Motivation & aim
Despite the existence of international frameworks – such as the UN Sustainable Development Goal 8.7 on ending modern slavery – an estimated 49.6 million people worldwide were affected by forms of modern slavery in 2021 (ILO, 2022). The phenomenon remains difficult to detect due to its inherent opacity, deliberate concealment, and the complexity of actor networks. This severally constraints empirical analyses of modern slavery practices (Caruana et al., 2021) and monitoring and exercising oversight over corporate behavior by civil society actors (Hartmann & Moeller, 2014; Gold & Heikkurinen, 2018).
From both a compliance and a sustainability perspective (Gutierrez‑Huerter O et al., 2023), organizations therefore require a tool that can systematically reveal risks of modern slavery, support the prioritization of preventive and corrective measures, and make actor‑specific expertise effectively usable. Existing tools offer valuable insights but typically operate on high levels of aggregation (e.g., countries or industries). However, companies operate in dynamic, globally interconnected business ecosystems that often transcend national and sectoral boundaries (Hanelt et al., 2021). As a result, there is a growing need for integrated, adaptive risk evaluation tools (Gold et al., 2021).
This research project responds to these limitations by leveraging machine learning methods to provide a learning risk‑analysis instrument capable of addressing this complexity. Machine learning provides new opportunities by enabling the processing of large, heterogeneous datasets, the detection of non‑obvious patterns, and the identification of novel risk indicators beyond traditional measurement methods (Tofangchi et al., 2021; Choudhury et al., 2021). Thus, machine learning offers a promising avenue for addressing existing shortcomings in risk identification and transparency related to modern slavery.
The project seeks to empower civil society actors (as well as other people/organizations that fight against modern slavery, such as journalists) to analyze new cases more effectively and to identify structural risk patterns for informed actions to protect vulnerable workers. It further aims to support academic research by providing structural empirical data for systematic analysis.
Parties involved
Designed specifically as a transdisciplinary research project, it combines expertise in digital transformation (Department of Digital Transformation Management at the University of Kassel) with with that in social sustainability and supply chain management (Chair of Sustainability Managemen at the TU Munich Straubing Campus).
By providing practical support for the active fight against the major societal challenge (‘grand challenge’) of modern slavery, the research project further contributes to the mission and objectives of the Hans Böckler Foundation, which funds the research project.
Process
The research project first developed an expert‑aligned, operationalized scientific definition of modern slavery.
Then, using various databases (e.g., Sherlock), cases of modern slavery were systematically selected and extracted from NGO reports, news articles, court case files, and research papers. Filtering and analysis were supported by large language models and underwent mulitiple modification rounds. Cases further passed through validation procedures conducted by international experts. The dataset covers the period from 1 January 2015 to 31 December 2024.
Finally, building on the database, the research project applied a user‑integrated Design Science Research approach to develop a machine‑learning method, guided by Hevner et al. (2004) and Peffers et al. (2007). The development process included multiple evaluation cycles. Recognizing that the effectiveness of technical solutions depends on their socio‑political embedding, the project further relied on intensive collaboration with intended users for developing and validating the prototype. Ethical requirements were incorporated throughout to ensure responsible and human‑centered design. The resulting research outcome is a Level‑1 software prototype that enables users to analyze cases in extended supply chains, receive risk indications and uncertainty assessments, and generate aggregated evaluations to support informed decision‑making.
