Fairwork

Fairwork is the participatory-action-research framing and worker-engaged ratings methodology through which a global network of academic and civil-society researchers — coordinated by the Oxford Internet Institute and the WZB Berlin Social Science Center — evaluates digital labour platforms against five principles of fair platform work: fair pay, fair conditions, fair contracts, fair management, and fair representation. The framing was seeded by Mark Graham in a March 2017 Oxford Internet Institute blog post, formally launched as the Fairwork project in January 2018, and operationalised into the first country-scorecard ratings in South Africa and India in 2019. By 2026 the project has rated more than 700 digital labour platforms across 38 countries under annually-updated country scorecards covering ride-hail, food and grocery delivery, domestic-work platforms, cloudwork and microwork sites, and the AI supply chain. The framing's role inside the corpus is structurally adjacent to the worker-authored register of msg-pay-up and the digital-rights-advocacy register of msg-bossware: like both, it operates one degree removed from the worker-organising vehicles that carry its substantive demand, but its worker-engaged research methodology — direct surveys and interviews with platform workers form the load-bearing evidence base for each scorecard — sits closer to grassroots organising than the EFF-anchored bossware register, and its continuous country-by-country audit instrument supplies worker-organising vehicles and unions with citable evidence for pressure campaigns and legislative-engagement work.

Origin

The framing's academic seed is the 24 March 2017 Introducing a FairWork Foundation post by Professor Mark Graham of the Oxford Internet Institute. The post proposed "a way of holding client firms in virtual production networks more accountable through the development of a 'FairWork Foundation'" — a certification model explicitly drawing on the fair-trade-foundation precedent, in which an independent body audits production networks against a public set of principles and confers a certification that downstream consumers can use as a procurement signal. The proposal was framed as a response to the digital-labour conditions Graham's research at Oxford had been documenting through field studies of ride-hail drivers in Lagos and Nairobi, microtaskers on Amazon Mechanical Turk, and freelancers on Upwork and Fiverr — a labour-economy in which the workers' conditions had become structurally invisible to the end-customers, end-clients, and platform owners who consumed the labour.

The Fairwork project formally launched in January 2018 with Mark Graham as Principal Investigator. The five principles — fair pay, fair conditions, fair contracts, fair management, and fair representation — were refined through multi-stakeholder workshops held in Cape Town, Johannesburg, and Bangalore in 2018, with participation from platform workers, trade-union representatives, researchers, policymakers, and platform-management representatives. The workshops are the framing's load-bearing originating moment: the principles emerged not from a single academic paper but from a collaborative process designed to ensure that the rated criteria reflected workers' own categorisations of fair and unfair platform work in their specific labour-market contexts.

The first scorecards — released for South Africa and India in early 2019 — covered ride-hail platforms (Uber, Taxify), on-demand delivery platforms, and freelancer platforms; in the South Africa pilot, the freelancer platform NoSweat led the rankings with eight out of ten points, while several large international ride-hail platforms scored at or near zero on the same criteria. Graham's framing of the launch — "these ratings will enable consumers to make informed choices about the platforms and services they need" — situated the project's theory of change in the same consumer-procurement register that the fair-trade-foundation precedent had operated through, while opening the audit-and-implementation cycle through which platforms could negotiate substantive labour-conditions changes before the next year's scorecard.

The five principles

The five Fairwork principles supply the framing's analytical decomposition of fair platform work. Each principle carries a maximum of two points — one for a documented minimum threshold and one for a higher commitment — for a maximum scorecard score of ten. The principles, in summary:

• Fair pay. Workers, irrespective of their employment classification, should earn a decent income in their home jurisdiction after taking account of work-related costs and active hours worked, paid on time and for all work completed.
• Fair conditions. Platforms should have policies in place to protect workers from foundational risks arising from the processes of work, and should take proactive measures to protect and promote the health and safety of workers.
• Fair contracts. Terms and conditions should be transparent, concise, and always accessible to workers; the contracting party must be subject to local law and identified in the contract; clauses must not unreasonably exclude liability or prevent workers from seeking redress for grievances.
• Fair management. A documented due process must exist for decisions affecting workers; workers must have appeal rights for disciplinary actions and deactivations; the use of algorithms must be transparent and result in equitable outcomes.
• Fair representation. Platforms should provide a documented process through which worker voice can be expressed; workers, irrespective of their employment classification, must have the right to organise in collective bodies, and platforms must be prepared to cooperate and negotiate with them.

The principle architecture supplies the framing's bridge from earlier worker-rights vocabulary — the International Labour Organization's Decent Work agenda, the established human-rights register of occupational safety, freedom of association, and access to remedy — into the specific labour-relation of platform work. The "fair management" principle in particular operationalises the msg-algorithmic-management framing into a continuous audit instrument: it requires platforms to evidence that algorithmic decisions affecting workers are transparent, contestable, and subject to human oversight, and supplies worker-organising vehicles with the working language for pressure-campaign demands on that substantive question.

The pressure-and-praise methodology

The framing's working methodology is a three-method audit that produces the load-bearing evidence for each scorecard. Desk research first maps the platforms operating in a country, identifies platform management contacts, and analyses the platform's contracts, terms, policies, digital interfaces, and public information for compliance with the five principles. Worker data collection then proceeds through direct surveys and interviews — recruited on-platform, through social media, and through snowball sampling — with a target of six to ten worker interviews per location-based platform and at least ten workers per continent for cloudwork platforms operating across geographic boundaries. Manager interviews close the cycle: Fairwork approaches platform leadership requesting evidence on each principle, opens a dialogue in which the platform may agree to implement improvements before publication, and substitutes desk-research-and-worker-data findings when the platform declines participation. Scoring is determined through peer review involving local research teams, the central Oxford team, and two external reviewers from other Fairwork country teams, with points awarded only on clear evidence that a threshold has been met.

The methodology's theory of change is what the project's 2025 Environment and Planning A paper calls the pressure and praise dynamic: publication of league-table scorecards generates reputational pressure on low-scoring platforms — visible across press coverage, procurement decisions, investor scrutiny, and worker-organising campaigns — while affording high-scoring platforms a competitive recognition that creates a market incentive to improve. The implementation period preceding publication is the methodology's substantive engagement surface: platforms can negotiate concrete labour-conditions changes during that window, document the implementations, and earn the corresponding score points at publication. The Fairwork project reports that more than 300 pro-worker policy changes have resulted from this dynamic across rated platforms — implementations of local-minimum-wage floors, occupational-safety-and-health provisions, formal grievance procedures, anti-discrimination policies, and recognition of worker associations.

The pressure-and-praise dynamic is the framing's principal departure from the digital-rights-advocacy register of msg-bossware and the worker-authored register of msg-pay-up. Where bossware operates as a critical-naming framing without a continuous evidence base, and where Pay Up carries a worker-side organising demand into named legislative text, Fairwork operates as a continuous, comparable, multi-year audit instrument that produces annually-refreshed country scorecards. Each country team's scorecard becomes a citable evidence base for unions, civil-society advocates, journalists, and legislators — supplying the framing's load-bearing data infrastructure across more than a decade of platform-economy expansion.

Expansion: cloudwork and AI work

The framing has extended beyond the location-based platform work that anchored its 2018-2020 launch wave into two further strands. The Fairwork Cloudwork ratings strand, launched with the first cloudwork report in 2021, audits global online-task and microwork platforms against the same five principles. The first cloudwork ratings drew on surveys with 792 workers across 75 countries between July and November 2020 on 17 platforms, producing a working finding that most cloudwork platforms fail the minimum fair-work thresholds — Amazon Mechanical Turk, Rev, PeoplePerHour, and Freelancer scored at the bottom of the league table with zero to one points, while Jovoto and TranscribeMe led with seven out of ten. The second cloudwork wave in August 2022 expanded the platform set to include Prolific, Workana, Appen, 5 Euros, Clickworker, Scale/Remotasks, 99designs, Fiverr, Soy Freelancer, Upwork, Microworkers, and PeoplePerHour, with four of the fifteen rated platforms failing to meet any threshold under any principle. The cloudwork strand has supplied the framing's load-bearing evidence base for the global microwork-and-freelance-platform layer of the platform economy, including the AI-data-annotation work concentrated on platforms like Scale, Remotasks, and Appen.

The Fairwork for AI workstream, running from July 2021 to March 2024 with INRIA funding, extended the framing into the broader AI supply chain. The workstream — led by Mark Graham with Funda Ustek-Spilda as a key researcher — examined the working conditions of data-annotation workers, content moderators, and the hidden human labour powering generative-AI systems, including detailed research on workers at Sama and parallel data-annotation contractors. The project's working aim, in its own framing, was to "determine AI best practice regarding working conditions" by developing "a set of AI fair work principles and operationalizable processes through which they can be applied, measured, and evaluated" — a translation of the location-based ratings methodology into the AI-supply-chain register that connects the Fairwork framing to the African Content Moderators Union's organising work and to the broader msg-modern-slavery-content-moderation framing.

Policy uptake: ILO, EU Directive, national legislation

The framing's most consequential supranational anchor is its uptake by the International Labour Organization's algorithmic-management workstream, in which Fairwork's on-the-ground evidence base across 38 countries has supplied the empirical foundation for ILO Convention-and-Recommendation drafting toward the 2026 International Labour Conference. Fairwork's audit data on minimum-wage compliance, occupational-safety provisions, contract transparency, algorithmic-management oversight, and freedom-of-association recognition has been a load-bearing evidence input into the ILO's working position on platform work — including the "human-in-command" policy line that ensures "final decisions affecting work are taken by human beings".

At EU level, the framing's policy engagement runs through the Fairwork project's formal response to the European Commission's proposal for a Directive on Platform Work, submitted during the legislative drafting process that produced Directive (EU) 2024/2831 of 23 October 2024. The directive's algorithmic-management chapter — which imposes transparency requirements on automated decision-making in platform work, requires human oversight of significant decisions, prohibits the processing of emotional and psychological data, and equips workers with the right to contest automated decisions — is the framing's clearest single legislative translation, alongside the parallel translation of the msg-algorithmic-management framing that the directive's working register also carries.

National-level legislative uptake has carried through the Fairwork country teams' direct engagement with national policymakers. The Fairwork India Ratings 2024 — the project's sixth consecutive annual India scorecard, covering Amazon Flex, bigbasket, BluSmart, Flipkart, Ola, Porter, Swiggy, Uber, Urban Company, Zepto, and Zomato — has been cited in Indian state-level platform-worker legislation including the Rajasthan Platform Based Gig Workers (Registration and Welfare) Act and parallel Karnataka and Telangana platform-worker-rights bills. The 2025 Fairwork US Ratings — published under the title When AI Eats the Manager — translated the framing's algorithmic-management substantive analysis into the US platform-economy register where the msg-pay-up framing's worker-authored organising had produced the named pay-floor laws in Seattle, New York City, Minnesota, Massachusetts, and California that the report's scorecard treats as the policy-baseline against which platform conduct is assessed.

SCOPE-EDGE: framing-authored-at-academic-edge

Fairwork sits at a structural edge of the corpus's grassroots-and-democratic scope. Its principal carriage runs through the Oxford Internet Institute and the WZB Berlin Social Science Center — academic institutions in the platform-economy research register — and its funder coalition draws on public-research and philanthropic funders (the European Research Council, the UK Economic and Social Research Council, the German Federal Ministry for Economic Cooperation and Development, the Ford Foundation, the John Fell Fund, the Minderoo-Oxford Challenge Fund in AI Governance, the Global Partnership on Artificial Intelligence, and IDRC's FoWIGS initiative) rather than from the grassroots-organising or trade-union register that anchors the corpus's worker-authored framings.

Unlike the worker-authored register of msg-pay-up — launched from inside Working Washington's Instacart-shopper organising vehicle with Mia Kelly as a load-bearing worker voice and Vanessa Bain's parallel Gig Workers Collective organising — Fairwork was coined inside academic platform-economy research at the Oxford Internet Institute and travels primarily through researcher-NGO-policy registers. It parallels msg-bossware in its civil-society-and-policy carriage, but operates one degree closer to grassroots organising via its worker-engaged research methodology: the worker surveys and interviews that produce each scorecard's evidence base are direct engagements with the platform workers whose conditions the audit assesses, not desk-research-and-policy-advocacy operations conducted from outside the labour-economy the framing names.

The framing's in-scope status for this corpus rests on three load-bearing properties. First, the worker-engaged research methodology — surveys and interviews with platform workers in each rated country, recruited through on-platform and snowball-sampling techniques — keeps the framing's evidence base grounded in workers' first-person reports of their working conditions, distinguishing it from purely desk-research-based platform-economy critique. Second, the framing has been adopted by worker-organising vehicles and unions as a citable evidence base in pressure campaigns — Fairwork scorecards have been cited by the App Drivers and Couriers Union (ADCU) in Uber-side litigation, by gig-worker unions in India in state-platform-worker-rights drafting, by domestic-worker organisations in South Africa in Fairwork Cleantech campaigning, and by platform-labour journalists across multiple jurisdictions as the working comparative reference for platform-worker conditions. Third, the framing has produced the documented 300-plus pro-worker policy changes that its pressure-and-praise methodology targets — substantive changes in pay floors, contract transparency, grievance procedures, anti-discrimination protections, and recognition of worker associations on rated platforms — which are the framing's load-bearing case that the academic-edge methodology produces material worker-side outcomes rather than purely scholarly description.

Why it has carried

Three features have made the framing durable.

First, the principle architecture is portable. Five principles, each scored against two thresholds, with a maximum of ten points per platform: the architecture survives translation across location-based work, cloudwork, AI-supply-chain work, and the prospective sectoral expansions the project has announced. The portability has carried the framing across more than 700 platform audits in 38 countries without requiring a fundamental redesign of the rating instrument — the same five principles operate across ride-hail in Lagos, food delivery in Bogotá, microwork on Amazon Mechanical Turk, and data annotation on Scale/Remotasks. That portability is the framing's principal feature for cross-jurisdictional comparison and for the supranational policy register the ILO and EU institutions occupy.

Second, the worker-engaged methodology — surveys and interviews with platform workers as the load-bearing evidence base for each scorecard — keeps the framing's empirical claims accountable to the workers whose conditions it audits. Where academic-research framings can drift into normative claims unmoored from the labour-economy they describe, and where civil-society-advocacy framings can substitute desk research and policy-position drafting for direct worker engagement, Fairwork's three-method audit requires direct contact with the worker side of every rated platform. The methodology is the framing's structural answer to the academic-edge SCOPE-EDGE problem: by binding the audit instrument to worker-survey evidence, the project commits to an evidence base that grassroots organising vehicles and unions can cite, contest, or build on, rather than to one that operates entirely outside their reach.

Third, the framing arrived with an explicit pressure-and-praise theory of change and has produced documented platform-side behaviour change against that theory. The methodological paper that the project's principals published in 2025 — Pressure and praise as an action research methodology — codified the working theory that scorecard publication produces both reputational pressure on low-scoring platforms and competitive recognition for high-scoring ones, and that the implementation window preceding publication opens a substantive dialogue through which platforms negotiate labour-conditions changes. The project's documented 300-plus pro-worker policy changes are the framing's load-bearing case that this theory of change produces substantive outcomes rather than purely reputational ones — a result that distinguishes Fairwork from rating-and-scoring methodologies in other domains that have produced reputational shifts without corresponding material change in the rated parties' substantive practices.