Between control and participation: The politics of algorithmic management

TL;DR

This qualitative case study reconstructs the politics of algorithmic management in a German logistics company, distinguishing goal formation, data production, and data analysis phases and showing that multiple actors’ knowledge shapes outcomes. Despite a process-oriented, non-surveillance strategy shaped by works council negotiation, the system centralised knowledge and excluded workers from optimisation.

Key findings

• The paper develops a three-phase process model—goal formation, data production, and data analysis/usage—in which actor constellations and knowledge contributions change by phase: central management, AnalyticsTech, local management, and works council in goal formation; process engineers, data scientists, local management, and workers in data production; process engineers and data scientists with local management in data analysis.
• In this German case, management adopted a process-oriented strategy rather than a workforce-oriented surveillance strategy: a works council agreement ensured voluntary participation, anonymous data, no individual performance monitoring, no measurements during breaks, and no staff cuts; but the outcome was still a centralisation of knowledge and disempowerment of workers, who were not involved in optimisation and referred to themselves as “lab rats.”
• Data production was crucial and labour-intensive: measurements lasted two weeks; sensors generated approximately 3000–4000 data points per second and sensor; workers’ domain knowledge was an important power resource, and workers sometimes undermined measurement by not wearing wearables, removing sensors, or working slower. Process engineers and local management were powerful due to domain knowledge, while data analysis became a centralisation phase led by process engineers, with data scientists in a service role.
• The study finds narrow limits to automating management decisions in conventional firms compared with platforms: simple analyses can be automated, but identifying problems and developing optimisation concepts require domain experts; LogisticsCorp initially depended on AnalyticsTech, then licensed the software and internalised analysis, showing decentring and recentring are both possible.

Methods

• In-depth qualitative case study of an algorithmic management system for process optimisation at LogisticsCorp, a German logistics company with more than 10,000 employees, using software from AnalyticsTech, a software developer with around 50 employees; the case is embedded in the German regulatory context.
• Data collection included 16 semi-structured interviews of one to two hours with AnalyticsTech management and LogisticsCorp central/local management, process engineers, and data scientists; four additional interviews with managers from other companies using AnalyticsTech software; a two-day field observation of implementation; participation in two AnalyticsTech workshops; and document analysis of workshop materials, information documents, process maps, protocols, and analysis outputs.
• Analysis used qualitative content analysis in MAXQDA with inductive and deductive coding, four thematic main categories, and a respondent-validation workshop with company actors.

Limitations

• As stated by the authors, the methodological characteristics of an in-depth case study mean they cannot draw conclusions about how common the described constellation is in the overall population of companies; this is left for future research.
• Evident from the text: the study is a single case involving two companies and one advanced implementation, limiting generalisability; workers were not formally interviewed but only spoken with informally and observed, so the worker perspective is less fully captured than management, engineer, and data scientist perspectives.