markets. Just as a company can use information to estimate the price a consumer may be willing to pay, a platform can, in principle, use what it knows about workers to tailor pay and incentives to reduce labor costs or encourage particular worker behavior. This system operates within a clear information disparity. The platform collects extensive data on workers and the market. It uses it to determine pay and allocate opportunities. In contrast, the worker typically does not know why they received this specific offer or how it compares to what was offered to others.
The concept of “algorithmic wage discrimination” is central to this paper because it shifts the analysis of algorithmic pay-setting from the technical question of how prices are calculated to the employment relationship itself. The concern arises when a platform’s accumulated knowledge enables it to differentiate among workers and individualize their pay, while workers lack the information needed to compare their offers, understand how they were determined, or negotiate them. Individualized pay-setting then becomes a way of exercising power within the employment relationship, rather than merely a more complex method of calculating pay.
This paper examines how platforms can shift from setting pay based on task characteristics and market conditions to individualizing pay and allocating work opportunities based on each worker’s data and behavior. It distinguishes dynamic pricing from individualized pricing and reviews cases from ride-hailing, delivery, and app-based nursing, while clarifying the limits of the available evidence.
The paper also analyzes how opaque pay-setting criteria and the absence of comparative data affect workers’ ability to assess their pay, detect discrimination, and negotiate their compensation. It proposes safeguards that would enable workers to know the range, median, and distribution of pay for comparable work; understand and challenge the factors influencing pay-setting; restrict the use of data unrelated to the value of the work; and strengthen oversight and collective bargaining.
From Dynamic to Individualized Pricing
In its non‑individualized form, dynamic pricing adjusts work compensation according to shared factors of the task or the market, such as peak hours, a shortage of workers in a particular area, or the need to complete the task quickly. The basis for comparison remains clear when workers who meet the same conditions are subject to a common pricing rule linked to time, location, or the difficulty of the work. Individualized pricing, on the other hand, incorporates an individual’s characteristics or history among the factors influencing the offer, and both types may coexist within the same system.
Individualized pricing incorporates the worker’s profile into the pay-determination process. Documented or disclosed inputs in some algorithmic management systems include ratings, task completion records, acceptance or rejection rates, and professional qualifications. The technical architecture may derive other indicators from response times, typical activity periods, previously accepted amounts, distances traveled, or device characteristics, without that implying they are actually used in pay pricing. Some inputs have a lawful connection to the work, such as a professional certification or a reliable quality standard. Others, however, if used to lower the offer, measure a worker’s bargaining power more than the value of the work they would deliver.
Individualization may take place at six interconnected points. The platform selects who sees the task; determines the price shown to each person; arranges offers so that better options are hidden behind weaker ones; adjusts incentives and bonuses; decides on fees or deductions; and then uses ratings and eligibility criteria to determine access to subsequent opportunities. The algorithm does not need to explicitly assign different amounts to each worker to create an individualized effect; sometimes it is enough to conceal the higher‑paid task from a category of workers, or to display it later, or to make access to it dependent on past behavior whose significance is not known to the worker.
This pattern is difficult to detect because each worker sees only the offers that appear to them. In some traditional workplaces, particularly where collective agreements, job grades, or published pay scales exist, workers have a common reference for comparison. They can sometimes learn their colleagues’ wages and compare them to their own or to what they are being offered for the same work.
On platforms, however, offers and their ordering may differ from one account to another, tasks may disappear quickly, and the application may not provide a complete record of what was shown and subsequently withdrawn. As a result, each worker sees only a limited part of the market, making it difficult to determine from their own account whether the offer they received was comparable to those received by others or part of a broader pattern of pay differentiation.
The assessment of individualized pricing begins with the right to just and favorable conditions of work, including fair remuneration and equal pay for work of equal value without discrimination, as recognized in Article 7 of the International Covenant on Economic, Social and Cultural Rights (ICESCR). The Committee on Economic, Social and Cultural Rights explains that fair remuneration should be based on objective criteria, including the responsibilities involved, the skills required, working conditions, and work-related risks.
Determining whether work is of equal value likewise requires an assessment based on objective criteria. Convention No. 100 of the International Labour Organization, in turn, underscores the importance of objective job evaluation based on the work performed, and permits differences in pay where they reflect objective differences in the work itself.
On this basis, workers should not bear the burden of proving how a pricing system operates when they have no access to the data underlying it. Where a worker or their representative presents reasonable indications of differences in pay for comparable work, the competent authority should be able to examine records of the offers and the factors that informed the determination of pay, while the worker should be able to understand the reason for the difference and challenge it. This is consistent with the Committee’s emphasis on the need for effective mechanisms of oversight, accountability, and remedy where the right to just and favorable conditions of work is violated.
When examining a pay disparity, the question is whether it is based on objective factors related to the nature and requirements of the work, or on characteristics of the worker that do not justify a difference in compensation for work of equal value.
It is not sufficient for a platform to attribute pay differences to how tasks are allocated or how offers are determined for individual workers. It should explain why one worker received less than another for comparable work. A difference in pay may, for example, be justified where a task involves exposure to psychologically harmful content or night driving under dangerous conditions, where the additional pay is linked to those conditions and applies to everyone performing the same task. Therefore, the assessment concerns whether the pay difference is related to the nature and requirements of the work or to worker characteristics the platform uses to reduce the amount offered.
This distinction also helps avoid conflating individualized pricing with wage differences across countries. Digital outsourcing companies may pay workers in different countries different rates, taking advantage of differences in the cost of living, currency values, and workers’ limited ability to relocate. This practice warrants separate criticism for the potential exploitation it may entail, but it does not prove that workers within the same market received different compensation because the platform determined that one would accept a lower wage. The analysis here focuses on the platform’s use of worker-specific data to determine different pay for comparable work.
What Do the Real-World Cases Prove?
One Shift, Two Offers
The app-based nursing market provides a clear example of this problem. Nurses interviewed by the Roosevelt Institute described their workday as beginning with searching through various shifts. They might be asked to state their acceptable pay or wait for an offer to appear shortly.
Some platforms may also offer different shifts or pay rates to different workers even when the hospital, timing, and nature of the work are similar. In some cases, nurses compete for a shift by each offering to accept a lower rate than the other until one of them gets the job. In this way, the application does not merely offer opportunities; it determines who sees them and how workers compete for them by leveraging its data on each worker’s history and behavior.
Fourteen out of the twenty‑nine female workers interviewed by the Roosevelt Institute reported that their income from the apps alone was insufficient to make a living. These statements do not prove that the platforms used the workers’ need to determine their pay, but they highlight the importance of economic vulnerability in this discussion; the more a worker relies on the platform for income, the greater the risk of using their data and past behavior to infer their willingness to accept lower pay. The problem is exacerbated when the platform possesses this knowledge while the worker remains unable to compare what they are offered with what others receive for similar work.
The Roosevelt report likewise does not establish that the platforms use specific personal data to determine each worker’s pay, nor does it disclose the technical reasons for differences between offers. It does, however, document repeated accounts of differences in the shifts and pay rates offered to workers operating under similar conditions, without providing a way to understand why these differences occur or to compare their offers with those made to others.
Uber and the Limits of Evidence
Researchers at the University of Oxford analyzed data from approximately 1.5 million trips completed by 258 Uber drivers in the United Kingdom, comparing their conditions before and after the introduction of the dynamic pricing system. They found that average real income per working hour dropped from £22.20 to £19.06 before deducting operating costs. This calculation was based on the court’s definition of working time, which includes the period during which a driver is logged into the app and available to accept trips, including waiting time.
The driver’s average share of the fare remained close to 75%, compared to approximately 25% for Uber. However, the median driver share fell to 71%, compared to 29% for Uber. The median here refers to the rate that falls in the middle of the distribution of driver shares, with half of drivers above it and half below it. The difference between the mean and the median indicates that the overall average may not reflect the share received by the driver at the middle of the distribution. The Oxford research also found that pay and trip allocation became less predictable, and that income differences among drivers increased after the introduction of dynamic pricing.
The study shows that a non-transparent dynamic pricing system can shift a larger share of a trip’s value to the platform while making drivers’ earnings less predictable. The researchers attributed a significant part of this outcome to the platform’s expanded power to set passenger fares and driver pay separately. This distinction matters because harm can arise from the platform’s exclusive control over market information and both sides of the pricing process, even before there is evidence that pay is individualized based on each driver’s vulnerability.
This study reveals the limits of what a driver can know from their individual experience. The simultaneous changes in price and trip allocation make it difficult to determine the reason for the decline in earnings from a single trip or a single account record. The change in the company’s share and the disparity between drivers only became apparent after collecting and comparing data from a large number of trips. This made it possible to distinguish between normal daily fluctuations and a recurring pattern that can be measured and analyzed.
Lyft: Disclosed Individualization of Incentives
Lyft (a U.S. ride-hailing service) provides a direct example of individualizing one component of compensation at the individual worker level. On its Earnings Challenges program help page, the company explains that earnings-based offers are personalized for each recipient. Under the program, the driver receives a bonus for reaching a specified earnings target within a given period, and the platform does not require all drivers to meet the same target or offer the same bonus. In this way, the individualization of incentives becomes a fact acknowledged by the platform itself, rather than merely a technical possibility.
The program does not reveal how Lyft determines the value of the offer each driver receives, or whether it uses the driver’s economic need or history of offer acceptance in this decision. It proves, however, that part of compensation may vary from one worker to another within the same platform, while the reasons for this variation remain beyond what the worker can discern from the offer alone.
Shipt: The Loss Revealed by Workers through Their Collected Data
Shipt, a U.S.-based on-demand shopping and delivery platform, illustrates the importance of collective comparison in assessing the effects of pay algorithms. The company initially used a relatively transparent pay formula but replaced it in 2020 with a system that calculated pay based on the estimated effort required. Because the details of the new formula were not disclosed, workers collaborated with researchers to develop a tool to collect and compare their pay data. A study based on data provided by 140 workers found that average pay per order increased, yet nearly half of the participants experienced undisclosed pay reductions. Some also recorded periods of work during which their pay fell below the minimum wage in their states.
The study shows that the increase in the average pay across the Shipt platform did not mean that all workers benefited from the new system, as its effects varied considerably among workers. While overall average pay increased, the income of a substantial group of participants declined. This disparity could not have been detected from the data associated with a single worker’s account; it became apparent only after data from multiple workers were collected and compared.
DoorDash and #DeclineNow: Rejecting Offers Until the Pay Rises
The #DeclineNow campaign reveals another aspect of the platform’s ability to adjust pay in response to workers’ responses to offers. DoorDash drivers (a U.S. on-demand food and goods delivery platform) organized to reject low-priced orders. The campaign relied on workers’ observation that a rejected order could subsequently be offered to other workers at a higher rate, and that repeated rejection could lead to a gradual increase in offers. According to reports on the campaign, the number of members in its associated forum reached approximately 40,000.
The campaign showed that DoorDash’s price offers are not necessarily a fixed amount tied to a task. The platform can begin with a low offer and adjust it as workers continue to reject the order, until it finds a worker willing to accept it. The campaign also demonstrates that collective refusal can partially disrupt this mechanism as the cost of finding a worker willing to perform the task at that rate increases.
“Talabat” in Cairo: Ratings Affect Workers’ Pay
“Talabat” in Cairo presents one of the clearest examples in Egypt of how data related to workers themselves is used to determine remuneration. In April 2022, the company’s delivery workers went on a three-day strike demanding an increase in the delivery fee.
An investigation by “Mada Masr” into the strike and working conditions revealed that the amount a worker received for a delivery was linked to their rating on the app: 15 EGP for workers with a “good” rating, compared to 13 or 12 pounds for those with a lower rating. The rating was thus not merely a visible indicator of a worker’s quality, but was directly linked to the compensation they received for delivery.
The investigation also noted that a driver’s rating could drop due to the restaurant’s delay in preparing the order or the driver contacting the customer to inquire about the address. Furthermore, ending a shift before its scheduled time or taking a break was treated as an absence or lateness, resulting in an automatic reduction in the rating.
The investigation documented that the rating, a feature linked to the worker’s profile within the application, determined the level of compensation they received, and that a change in the rating would move the worker from one pay tier to another.
The rating was also influenced by the quality of delivery, compliance with platform rules, and circumstances beyond the worker’s control. This example illustrates how categorizing workers based on individual data led to a direct difference in the compensation they received for the same work.
“Uber” in Egypt: Variable Commission and an Unclear Pricing Rule for Drivers
The experience of Uber drivers in Egypt provides a less conclusive example, but it highlights another difficulty: identifying the rule that determines workers’ net earnings. In its review of platform workers’ protests, a Mada Masr report noted that drivers gathered outside the company’s headquarters in downtown Cairo in 2021 to demand lower commission rates. One driver reported that Uber’s commission averaged 27% and reached 35% in some cases, adding that the rate was determined by a company policy whose basis drivers did not understand.
The commission directly affects the share of the fare retained by the driver. Variations in commission rates therefore raise the same concerns as differences in pay. Drivers need to know which rule determines their actual earnings and whether the same rule applies to others.
- Pay Requires a Basis for Comparison
The International Covenant on Economic, Social and Cultural Rights (ICESCR) recognizes everyone’s right to just and favorable conditions of work, including fair remuneration, equal pay for work of equal value, and a decent standard of living for workers and their families. The General Comment No. 23 explains that the assessment of fair remuneration should be based on objective criteria and that equality concerns the actual value of the work, not merely the similarity of job titles. However, applying these rules requires the ability to determine whether pay differences reflect genuine differences in the work or an unjustified disparity between workers.
It is difficult for a worker to prove that they received lower pay for work of equal value if they do not know what was offered to others or how their own pay was determined. Some systems may allow the use of statistical data or require the employer to provide records. Therefore, proof does not always require knowing a specific colleague’s pay. Neither the Covenant nor the General Comment provides for pay comparability as a standalone right. However, this option represents a practical requirement that helps in proving unjustified pay disparities.
Comparing pay does not require identifying another worker or disclosing their individual earnings. In platform work, workers need to know the pay range for comparable tasks and the median, the midpoint when those amounts are arranged from lowest to highest. The average can also help indicate the general level of the offers. Workers also need information on the number of offers, the reasons for differences in time, location, or risk, and any fees, deductions, or unpaid time. This information can be provided in aggregated form to protect workers’ privacy.
A task title alone is not sufficient to determine whether work is of equal value. A platform may use different titles for similar work, or the same title for tasks that differ in their requirements. Comparisons should therefore be based on skill, effort, responsibility, and working conditions. If two tasks require the same level of training, effort, and responsibility, they may be comparable even if they have different titles. Conversely, differences in risk or responsibility may justify pay differences between two tasks with the same title.
The European pay transparency rules provide a partial example of this logic. The Pay Transparency Directive requires Member States to enable workers to request information about their pay level and the average pay levels for categories of workers performing the same work or work of equal value, broken down by gender, and to provide access to the criteria used for determining pay and career progression. These safeguards are not automatically available to every contractor working through a platform, but they illustrate the feasibility of providing aggregated pay data while protecting privacy.
In platform work, comparative pay information should be available to workers before they accept a task, so that they can assess the offer promptly. They should also be informed of the reason why their offer deviates from the usual rate, and, upon completion of the task, receive a record detailing the pay, fees, actual time worked, and any decision that resulted in a reduction in compensation.
This information is also useful for collective bargaining. An individual worker typically sees only the offer displayed on their screen. In contrast, a group of workers, by comparing their data, may identify a persistent reduction in pay for a particular group, discover that a reduction in the base rate accompanied a bonus, or find that waiting time has become a cost borne by workers. This can shift bargaining away from isolated individual offers toward the rules that determine pay.
However, access to this information does not eliminate the power imbalance between workers and platforms. A worker may know that their offer is below the prevailing rate and still be compelled to accept it. Transparency therefore needs to be accompanied by other safeguards, including a minimum wage, workers’ right to organize, limits on the data that may be used to determine compensation, and an oversight authority empowered to examine records and require corrective action or redress.
Pay comparison should not result in the disclosure of workers’ personal data. Data relating to a small number of workers in a particular area or occupation may make it possible to identify them. Workers should not be compelled to disclose information about their health, income, or family circumstances to prove pay disparities. The platform should therefore be responsible for aggregating and anonymizing data, while making more detailed information available to oversight authorities and workers’ representatives in accordance with appropriate safeguards.
In all cases, performance should not be used as a general justification for differences in pay. Ratings may be affected by customer bias, while a worker’s task acceptance rate may reflect their willingness to accept low pay more than the quality of their work. Response times may also be affected by disability, caregiving responsibilities, or the quality of the worker’s internet connection. The platform must therefore demonstrate the relevance of each criterion to the work, periodically review its impact, and enable the worker to correct their data and challenge the outcomes based on it.
Data and Bargaining Power
The platform may record the worker’s acceptance and rejection of offers, their waiting time, and the frequency of their return to the application. A distinction must be made between the platform’s ability to collect this data, its actual collection, and its subsequent use in determining pay or allocating tasks; proof of one stage does not prove the others. However, having data on a large number of workers enables the platform to test different rates and monitor workers’ responses, while each worker sees only the offers displayed to them. Therefore, accountability requires knowing what data was collected and for what purpose it was used.
The algorithm does not need explicit information about the worker’s poverty or family responsibilities to infer their economic conditions. Data such as the timing of offer acceptance, the frequency of returning to the application, the postal code, or the type of device may be correlated with income or the degree of dependence on the work. Accordingly, the oversight authority should examine the data collected by the platform, the way the platform uses this data to determine pay, and whether it infers information from it that is not related to the work itself.
Pricing based on workers’ vulnerability may also intersect with legally prohibited discrimination, though the two do not always coincide. Using certain indicators may harm legally protected groups, such as women, people with disabilities, or specific ethnic groups. A worker burdened with debt or who has lost another source of income may also be harmed, even if their case does not fall under anti-discrimination laws. Therefore, these cases require safeguards for wages and data to prevent exploitation of workers’ vulnerability and inability to refuse offers.
Individualized pay can also weaken workers’ ability to establish a shared reference rate. When each worker receives different offers and incentives without knowing why, it becomes difficult to determine whether lower pay is an isolated experience or part of a broader pattern affecting many workers. This makes it harder to identify the platform’s overall pay-setting practices and develop collective demands concerning pay.
In a study on pay reductions in platform work, researchers presented a theoretical mathematical model to examine the effects of information asymmetry between platforms and workers. The model assumes that the platform offers sequential prices to accomplish a set of tasks, and that a worker accepts a task if the offer reaches the threshold they deem appropriate for performing it.
The researchers demonstrate that if the platform has information about the distribution of prices workers are willing to accept and can wait for someone to accept a low offer, it can, under certain conditions, reduce the total amount it pays. They also demonstrate that the existence of a minimum price, or coordination among workers to reject offers below that price, limits this ability. The model does not prove that a real-world platform uses this mechanism. Still, it illustrates theoretically how information asymmetry and differences in workers’ willingness to accept offers can give the platform greater power in determining pay.
The impact of individual refusal remains limited if the platform can record the refusal, then provide another offer or assign the task to a different worker. By contrast, sharing pay and offer data among workers allows them to see what is not visible through an individual account. Anonymized collective records, union-based comparison tools, and workers’ representatives’ right to examine the system can help bridge the gap between what the platform knows and what workers know about the market.
This does not mean that worker data should be excluded from all pay-setting decisions. Professional qualifications, experience, or longer travel distances may be legitimate factors when they are directly related to the work. By contrast, using a worker’s history of accepting offers to reduce a later offer does not measure the value of the task. It uses the worker’s previous response to a price to predict what they may accept in the future. Any data used to set pay should therefore be necessary for a legitimate purpose, relevant to the work, and subject to review and challenge.
The Law Addresses Decisions but Not Wage Distribution
Rules and regulations governing labor and wages, data protection, and algorithmic governance address separate aspects of the problem of platforms using workers’ data to determine compensation or allocate job opportunities. However, they do not combine these into a single protective framework.
Some rules require platforms to inform workers of their pay and deductions, while others allow workers to request an explanation of automated decisions, restrict the use of certain types of data, or enable pay comparisons in specific contexts, such as equal pay claims. Nevertheless, these safeguards remain fragmented across different legal frameworks and do not, together, provide comprehensive protection for workers.
Convention No. 193 on Work in the Platform Economy
On 12 June 2026, the International Labor Conference adopted Convention No. 193 on work in the platform economy, the first international labor standard dedicated specifically to this sector. Some of its protections apply to platform workers regardless of their contractual status and address remuneration, freedom of association and collective bargaining, non-discrimination, occupational safety and health, and access to remedies. As of the date of this paper, the NORMLEX database did not record any ratifications of the Convention. According to Article 27, the Convention enters into force twelve months after the registration of ratifications by two states.
Articles 10 and 11 regulate aspects of remuneration and transparency. Article 10 provides for the payment of remuneration in full and on time and, with respect to workers in an employment relationship, requires the guarantee of a minimum wage and the reimbursement of work-related expenses. It also requires states to consider extending the guarantee of a minimum wage to platform workers who are not in an employment relationship.
Article 11 further guarantees that workers receive clear and timely information about remuneration and deductions. Additionally, Article 13 establishes their right to be informed about automated systems that monitor, evaluate, or make decisions affecting their working conditions or access to work opportunities.
Articles 14 to 16 add safeguards concerning automated systems and data. They call for the use of such systems in a manner consistent with fundamental rights, enable workers to request a written explanation for certain significant adverse decisions, and, in specified cases, provide the possibility of reviewing decisions such as non-payment, account suspension, or deactivation. They also require specifying the purposes for processing workers’ data and ensuring workers’ rights to access, correct, and delete such data.
However, the Convention does not expressly grant workers a right to know the pay offered to other workers for comparable tasks, or to access information on the median and distribution of such pay. Nor does it expressly prohibit the use of a worker’s history of accepting offers to infer their willingness to accept a lower rate.
The Convention therefore enhances workers’ right to know their own pay and understand how automated systems affect them, but it does not fully bridge the pay comparability gap discussed in this paper. The extent to which this gap is addressed will depend on national legislation, inspection mechanisms, and collective bargaining.
European Union
The European Platform Work Directive regulates the use of automated systems in the management of workers. Its provisions include decisions affecting income, task pricing, and work allocation. The Directive requires platforms to inform workers and their representatives about the systems they use, the categories of data processed, and the decisions affected by those systems.
It also requires human oversight, periodic assessments of certain impacts of these systems, and the provision of explanations and review of certain significant decisions. Member states are required to implement the Directive into their national laws by 2 December 2026. Therefore, not all of its provisions will be applied in the same way in every country until this transposition is complete.
The Directive restricts the types of data that may be used in algorithmic management. Among other things, it prohibits the processing of data concerning workers’ emotional or psychological state, private conversations, certain data collected outside working hours, and data used to predict the exercise of fundamental rights, such as trade union activity.
It also prohibits the inference of certain sensitive information, including racial or ethnic origin, beliefs, disability, health status, and trade union membership. However, it does not expressly prohibit the use of indicators such as poverty, debt, or a history of accepting low pay. The potential use of data that reveals worker vulnerability must therefore be addressed through specific safeguards.
The European AI Act, by contrast, focuses on the risks of the systems themselves. It classifies certain AI applications used in recruitment, worker management, task allocation, and performance monitoring as high‑risk and imposes requirements on them relating to risk management, data quality, documentation, and human oversight. These requirements, however, do not, by themselves, determine whether a pay policy is fair: a system may be accurate and well-documented while still applying an unfair pay rule.
These rules complement the Pay Transparency Directive, but each text has a different scope. The Platform Work Directive regulates data and automated decisions; the Pay Transparency Directive enables comparison within the framework of equal pay between women and men; and the AI Act regulates the risks and governance of systems.
The Colorado Initiative
Colorado’s experience provides a direct example of a legislative attempt to regulate individualized pay based on worker data. A 2025 bill sought to prohibit automated systems from using data derived from worker monitoring or from inferences about workers’ characteristics and behavior to determine individual pay. The bill defined monitoring data to include information about personal characteristics, behavior, and biometric data. However, it did not become law after the House Judiciary Committee voted on April 22, 2025, to postpone it indefinitely.
In 2026, Colorado lawmakers revisited the issue with a more detailed bill. It prohibited the use of algorithms that made monitoring data a factor in determining an individual worker’s pay. However, it permitted pay differences based solely on seniority or the nature of the tasks performed, provided that workers were informed about the data used and how it affected their pay. The bill also required entities that use such systems to establish procedures by which workers could access, correct, or challenge the data collected about them. Both chambers of the state legislature passed the bill, but the governor vetoed it on June 2, 2026, preventing it from becoming law.
This approach is consistent with the distinction proposed in this paper that not all individualized pay is unlawful. A pay difference may be justified where it is based on experience or the nature of the work. In contrast, the concern arises when data about a worker’s behavior or conditions are used to determine what they are likely to accept.
The bill demonstrates that regulation should not be limited to prohibiting specific categories of data; it should also require platforms to justify the relevance of each factor to the work, disclose the data that affect pay decisions, and enable workers to correct and challenge it. At the same time, the failure of both bills illustrates that translating this principle into an enforceable legal prohibition remains a matter of controversy regarding the limits of legitimate individualization of pay and the exceptions that should be permitted.
Limits of Legal Protection for Workers against Individualized Pricing in Egypt
Egypt’s Labor Law No. 14 of 2025 creates a potential basis for protecting some platform workers if they can establish that they work for pay, for an employer, and under that employer’s management or supervision. Article 96 does not limit “new forms of work” to remote or flexible work; instead, it allows other forms to be included.
Article 97 extends the rights applicable to traditional employment relationships to workers in these new forms, including minimum-wage protection, collective bargaining, and freedom of association. Article 99 also allows an employment relationship to be established by all forms of evidence. The law also requires employers to retain data and records concerning workers’ pay and any subsequent changes to it.
However, the law does not explicitly specify how supervision is to be established when the platform exercises it through price‑setting, the ordering of work opportunities, rating, or control over access to tasks. Even if an employment relationship is established, the law does not grant the worker a clear right to know what was offered to others for comparable work, nor does it establish specific rules governing the use of their behavioral data in pay determination.
Protection, therefore, requires that the platform’s control over price, rating, and access to work be considered among the indicators that may be relied upon to establish management or supervision, and that the platform retain data enabling the interpretation and comparison of pay differences.
Egypt’s Personal Data Protection Law No. 151 of 2020 provides another avenue, as it requires that data collection and processing be for legitimate, specific, and declared purposes, and establishes rights of access, rectification, and objection to certain forms of processing. These rules can help the worker know what data the platform holds about them, but they do not determine when the use of such data in pay setting is legitimate, nor do they grant the worker a right to compare their pay with that of other workers. This gap requires specific rules governing inferences that affect pay, as well as records that clarify the data used and its impact on pay determination.
By contrast, Egypt’s Competition Law No. 3 of 2005 on the Protection of Competition and the Prohibition of Monopolistic Practices addresses market power and practices that restrict competition. These rules could become relevant where a platform exercises substantial buyer power in the labor market or controls workers’ access to demand.
The law, however, does not establish a specific framework for buyer power in labor markets, does not address every individualized difference in pay, and does not give workers direct access to comparative pay data. Its application to platform labor markets therefore requires integrating the impact of platform power on the terms of labor purchase, and clearer coordination between the Competition Protection Authority and the authorities competent for labor and data.
The gap in the Egyptian legal framework is that each law addresses a different part of the problem. Labor law protects pay where an employment relationship is established; data protection law regulates the collection and use of workers’ data; and competition law may intervene where market power results in a restriction of competition. However, no framework brings these dimensions together when a platform uses workers’ data to determine individualized pay, while workers lack a right to know where their pay stands relative to offers for comparable work and to challenge the basis for the difference.
Feasible Protection Conditions
The principles proposed in this paper can be framed as a rule against pricing that relies on workers’ vulnerability. This is intended to prevent the platform from lowering pay or withholding better opportunities based on data revealing the worker’s need for income or their prior willingness to accept lower pay.
A Rule for the Lawfulness of the Data Used
A distinction can be made between three types of data. First, data that does not measure the value of the work, such as the worker’s history of accepting low offers or inferences about their income and debts. Such data should not be used to reduce their pay.
Second, factors directly related to the task, such as qualifications, experience, responsibility, risk, and distance, may be used if applied under a single rule to workers performing comparable work. Third, data whose meaning may change depending on how it is used, such as response speed, location, device type, and timing of activity, should not be used.
A fixed list of prohibited data is insufficient, as a system may infer the same information from other indicators. Each type of data should therefore be assessed according to the purpose of its use, its direct relevance to the work, its necessity, and whether the same purpose could be achieved through a less intrusive means. Data used to reduce pay should receive particular scrutiny when it indicates a worker’s need for income or limited ability to reject an offer but is unrelated to the value of the task.
Information That Can Be Used in Practice
This paper has already highlighted the need to enable workers to compare their offers with offers for comparable work before accepting a task. Added to this is the need to retain a record that can be referred to after the task is completed, including the offer presented to the worker, the time it appeared and was accepted, the final compensation and deductions, and the main factors that influenced it. These records should be retained for the period necessary to file complaints or claims, and their deletion should be suspended during any dispute or investigation to prevent evidence from being lost through system updates.
Not everyone needs the same level of access to data. Workers should receive the information necessary to understand and challenge their offers. Workers’ representatives and trade unions should have access to aggregated data showing the distribution of pay and offers and the factors used, without identifying individuals.
The oversight authority or an independent auditor, in turn, should be able, where necessary, to examine the relevant records, how the system operates, and the factors influencing its decisions. Trade secret protections should not be used to prevent the competent authority from determining the basis for differences in pay.
Collective Auditing and Enforcement
Systems capable of changing pay or workers’ access to work opportunities should be subject to independent audits before deployment, at regular intervals thereafter, and after any substantial changes. Audits should examine whether the factors used to determine pay are related to the work and whether they produce consistent differences among workers performing comparable work.
Workers’ representatives should participate in determining which issues to examine, while the oversight authority should have access to the complete data and methodology underlying the system. General findings should be made public without disclosing individual workers’ data.
When a worker or their representative presents reasonable evidence of an unexplained pay disparity, the platform should be responsible for providing the relevant records and explaining the reasons for the difference, since it controls the data. If the platform unjustifiably fails to provide the records, the law could allow an adverse inference to be drawn against it. Remedies should include payment of the pay difference, correction of data, redress for the harm, and suspension of the use of the unlawful system used to determine pay.
It must also be possible to file a complaint both within and outside the app, at no cost, with human review of important decisions. Reducing job offers, freezing accounts, or limiting job opportunities as a result of a complaint, the sharing of pay information, or union activity should be prohibited. A worker may not be formally terminated, but could be penalized with fewer assignments; therefore, oversight must include monitoring changes in the volume of available work following an objection complaint.
A Minimum Pay Floor Against Need-Based Pricing
Transparency is not sufficient if the base rate itself is so low that it forces the worker to accept it. For those with a proven employment relationship, the guarantees regarding minimum wage, working hours, and expenses apply in accordance with the relevant law.
Not all waiting or travel time automatically counts as working time. Still, the basis for its calculation is strengthened when the platform requires the worker’s presence or restricts their freedom during that time, as demonstrated by the issue of on-call time in the British Uber case.
As for platform workers who are legally classified as self-employed or independent contractors, Convention No. 193 does not directly guarantee them a minimum wage, but it does require States to consider extending the minimum wage established for workers in an employment relationship to them.
This paper proposes a broader rule to ensure an effective minimum level of compensation that accounts for the time and necessary costs involved in performing the task. This minimum rate would not prevent higher compensation based on experience, responsibility, or risk; rather, it would prevent platforms from lowering pay to a level that exploits a worker’s need for income or limited ability to refuse offers.
Public entities and large corporations that purchase services through platforms can impose these safeguards in their contracts. They may require a minimum level of pay, disclosure of the contractor chain, a prohibition on the use of demand indicators in pricing, and an audit right that extends to subcontractors. These conditions become all the more significant when an individual works through a platform on behalf of another organization with which they have no direct relationship.
No Pricing Based on Worker Vulnerability
Pay comparability reveals what workers cannot learn from their individual offers alone: whether they received different pay because of the work itself or because the system treated them differently. This distinction becomes especially important when a platform collects data on workers’ acceptance and rejection of offers and possesses extensive market knowledge that workers do not share.
Understanding how an algorithm works is not enough if the rule it applies is unfair. The key consideration is the reason for the difference in pay. Higher pay may be justified by skill, responsibility, or risk, but pay should not be reduced because data indicate that a particular worker is in greater need of income or less able to reject an offer. This is the boundary established by the principle of “no pricing based on worker vulnerability.”
Protecting this boundary requires three connected safeguards: enabling workers to compare their offers, requiring platforms to justify the data they use, and establishing collective and regulatory mechanisms capable of examining these systems and stopping unlawful practices. With these safeguards, differences in pay become decisions that can be explained and challenged, rather than outcomes that workers cannot understand or compare with offers made to others.
Conclusion
The problem with individualized pricing is not simply that pay differs across tasks or workers. Some differences may be justified by skill, responsibility, risk, or the cost of performing a task. The concern arises when a platform’s knowledge about an individual worker influences the offer they receive, allowing their past behavior, reliance on platform work, or willingness to accept offers to be used to reduce their pay.
The cases examined in this paper do not establish that all platforms use this pricing model. They do, however, show that the necessary infrastructure already exists in several forms, including personalized incentives, pay linked to ratings, price adjustments in response to workers’ acceptance or rejection of offers, and systems that collect extensive data about workers’ performance and behavior.
The impact of this capacity is greater because workers usually see only their own offers. When pay varies across accounts, an individual worker cannot determine whether the difference arises from the task or from the data held by the platform. Pay comparability is therefore essential to enforcing the right to fair pay and equal pay for work of equal value.
Workers need to know where their offers fall relative to comparable work and which factors explain the differences. This information also allows workers to identify patterns that remain hidden in individual accounts and to negotiate the rules governing pay instead of treating each offer as an isolated event.
The legal rules examined in this paper contain elements of protection, but these are divided across different fields. Some rules regulate pay, others govern data and automated decisions, and others address pay transparency or market power. Together, they still do not establish a clear right for workers to understand why their offers differ or to compare them with offers for comparable work.
This gap is particularly clear in Egypt. Labor law, data protection law, and competition law may each provide some protection. Still, no single framework governs the use of worker data to set individualized pay while also giving workers the means to detect and challenge unjustified differences.
The principle of “no pricing based on worker vulnerability” limits this power. It does not prevent platforms from using factors genuinely related to the value of the work. It prevents them from using a worker’s need for income or limited ability to reject an offer as a reason to pay less.
Enforcing this principle requires that pay differences remain comparable and explainable, that platforms maintain records that can be examined, and that workers, their representatives, and oversight authorities be able to challenge the use of data unrelated to the work.
The more accurately a platform can estimate the rate each worker may accept, the greater the need to prevent that knowledge from being used to obtain their labor at the lowest rate their personal circumstances compel them to accept. Pay should be justified by the work performed and the task’s requirements, not by what the platform knows about the point at which a worker can no longer afford to refuse.