Product1 publisher3 min readPublished
US firms sanction poor performance by software at nearly 17 times the European rate
An OECD survey of 6,047 employers found this software in most workplaces on both continents. The countries split over what firms are allowed to do with it, and most of these tools are not AI at all.
The Product Desk · Product desk

What happened
- The OECD surveyed 6,047 firms in six countries on whether they use software to instruct, monitor or evaluate workers, with Ipsos interviewing mid-level managers at workplaces of 20 or more in mid-2024.
- Adoption is high nearly everywhere: 90 percent of US firms, 81 percent in France, 78 percent in Germany and Spain, 76 percent in Italy and 40 percent in Japan.
- Software that sanctions poor performance is used by 67 percent of US firms, 4 percent across the four European countries surveyed and 1 percent in Japan.
- Of the 15 use cases the OECD measured, more than three quarters of American firms run ten or more, and very few run fewer than eight.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Because the scoring sits inside software a firm already licenses, a review triggered by the arrival of an AI feature will not fire at all. The trigger has to be the use case, not the vendor's product label.
- decision A configuration built for US sites cannot be cloned into Italy or Spain untouched. Someone has to go feature by feature and decide which evaluation uses are switched off and which are written up for the works council.
- exposure The penalty on record here came under data protection law, which puts the liability with whoever signed off on the data flows rather than with whoever bought the licence.
- contradiction Managers report 91 percent worker awareness and the OECD declines to believe them. A rollout notice is therefore thin evidence that anyone told the workforce what the software scores.
The screen where this gets decided is a permissions tab in software the company already owns. Managers in the survey named SAP, Workday, Oracle, Jira, Asana, Trello and time-tracking products, and the OECD notes the practice is often not recognised or branded as algorithmic management at all [18].
The divergence widens as the use case moves from scheduling work to judging people. Any evaluation tool at all runs at 90 percent of US firms and 35 percent of European ones, about 2.6 to 1 [5][3]. Sanctioning poor performance runs at 67 against 4, close to 17 to 1 [3][1]. Monitoring the content and tone of conversations, calls or emails, 55 against 6, is about 9 to 1 [6][2]. Speed tracking is 72 against 15 [7]. Leaderboards visible to other staff sit in half of US workplaces and 7 percent of European ones [8]. Rewarding good performance runs 83 against 13 [4]. On adoption, the US and France are nine percentage points apart [2][4].
European firms typically use three to five of the OECD's use cases, and nearly a third of Japanese adopters use exactly one [10]. So a French rota tool and an American stack running conversation analysis, speed tracking and automated sanctions both land in the same adoption column.
The OECD attributes the difference to regulatory architecture, setting the EU's centralised, rights-based approach against a patchwork of agency enforcement and state and local rules in the US [11]. Consultation law does specific work inside that. An EU directive requires information and consultation of worker representatives, while US bargaining obligations attach only to particular subjects [12]. Among US firms that consult at all, almost all managers say they were consulted and about a third say other employees were [13].
GDPR alone does not explain the spread inside Europe. Italy and Spain have markedly more governance measures in place than France and Germany, with audits at 83 and 78 percent [14]. The OECD credits Italy's Transparency Decree of 2022, which explicitly covers automated decision-making, and Spain's 2021 Riders' Law, which gives works councils a right to be informed about algorithmic management [15]. It also puts the Foodinho case behind those audit numbers, in which Glovo's Italian subsidiary was fined 2.6 million euros under data protection law [16].
The OECD says plainly that algorithmic management tools vary in sophistication and are not necessarily AI-powered, and it warns that rules written specifically for AI leave gaps open for tools that do not use it [17].
Ninety-one percent of managers said employees or their representatives are made aware of the software. The OECD does not accept the figure, and suggests most managers are unaware of how little workers know [19].
For whoever has to configure one of these systems across US and European sites, the useful exercise is a two-column list. Put every enabled feature into coordination (rota, ticket assignment) or evaluation (ranking, sanctioning, tone analysis), then, for each item in the second column, name the document that told worker representatives it exists and the date it went out. More than three quarters of US firms run ten or more of the OECD's fifteen use cases and European firms typically run three to five, so a template built in the US arrives in Milan or Madrid with five to seven extra items in the evaluation column [9][10][5].
What to watch
- Whether France and Germany copy Italy's Transparency Decree or Spain's Riders' Law, the two laws the OECD credits for higher audit rates.
- Whether enforcement bodies keep reaching non-AI management tools through data protection law, as in the 2.6 million euro Foodinho fine.
- Whether any worker-side data emerges to test the 91 percent awareness figure managers reported.