A gig economy algorithm found discriminatory
Forum: Tribunale Ordinario di Bologna
This case concerned a scoring algorithm that was used by gig economy platform Deliveroo in Italy.
Deliveroo is a well-known platform that offers food delivery jobs to a network of riders, who normally carry out the delivery jobs on bicycles or mopeds.
At the time the case was brought, Deliveroo riders in Italy were able to book shifts in advance via an app, meaning that they could indicate what time periods they would be available to accept jobs.
These shifts would be made available to riders to book in a staggered way, prioritising those who were scored by an algorithm as being most “reliable” and “participatory.” Those with low “reliability” and “participation” scores would have less available shifts, and so less job opportunities, than drivers who scored higher.
Riders found that their “reliability” scores were negatively affected by their failure to “turn up” and “log in” for pre-booked sessions.
A group of trade union bodies challenged this before the Italian courts, arguing that such a system disproportionately discriminated against trade union members who might exercise lawful and legitimate abstention from their work.
The Ordinary Court in Bologna agreed, ruling that the platform’s system for setting conditions on access to work amounted to indirect discrimination. In doing so, the court reiterated that (in Italy) gig economy workers received the same legal protections from discrimination as employees.
The case was taken by three affiliate bodies of the Italian General Confederation of Labour, Italy’s largest trade union. These affiliates were working with gig economy workers across a range of sectors, including logistics, transport and catering. A number of their members were Deliveroo riders, who would carry out delivery services in urban areas by bicycle or moped.
Deliveroo was a global company that provided home delivery of food through a digital platform and a network of riders. The riders who carried out work through the platform had to download the Deliveroo app on their smartphone. This app generated a personal profile that allowed riders to access the platform and accept delivery jobs. Deliveroo treated riders as being self-employed rather than employed by Deliveroo.
Deliveroo riders had a choice of two different ways to take on delivery jobs on the platform.
For the first method, a rider could log on to the app to accept job proposals in real time. The second method was through the “self-serve booking system.” This allowed riders to book work sessions in advance for certain locations through a calendar of “time slots” on the app.
Work sessions were periods of time in which a rider intended to accept job proposals. There was no obligation on a rider to accept a specific job proposal during a work session. Nevertheless, there were incentives for riders to turn up for their selected work sessions.
The booking calendar on the “self-serve booking system” was not fully available to all riders at the same time. Instead, the calendar of “time slots” for a given week would be progressively made available to groups of riders. On the day when “time slots” were made available, some riders could start booking slots at 11:00 while others had to wait until 15:00 or 17:00 to book.
Those who had to wait until the afternoon to book work sessions noticed that they had less available “time slots” available to them than those who could book at 11:00.
It emerged that the priority treatment of a rider to book “time slots” was determined on the basis of a score that was assigned to them through an algorithm. This score consisted of indicators on:
“Reliability:” this was calculated from the number of occasions a rider failed to log in for a work session within fifteen minutes of its start time. For work sessions, riders were expected to cancel more than 24 hours before the sessions that they could not make.
“Participation:” this was calculated from how available a rider was for the most popular times for Deliveroo orders (from 20:00 to 22:00 from Friday to Sunday).
This meant that a “reliability” score would be negatively impacted if a rider failed to “turn up” to a working session that they did not cancel over 24 hours before that working session started.
This meant that, in order to avoid a negative impact on their score, riders who did not cancel more than 24 hours in advance of a session would have to drive to the relevant area for the work session and log in to the app within fifteen minutes of the session start time.
The trade union affiliates taking the case alleged that Deliveroo’s use of this system was discriminatory because it penalised riders for legally protected abstentions from work. For example, riders could be penalised for not turning up for sessions because they were taking collective action (i.e. a strike) or because of illness or family needs.
Evidence was presented to the court, through a witness statement, on the fact that there were difficulties in calling for strikes in the sector as riders were reluctant to participate in such forms of protest because it undermined their future job opportunities.
While the case was pending before the Italian courts, Deliveroo ceased using the “self-serve booking system” and released a statement saying that statistics would no longer have an impact on rider’s accessibility to the booking calendar on the system.
Directive 2000/78/EC of 27 November 2000 establishing a general framework for equal treatment in employment and occupation
This EU law set to prohibit discrimination across the region within certain sectors, including in relation to “conditions for access to employment, to self-employment or to occupation.”
The court in Bologna first clarified, with reference to legislation and case law, that labour law protecting employees from discrimination also extended to self-employed gig economy workers.
That meant that gig economy workers were protected under the law from all forms of discrimination in their access to work through online platforms. Meaning that gig economy workers should have equal access to work through these platforms regardless of their religion, personal convictions, disability, age and sexual orientation (i.e. their protected characteristics). The court went on to reiterate that “personal convictions” included trade union membership.
The court then went on to note the difference between direct and indirect discrimination.
With direct discrimination, the actual act of treating a person less favourably on the basis of a protected characteristic (i.e. religion, personal convictions, disability, age or sexual orientation) would be the prohibited discrimination.
With indirect discrimination, a general measure or policy, even when formulated in neutral terms, that had a disproportionately detrimental effect or impact on a particular group would amount to prohibited discrimination.
The court also reiterated that the will or intention of a party, such as an employer, was irrelevant to the question of whether discrimination had occurred.
The court reasoned that a specific victim of discrimination did not have to be identifiable. Instead, the party claiming that discrimination had occurred could provide facts, such as facts of a statistical nature, from which discrimination could be inferred. In other words, the party taking the case only had to show statistical data that indicated that groups of workers were being disadvantaged. Once that had been shown, it was for the other side to show that their choices were actually based on objective and non-discriminatory factors.
Having reviewed the facts presented to it, the court reasoned that there was adequate proof that taking part in a trade union’s activities involving abstention from work, such as a strike, was capable of compromising a Deliveroo rider’s statistical scores in this case. On the “reliability” score, the rider would suffer a decrease for not participating in a work session that they had booked. On the “participation” score, the rider would also be penalised if the relevant work session was between 20:00 and 22:00 Friday through Sunday.
Deliveroo had argued that a rider who wanted to abstain from work could still log in and simply not accept a job without being penalised. The court reasoned that this was of little relevance since riders were still expected to travel to a specific geographical area, which the app tracked through geolocation, where they had to log in for the work session. Therefore, to avoid penalisation, a rider who wanted to strike in such circumstances would have to “show up at the workplace.” The court reasoned that this seemed incompatible with the exercise of the right to strike, which usually involved the total abstention from work.
Late cancellation & sudden strike action
The parties to the case were in disagreement about whether cancelling a work session within 24 hours of the start time would negatively impact a rider’s score.
When considering this, the court was critical of the fact that it did not have access to more information on the workings of the statistical model. It stated that the failure of Deliveroo to disclose the operating rules of the algorithm basically precluded a more in-depth examination of this issue. The court also observed that Deliveroo had not provided information on the specific calculation criteria adopted to determine the statistics of each rider, and they were not made available on the website of the company.
Nonetheless, with reference to information disclosed by Deliveroo’s witness and its model contract with riders, the court was able to determine that a “late cancellation” had the potential to negatively impact a rider’s score.
Therefore, a rider who wanted to join a strike action, and cancelled within the 24-hour window before a work session, would be penalised by potentially losing priority access to “time slots.” The court reiterated that sudden strike action, without notice, was a recognised form of collective action. If a rider had to cancel a work session over 24 hours before it started in order to attend a strike action, this would put Deliveroo on notice to replace the striking rider. This would minimise economic damage to the company, which was part of the purpose of this form of collective action.
The court observed that Deliveroo did not want to know the reasons why a rider did not turn up for a work session or had to make a “late cancellation.”
The court reasoned that this “blindness” gave effect to the system’s discriminatory potential. It observed that, by treating in the same way those who did not participate in a session for futile reasons and those who did not participate because they were on strike, Deliveroo had discriminated against the latter in practice. Potentially marginalising these groups from the priority group for access to “time slots” and reducing their opportunities for access to work.
The court concluded with critical remarks on the fact that, although capable of doing it, Deliveroo would not take corrective action with regard to a rider’s score where the reason for a rider’s failure to work was protected by law.
The court, therefore, found the “self-serve booking system” to have amounted to indirect discrimination since it applied an apparently neutral rule which placed a certain category of workers (those participating in trade union activities) in a position of potential and particular disadvantage.
It was left for the court to consider whether this indirect discrimination could be justified because it pursued a legitimate purpose and was necessary and proportionate.
Deliveroo was unable to convince the court.
The court stated that the discrimination could not be justified because the system encouraged riders to cancel bookings so other riders could book them instead. The court highlighted that its decision rested on the fact that a rider’s non-participation could not be justified with reference to a legally protected right (i.e. the right to strike), rather than on the fact that riders were encouraged to cancel in advance the sessions that were not going to be worked.
The court declared that the conduct of Deliveroo in this case was discriminatory in how it set the conditions of access to the booking of work sessions through their platform.
Deliveroo was ordered to publish an extract of the decision on the “frequently asked questions” part of its website and was ordered to pay for an extract of the decision to be published in a national newspaper.
They were also ordered to pay €50,000 in compensation to riders.