
The EU Platform Work Directive is usually discussed as a labour law question for the large platforms: the ride-hailing apps, delivery networks and freelancing marketplaces that set prices, rank workers and deactivate accounts at scale. That framing is accurate as far as it goes, and it leaves out the part of the market where many platform businesses sit. Member states must transpose the directive into national law by 2 December 2026, and the obligations it creates around algorithmic management are not reserved for companies with thousands of workers or in-house legal teams.
For small digital labour platforms and the SME intermediaries that supply software, staffing and scheduling to them, the practical question is narrower and more concrete: can the business explain, in writing, how its automated systems affect the people who work through it, and where a human being sits in that chain? Answering that question is not a communications exercise. It is a documentation project with a fixed end date, and the firms that start it in 2026 will find it considerably cheaper than those that begin in December.
The directive's core purpose is to improve working conditions for people who work through digital labour platforms. It establishes a presumption of employment that can be rebutted, and it sets out rules on the use of automated systems in managing that work. The algorithmic management provisions are the ones that reach furthest into ordinary business operations, because they are not confined to employment relationships. They apply to platform work more broadly, including arrangements formally classified as self-employment.
That matters for smaller operators. A regional platform matching cleaners to households, a niche marketplace for freelance translators, or an SME agency that runs a scheduling and scoring tool on behalf of client platforms may not think of itself as an algorithmic manager. But if software decides which worker sees which task, how a rating is calculated, whose profile appears first, or whose account is suspended, then the business is making automated decisions about people's livelihoods. The directive treats those decisions as something workers are entitled to have explained.
The obligations are procedural rather than prescriptive. They concern transparency about automated decision-making, human oversight of significant decisions, and the ability of workers to contest outcomes. None of that can be satisfied by a general statement of principle. It requires an account of specific systems, specific data and specific people.
The first task is an inventory, and it is usually larger than expected. Platform operations tend to accumulate automated tools incrementally: a task-allocation rule added to reduce idle time, a ratings formula adjusted after complaints, a fraud model that quietly triggers suspensions, a pricing engine that sets rates by demand. Each may have been introduced by a different manager, at a different time, for a defensible reason. Few will have been documented as a system that affects workers.
A workable inventory covers at least five categories:
For each entry, the file needs the system's purpose, the logic it applies in plain language, the data it draws on, the role of any human reviewer, and the route by which a worker can ask about or challenge an outcome. A generic HR policy or a broad statement about responsible technology will not answer a worker's question about why their acceptance rate dropped, and it will not satisfy a labour inspectorate either.
Most small platforms do not build their own ranking or scheduling engines. They license them. A workforce management suite, a marketplace plugin, a scoring module bolted onto an existing booking system — these arrive with contract terms drafted to protect the vendor, not to help the customer explain the system to a regulator.
That gap becomes a compliance risk the moment a worker asks for an explanation. If the platform cannot describe how a score is calculated because the logic sits inside a third-party product, the obligation still falls on the platform. Contracts signed before 2026 are unlikely to contain the necessary rights, and renegotiating them takes time.
Three clauses are worth pursuing in any vendor review: a right to receive a meaningful explanation of how the system produces outcomes affecting workers, a duty on the vendor to notify the platform of material changes to that logic, and a right to audit or obtain evidence of the system's behaviour. Where a vendor will not concede these, the platform should at least record the limitation and identify how it will manage the resulting gap.
Documentation that nobody owns decays within a quarter. The file needs a named accountable person — not a committee — with the authority to require answers from operations, product and procurement. In a small company that may be the operations lead or a founder, but the responsibility should be written down.
Version control matters as much as ownership. Algorithmic systems change: thresholds are tuned, new data sources are added, a model is replaced. A file that describes the platform as it operated in the month it was assembled will be wrong within weeks and misleading when it is most needed. Recording changes, with dates and reasons, turns a compliance snapshot into an operating record.
There is a commercial argument alongside the legal one. Clients procuring platform services increasingly ask suppliers how automated decisions about workers are governed, and workers themselves choose between platforms partly on how predictable and contestable the rules feel. A maintained file answers both audiences with evidence rather than assurance, and it shortens the response time when a dispute, an audit or a due diligence questionnaire arrives.
The work is not evenly distributed across 2026. A sensible sequence begins with the inventory, because everything else depends on knowing which systems exist. The next step is to test the file against real cases: take three recent decisions — a suspension, a rating drop, a shift allocation — and check whether the documentation would let the platform explain what happened and show where a human was involved. Where it would not, the gap is the work.
Vendor contracts should be reviewed in parallel, since renegotiation has its own timeline. Ownership and versioning can be settled early and cheaply. The final months before 2 December 2026 are better spent testing and refining than assembling.
The directive will be transposed differently across member states, and national implementing rules will add detail that no single template can anticipate. What will not change is the underlying test: whether a platform can account for the automated decisions it makes about people who work through it. That account is not a policy statement. It is a maintained record of how the software actually behaves, who oversees it, and how a worker can push back — and for small platforms and their SME suppliers, building it is the substantive preparation the 2026 deadline demands.