Platform work: Algorithmic management

A distinctive feature of platform work is that a large part of the managerial functions is carried out by algorithms. While increasingly present in traditional work settings - especially in logistics and warehousing - algorithmic management is a core component of digital labour platforms: without it, they wouldn’t be able to swiftly and efficiently match clients with service providers, allocate tasks, or supervise large and dispersed workforces. However, the fairness, transparency, and accountability of algorithmic decisions have raised concerns among stakeholders and policy makers, as highlighted by the Platform Work Directive(opens in new tab)This link opens in a new tab and national legislation in countries such as Belgium, Croatia, Greece, Malta, Portugal, and Spain.

Algorithmic management is defined as the use of computer-programmed procedures to monitor, organise, direct and evaluate workers. It relies on the collection and processing of worker data to enable automated or semi-automated decision-making.

Algorithmic management can be both AI and non-AI based: in non-AI based algorithmic management, algorithms follow pre-defined, fixed rules and operate based on logic programmed by humans: 'If X happens, then do Y' - for instance, if a worker is within a mile from the client, send request to accept task. In these systems, the code is human readable, and it is usually possible to trace why a certain decision was made. However, rule-based algorithms cannot learn or adapt to new situations. Notably, even if the rule-based algorithms and their codes can be traced back and/or predicted by humans, it doesn’t necessarily imply that workers know the code and understand the rationale of an algorithmic decision that affects them, as they may not be aware of all the rules and input data determining that decision.

In AI-based algorithmic management, algorithms use machine learning to make decisions: instead of adhering to fixed, pre-defined rules, they learn from training data, real-time interactions, and feedback to identify patterns and relationships. The AI builds a model (e.g. a logistic regression model, a classification model, a clustering model, a neural network, etc.) that represents these learned patterns and uses it to make decisions when faced with a new situation (e.g., assigning a task to a worker, deciding whether to offer ‘surge’ prices). In practice, algorithms feed relevant information (worker skills, task requirements) into the model and the model generates predictions (e.g., the likelihood of success that worker with a given set of skills may complete the request on time). While more flexible and able to handle complex, changing environments, AI based algorithmic management is less transparent and its decisions can be hard to understand, not only by workers, but even by developers themselves – for this reason it is often referred to as a ‘black box’. In addition, the quality and representativeness of the input data directly affects the fairness of the algorithm’s decisions.

To understand how algorithms are used to manage workers, it’s helpful to break down the concept of management into its key functions and assess the level of possible automation in each.

Traditional management theory (Koontz and O’Donnell) identifies five main managerial functions : planning, organising, staffing, directing (or leading), and controlling. Across these functions, tasks range from simple and standardised, therefore suitable for automation, to highly complex, requiring human intervention. In addition, some managerial tasks may not be automated because of regulatory constraints or company policies, regardless of technical feasibility. For instance, account deactivation requires human intervention, as stated by the Platform work directive.

According to traditional management theory, planning involves assessing the present, setting goals for the future, choosing strategies, and evaluating options. As of today, strategic planning cannot be fully automated as it requires vision and creative thinking. However, algorithms can formulate short term plans: for instance, they plan optimal routes in transportation or food delivery platforms; predict demand and adjust pricing by analysing historical performance data, real-time events, and external factors (e.g. weather and traffic for transportation tasks).

Organising is the process of identifying and classifying tasks, allocating responsibilities and enabling to perform them. In digital labour platforms, algorithms perform the organising function by matching clients with workers based on their skills, experience, and availability. For instance, in platforms mediating online tasks, algorithms facilitate matching by identifying the required skills in a client’s request, selecting best-matching workers based on profile and performance analysis, and finally only showing the client a pool of suitable candidates, therefore greatly reducing search time. In addition, platforms use algorithms to facilitate communication and coordination between workers, clients, and the platform itself (e.g. by sending a notification when a client approves a milestone or asks a question about a delivered file).

Staffing is the process of filling positions within the organisation through an effective selection; it includes personnel acquisition and personnel retention. Algorithms automate standard procedures, for instance verifying documents and conducting background checks, as well as more complex tasks, such as evaluating worker applications and profiles, based on predefined criteria and machine learning models. In theory, by focusing on objective criteria such as skills, qualifications, or experience, algorithms can promote fairer and more equitable hiring decisions. In practice, a lot depends on the potential bias present in training data and on algorithm’s architecture.

Directing and leading is the process by which a manager instructs, guides and oversees workers’ performance to achieve the planned goals. It involves influencing and motivating and can be considered the executive function of an organisation. Algorithms direct workers by providing detailed, step-by-step instructions on how a task must be carried out; they can lead and motivate them through gamified incentives, such as points, badges, or financial bonuses, or nudges towards a desired behaviour.

Controlling is the process of ensuring the objectives of the organisation are met, that all other functions are executed correctly, and activities comply with established standards. As well as setting performance standards, it involves measurement, evaluation, and corrective action. Algorithms can be used to control workers’ activities through various means, both during and after the execution of tasks. Algorithms can be created for continuously processing the performance metrics, such as delivery times, order accuracy, and client satisfaction; they track real-time worker location and task completion to enforce adherence to platform standards. Through this monitoring activity, algorithms collect large amounts of data on workers’ activities and performance, which are then used to evaluate, reward or discipline workers and eventually influence future work assignments.

The degree of direction and control exerted by the platform is a key determinant for the legal classification of platform workers’ employment status. The algorithmic matching of workers and clients is typical of an intermediary and therefore compatible with self-employment. However, if platforms use algorithms to give instructions, assign tasks, monitor workers’ locations and performance, and setting performance standards, they essentially do what an employer does, therefore indicating a dependent employment relationship.

A useful illustration of this point can be found in many of the court rulings collected in the Platform economy database. In general, when courts rule in favour of employee status of a platform worker to distinguish it from self-employment, they consistently cite factors such as the platform's control over work organisation, direction through algorithmic management, and the ability to impose sanctions, as seen in decisions involving on-location platform workers in France, Italy and Spain. These court rulings also emphasise the workers' lack of genuine autonomy, their dependency on the platform for income, and the platform's role in setting service conditions, exemplifying how algorithms can take over the ‘organising’, ‘directing and leading’ and ‘controlling’ functions. Conversely, rulings against employee status, notably in the Swedish Supreme Labour Court ruling and the Hungarian Supreme Court ruling on food riders, hinge on the perceived autonomy of the workers, the absence of direct hierarchical control, and the lack of a formal employment contract, as well as the distinction between coordination and subordination.

Arguably, one of the most positive aspects of algorithmic management is an increase in efficiency and the reduction in search times, both from the client and the workers’ perspective. In other words, clients can get faster access to large pools of suitable candidates based on the algorithm’s ability to match skills and requirements, while workers can get access to a larger selection of jobs, also across the globe (for online tasks), with potentially better pay. Another positive impact on working conditions is that algorithms can analyse real-time data to identify and mitigate safety risks (e.g. dangerous routes or bad weather conditions). In addition, gamified incentives may increase workers’ satisfaction and engagement, while instant feedback may help them improve their performance or identify skills gap.

However, algorithmic management has also been associated with some negative impacts, such as increased workload and time pressure, decreased autonomy, a sense of unfairness and lack of trust (EU-OSHA, 2024). In addition, even seemingly straightforward decisions regarding tasks allocation made by algorithms on the basis of performance metrics or ratings may be discriminatory, as they may fail to account for specific workers’ circumstances, such as care responsibilities and disabilities, which are protected by law in the EU, as exemplified by the decision made by two Italian courts, in Palermo(opens in new tab)This link opens in a new tab and Bologna(opens in new tab)This link opens in a new tab.

Several platforms across the EU have faced fines from data protection authorities for privacy infringements, specifically due to monitoring practices that exceed permissible levels for traditional employers – for instance processing biometric or geolocation data, or transferring sensitive personal data (including location, photos, payment details, and criminal/medical data) to the USA.

EU-OSHA (2024). Digital technologies at work and psychosocial risks: evidence and implications for occupational safety and health.