Authored By: Zulaikha Khan
STADIO Higher Education
Introduction
A 2017 decision by a CCMA commissioner brought to light an unusual question of status. In what she termed the ‘realities of the relationship’ test, the commissioner ruled that certain Uber drivers were in fact employees of Uber South Africa Technology Services (Pty) Ltd.[1] She was moved by the fact that while no one was ever there to supervise the drivers in person, control was nonetheless exerted via technology. As she put it, the system could even pick up on the movement of a cell phone as a sign of reckless driving.[2]
The Labour Court rejected this finding and set aside her ruling within the year. The court’s objection was not that the drivers should be seen as independent contractors; rather, it was because the commissioner had made the error of conflating Uber SA with its Dutch holding company, Uber BV, which is the true source of control over the drivers.[3] Consequently, the merits of whether such algorithmic direction amounts to control in the eyes of South African employment law remain an open question.
It is a gap worth noting. For the most part, South African law relies on the dominant impression test to establish employment status, and at the centre of that is the matter of control.[4][5] Yet this article will make the case that, in practice, the courts have treated control as the domain of a human supervisor and have not been willing to acknowledge algorithmic control, for all that the CCMA has found otherwise. We are concerned here only with the factor of control and do not intend to wade into the organisational or economic-dependence issues that feature in other case law, nor the question of whether Uber BV might be the employer.
What follows is divided into five sections. After Section II lays out the legal framework and Section III reviews the relevant case law, Section IV provides a critical look at the control factor. Section V considers how things are being handled in the United Kingdom and European Union before Section VI puts forward a way ahead.
The Legal Framework
The Dominant Impression Test
In South Africa, the parties’ own description of their arrangement does not dictate its legal character; the courts will see past the label to the substance of the work.[6] An ’employee’ is defined in section 213 of the Labour Relations Act 66 of 1995 as any person who is not an independent contractor and who works for another for remuneration.[7] When it comes to drawing the line between the two, one looks to the Labour Appeal Court’s judgment in South African Broadcasting Corporation v McKenzie. There, the court points to such features as the worker’s subservience to the employer’s will and the obligation to follow lawful directions.[8]
But the McKenzie judgment, in a passage from Mureinik that it endorses, is careful to note that the dominant impression test can be of little help in the more intractable cases.[9] It is a circular proposition: an employment contract is one that has the look of employment, but the test itself does not say what that likeness is.
The Statutory Presumption
A rebuttable presumption of employment was put in place for those whose earnings fall below the Minister’s threshold by way of section 200A of the LRA and a like provision in section 83A of the Basic Conditions of Employment Act 75 of 1997 (BCEA). This has been the case since 2002.[10] One need only have any of a number of enumerated factors to invoke the presumption; among them is whether another party dictates the hours or manner of one’s work, or if the person is part of the organisation for which the work is performed.[11] Should the other side wish to counter this, they are at liberty to put forward evidence as to what the relationship really is.
The Code of Good Practice: Who is an Employee?, promulgated under section 200A(4), goes into some detail on these matters. It would have a decision-maker form a ‘dominant impression’ from an appraisal of the realities between the parties and all the relevant factors that come to light.[12] What is notable is that neither the Code nor the statute requires this control to be exerted by a human being in real time. In the end, the test is functional, not formal.
A Broader Statutory Definition
One finds a different tack with the National Minimum Wage Act 9 of 2018. Here, a ‘worker’ is defined in plain terms as anyone who does work for another and is due payment for it.[13] There is no room for the control-oriented language of section 200A and, on its face, independent contractors are not left out. It is a much wider definition than the LRA has for an ’employee’, and it produces a certain asymmetry. A platform worker could be considered a ‘worker’ for the purposes of the Minimum Wage Act yet not fit the LRA’s more restrictive mould, given that the courts have yet to apply their notion of control to platform work. How much weight can be placed on that asymmetry is a matter for the next section, depending on the practical approach the courts have taken to the question of control.
III. Case Law Analysis
In the matter of algorithmic control, the CCMA’s 2017 award in the Uber case is still the most direct example from South Africa. The commissioner put aside the statutory presumption of section 200A as inapplicable to the drivers and based her conclusion on the common-law ‘realities of the relationship’ test (see Section II.A) rather than the statutory factors in Section II.B.[14] While she acknowledged that the drivers were free to set their own hours and could be at liberty to accept or even ignore a trip request, she was of the view that Uber dictated how they worked by imposing performance requirements and standards. ‘Even though there is no direct or physical supervision, control is exercised through technology,’ she noted, pointing to the deactivation policy.[15] She saw the platform’s ability to suspend app access and cut off a driver’s income as tantamount to conventional supervisory control, with no need for a human to give real-time instructions.
The merits of that finding were never put to the test on appeal. In 2018 the Labour Court set it aside, but for reasons unconnected to the issue of control.[16] Van Niekerk J ruled that the commissioner had not made the necessary distinction between Uber SA, the respondent in the matter, and Uber BV, the entity that in fact contracted with the drivers, set fares and ran the deactivation policy. Since the drivers had conceded any contractual link with Uber SA, the court held this was fatal to their claim on binding authority.[17] The court was clear it was not ruling on whether the drivers were employees of Uber BV in any capacity.[18] Consequently, the commissioner’s take on algorithmic control was left neither approved nor disapproved; it was rendered moot by a disagreement over the proper respondent.
For a more authoritative judicial view on the control factor one has to look to McKenzie, although the circumstances of that case are markedly different from platform work. There the court dealt with a radio presenter on fixed-term contracts who had some say in content and was under a form of supervision that had become ‘diluted’ with time as trust was built with the broadcaster.[19] A like analysis was applied in Goliath v South African Broadcasting Corporation to a music show presenter with an annually renewed contract; the court used the same indicia to find he was an independent contractor.[20] What these cases demonstrate is that when South African courts have come to grips with the control factor, they have done so by looking at instruction and supervision between two people, where one can direct and the other may be persuaded to let up. It is a model that the CCMA’s 2017 decision implies is ill-suited to the realities of platform work.
Critical Evaluation
One might be tempted to think the case law reveals a deficiency in South African law, as if it were without the concepts to deal with algorithmic control. The CCMA’s 2017 finding disproves this.[21] There a commissioner readily addressed the issue, using the ‘realities of the relationship’ framework as it stands and determining that automated deactivation and monitoring are, in substance, forms of control. The difficulty is that her conclusion has no precedential value. On jurisdictional grounds wholly unrelated to the merit of her reasoning, the finding was set aside and to date no South African court has seen fit to re-examine the matter. As for McKenzie, the only appellate authority on the control factor that is binding, it was decided in the context of old-fashioned human supervision and gives one no indication as to whether its logic holds for the kind of impersonal, continuous mechanisms at play here.
The position in the United Kingdom serves to put this in perspective. In Uber BV v Aslam the Supreme Court had little trouble in labelling the ratings-based management by Uber as ‘a classic form of subordination’ typical of an employment relationship.[22] It noted how the company fixed fares and contractual terms and policed acceptance and cancellation rates through what it termed an ‘escalating series of warning messages’ that could end in automatic exclusion from the app. Whether the subordination came from a manager or software made no difference to the Court’s view of worker status. Adams-Prassl comes to much the same end in his analysis of the judgment, noting that while the technology may alter the means of control, it does not change its legal import.[23] A functional approach is also evident in the European Commission’s proposal for the Platform Work Directive, which sees electronic supervision as an indicator of control in its own right, equivalent to the underlying phenomenon regardless of remuneration or conduct rules.[24]
There is ample support for such a reading in the literature on platform work. Wood, Graham, Lehdonvirta and Hjorth, in their comparative study of gig workers, make the case that algorithmic control is genuine workplace control; they contend that even with nominal flexibility over hours, systems like automated task allocation can be highly constraining.[25] Newaj offers a different perspective in her review of Uber litigation across South Africa, the UK and New Zealand. She points to the South African Labour Court’s failure to come to the aid of drivers, despite the CCMA having already found them to be employees, as proof of a wider doctrinal weakness in the way the law classifies platform workers.[26] The argument here would refine that: the test is not unworkable, but the one time it yielded a workable result, it was not allowed to stand on appeal. Masikane and Webster add to this in their account of South African platform capitalism, suggesting the flexibility on offer often masks a structural dependence that traditional labour categories were never designed to encompass.[27]
In many ways the facts in the two jurisdictions are not so far apart; the commissioner’s 2017 account of technological control fits well with the reasoning in Aslam. South African law is not hampered by any textual problem in section 200A or the common-law test. What is missing is binding authority to apply the control factor to the realities of the platform. An operator of a platform would have cause to object to this interpretation. For one, there is the matter of autonomy: the commissioner conceded drivers were free to pick and choose their hours and ride requests.[28] For another, the arrangement looks more entrepreneurial in nature. Drivers put up their own vehicles and running costs and are free to sign up with competitors, all hallmarks of independent contracting. These are objections with some substance and they should not be dismissed lightly. The commissioner’s own findings, however, tell a different story. In deactivating drivers for low acceptance or cancellation rates, the policy exacted a real behavioural cost from what was otherwise a formal right to say no.[29] Wood and his colleagues put it in broader terms: the sort of nominal autonomy that is hemmed in by algorithmic constraint is typical of platform work as a whole, not some quirk of Uber’s way of doing things.[30] The objection, then, speaks to a structural feature of the sector and hardly a case for denying it protection.
One comes to much the same conclusion from the empirical work of Fredman and her co-authors on the industry. They note that with their automated systems for monitoring and making decisions, platforms are able to obscure the control they have over workers and rework the terms of the relationship to sidestep any indicia of employment.[31] There is a certain combination at play here, where flexibility is exercised at a financial or reputational price, and the LRA’s control factor in its present judicial incarnation does not account for it. The test is concerned with whether control is there; it has yet to consider how it might be structured or put on display. One might do well to see if other jurisdictions have managed to pose that question.
Comparative Perspectives
In the Aslam case, the Supreme Court put an end to the question of algorithmic control by way of interpretation, eschewing any need for legislative amendment.[32] The Court’s ruling was clear: whether a person is a worker as defined in section 230(3)(b) of the Employment Rights Act 1996 is a matter of statutory interpretation, not of how one chooses to label the relationship in a contract. Uber’s de facto control over its workers, the algorithmic elements of which were taken into account, was sufficient in itself to confer worker status. In doing so, Aslam has made it plain that a court can reinterpret an existing test based on control; South African judges could well follow suit and need not stand by for Parliament to act.
The European Union has seen fit to go down the road of legislative reform. With Directive (EU) 2024/2831 in force since 1 December 2024 and member states having until 2 December 2026 to transpose it, there is now a requirement to put in place a rebuttable presumption of employment where the facts point to control or direction by a digital platform.[33] The Directive is also prescriptive when it comes to algorithmic management, mandating human oversight of automated systems and ensuring that no human being is left out of the decision to terminate or suspend a worker’s account.[34] One does not have to be a judge to make the link between an algorithm and traditional control under the Directive; the statute spells it out. Moreover, it goes further than the UK approach by regulating the operation of such management after an employee has been classified.
South Africa’s own framework has the makings of either course of action. There is nothing in McKenzie or the Code of Good Practice to suggest the common-law notion of control must be confined to real-time, human supervision, leaving the door open for the sort of judicial reinterpretation we saw in Aslam.[35] Alternatively, one could look to the EU model and amend the rebuttable presumption in section 200A to treat algorithmic monitoring as evidence of control. It would be a more measured step than the EU’s broad reforms, but it would serve to close the gap all the same.[36]
Conclusion
The position taken in this article is that the way South African law deals with algorithmic control is not a doctrinal vacuum so much as an unproven possibility. The CCMA made its determination in 2017 that automated deactivation and monitoring were, in substance, forms of control. Yet the merits of that finding were never put to the test; it was put aside on a jurisdictional point entirely unrelated to the kind of control at issue. As for SABC v McKenzie, the sole binding appellate authority on the matter of control has yet to be brought to bear on platform work, offering no grounds to suppose its logic would or would not apply to algorithms.
One need only look at the UK’s handling of Aslam to see how such a gap can be bridged by interpretation if the courts are prepared to view algorithmic mechanisms as the functional equivalent of conventional supervision, without recourse to new laws. The EU has chosen a different course with its Platform Work Directive, a legislative option that makes the equivalence plain by means of a statutory presumption and places continuing duties on algorithmic management once an employment relationship is in place. A proper reading of the materials at hand in South Africa would allow for either approach.
There are three ways to put an end to the uncertainty. For one, the courts should make use of the next occasion to put the CCMA’s 2017 reasoning on record as sound application of the control factor, instead of letting it stand as an orphaned award. The legislature ought to revise section 200A to put algorithmic monitoring on the books as an indicator of control and insist on some form of documented human oversight before an automated deactivation is allowed to determine a driver’s livelihood. Absent these measures, gig workers in South Africa are left to a test designed for human supervisors, while the one instance where algorithmic control was named for what it is remains unreviewed and apart from the law it ought to have influenced.
Reference(S):
Table of Cases
Goliath v South African Broadcasting Corporation SOC Ltd and Others (C581/2019) [2022] ZALCCT 10; (2023) 44 ILJ 185 (LC)
South African Broadcasting Corporation v McKenzie (CA8/98) [1998] ZALAC 13; (1999) 20 ILJ 585 (LAC)
Uber BV and Others v Aslam and Others [2021] UKSC 5
Uber South Africa Technology Services (Pty) Ltd v NUPSAW and SATAWU obo Morekure and Others [2017] ZACCMA 1; [2017] JOL 55085 (CCMA)
Uber South Africa Technology Services (Pty) Ltd v National Union of Public Service and Allied Workers (NUPSAW) and Others (C449/17) [2018] ZALCCT 1; (2018) 39 ILJ 903 (LC)
Table of Legislation
Basic Conditions of Employment Act 75 of 1997, s 83A
Code of Good Practice: Who is an Employee? GN 1774, GG 29445 (1 December 2006)
Directive (EU) 2024/2831 of the European Parliament and of the Council of 23 October 2024 on improving working conditions in platform work
Employment Rights Act 1996 (UK), s 230(3)(b)
European Commission, Proposal for a Directive of the European Parliament and of the Council on improving working conditions in platform work, COM(2021) 762 final (9 December 2021)
Labour Relations Act 66 of 1995, ss 200A, 213
National Minimum Wage Act 9 of 2018, s 1
Bibliography
Adams-Prassl J, ‘Uber BV v Aslam: “[W]ork relations … cannot safely be left to contractual regulation”‘ (2022) 51(4) Industrial Law Journal 955
Fredman S, Du Toit D, Bertolini A, Valente J and Graham M, ‘Fair Work for Platform Workers: Lessons from the EU Directive and Beyond’ (2025) 54(3) Industrial Law Journal 425
Masikane F and Webster E, ‘Workers’ Power and Platform Capitalism: The Embryo Towards an Alternative’ (2025) 30(3) New Political Economy 325
Newaj K, ‘Re-Evaluating the Employment Status of Uber Drivers in South Africa: Lessons from the United Kingdom and New Zealand’ (2023) 55(2 & 3) Comparative and International Law Journal of Southern Africa
Wood AJ, Graham M, Lehdonvirta V and Hjorth I, ‘Good Gig, Bad Gig: Autonomy and Algorithmic Control in the Global Gig Economy’ (2019) 33(1) Work, Employment and Society 56
[1]Uber South Africa Technology Services (Pty) Ltd v NUPSAW and SATAWU obo Morekure and Others [2017] ZACCMA 1; [2017] JOL 55085 (CCMA) para 52.
[2]NUPSAW obo Morekure (n 1) para 45.
[3]Uber South Africa Technology Services (Pty) Ltd v NUPSAW [2018] ZALCCT 1 (LC) paras 79-97.
[4]South African Broadcasting Corporation v McKenzie (1999) 20 ILJ 585 (LAC) 590E-F.
[5]Uber South Africa Technology Services (Pty) Ltd v NUPSAW (n 3) para 73.
[6]ibid para 75.
[7]Labour Relations Act 66 of 1995, s 213.
[8]McKenzie (n 4).
[9]McKenzie (n 4) 589, citing Mureinik with approval.
[10]Labour Relations Act 66 of 1995, s 200A(1); Basic Conditions of Employment Act 75 of 1997, s 83A.
[11]Labour Relations Act 66 of 1995, s 200A(1)(a)-(b); see also Code of Good Practice: Who is an Employee? GN 1774, GG 29445 (1 December 2006).
[12]Code of Good Practice: Who is an Employee? (n 11).
[13]National Minimum Wage Act 9 of 2018, s 1.
[14]NUPSAW obo Morekure (n 1) para 40.
[15]NUPSAW obo Morekure (n 1) paras 45-46.
[16]Uber South Africa Technology Services (Pty) Ltd v NUPSAW (n 3).
[17]ibid para 97.
[18]ibid para 98.
[19]McKenzie (n 4) para 34.
[20]Goliath v South African Broadcasting Corporation SOC Ltd and Others [2022] ZALCCT 10; (2023) 44 ILJ 185 (LC) para 25.
[21]NUPSAW obo Morekure (n 1) paras 45-46.
[22]Uber BV and Others v Aslam and Others [2021] UKSC 5 [93]-[101], especially [97] and [99].
[23]Jeremias Adams-Prassl, ‘Uber BV v Aslam: “[W]ork relations … cannot safely be left to contractual regulation”‘ (2022) 51(4) Industrial Law Journal 955.
[24]European Commission, Proposal for a Directive of the European Parliament and of the Council on improving working conditions in platform work, COM(2021) 762 final (9 December 2021).
[25]Alex J Wood, Mark Graham, Vili Lehdonvirta and Isis Hjorth, ‘Good Gig, Bad Gig: Autonomy and Algorithmic Control in the Global Gig Economy’ (2019) 33(1) Work, Employment and Society 56.
[26]Kamalesh Newaj, ‘Re-Evaluating the Employment Status of Uber Drivers in South Africa: Lessons from the United Kingdom and New Zealand’ (2023) 55(2 & 3) Comparative and International Law Journal of Southern Africa (online-first publication, unpaginated).
[27]Fikile Masikane and Edward Webster, ‘Workers’ Power and Platform Capitalism: The Embryo Towards an Alternative’ (2025) 30(3) New Political Economy 325.
[28]NUPSAW obo Morekure (n 1) para 45.
[29]NUPSAW obo Morekure (n 1) para 39.
[30]Wood, Graham, Lehdonvirta and Hjorth (n 25).
[31]Sandra Fredman, Darcy Du Toit, Alessio Bertolini, Jonas Valente and Mark Graham, ‘Fair Work for Platform Workers: Lessons from the EU Directive and Beyond’ (2025) 54(3) Industrial Law Journal 425, 426.
[32]Uber BV v Aslam (n 22) [69]-[76].
[33]Directive (EU) 2024/2831 of the European Parliament and of the Council of 23 October 2024 on improving working conditions in platform work, art 5.
[34]ibid, arts 9(1),(5), 10(5) and 11(1)-(3) (obligations concerning transparency, information, and human oversight of automated monitoring and decision-making systems, including the requirement that any decision to restrict, suspend, or terminate a worker’s account be taken by a human being).
[35]Uber BV v Aslam (n 22) [69]-[71].
[36]Directive (EU) 2024/2831 (n 33).
