Background

Large language model-based autonomous agents are moving organizations from AI-assisted towards AI-agentic workflows. Standard BPMN 2.0, however, treats activities as atomic: it cannot express what an agent actually does, how reliable it is, when a human must review its output, or how several agents coordinate with one another.

Recent work (e.g., Ait et al. 2024) proposes agentic modeling constructs as an extension of the standard BPMN modeling language to represent human-agentic collaborative workflows in business processes. However, the existing extension is explicitly presented as exploratory and is validated by re-modeling illustrative examples from the software engineering literature; it has been subjected to neither a representational analysis nor a study of whether modelers actually use it correctly, and its requirements were not derived from any application domain outside software development.

Retailers are beginning to explore agentic AI for business process automation, with early pilots and vendor deployments targeting forecast- and exception-heavy processes such as replenishment, returns handling, and product onboarding. However, few agentic use cases reach production, and both BPM scholars and practitioners identify human-in-the-loop collaboration, fallback options, and governance as core design requirements. This makes exception-prone retail processes a promising but under-researched case for human-in-the-loop agentic process-modeling constructs.

Object of investigation

The thesis is deliberately open in its emphasis. Depending on the student's interests, the weight can lie on a modeling case study with a practice partner, validating existing modeling language extensions on their ability to express agentic workflows in retail companies, on interviews with modelers or process owners, on a comprehension experiment, or on the development of retail-specific modeling extensions. Specific, adjacent aspects such as responsibility and traceability, oversight and escalation, auditability, or regulatory requirements may also be taken up as a focus.

Possible research questions

• How well can existing agentic BPMN extensions represent the characteristics of AI-supported retail processes?
• Which aspects that matter in practice cannot be expressed, or only ambiguously?
• Which refinements or additional constructs follow from this, and how could they be evaluated?

Possible results

• A set of retail processes modeled with the existing notation.
• A documented assessment of its strengths, gaps and ambiguities.
• Proposals for refinement and, depending on the chosen focus, an initial evaluation.

Starting literature

• Dumas, M., Fournier, F., Limonad, L. et al. (2023). AI-augmented Business Process Management Systems: A Research Manifesto. ACM Transactions on Management Information Systems, 14(1), Art. 11. DOI 10.1145/3576047.
• Ait, A., Cánovas Izquierdo, J. L. & Cabot, J. (2024). Towards Modeling Human-Agentic Collaborative Workflows: A BPMN Extension. arXiv:2412.05958 (also SEAA 2025, Springer LNCS 16081, DOI 10.1007/978-3-032-04190-6_22).
• Cao, L. (2021). Artificial intelligence in retail: applications and value creation logics. International Journal of Retail & Distribution Management, 49(7), 958–976. DOI 10.1108/IJRDM-09-2020-0350
• Vu, H., Klievtsova, N., Leopold, H., Rinderle-Ma, S., Kampik, T. (2026). Agentic Business Process Management: Practitioner Perspectives on Agent Governance in Business Processes. DOI 10.1007/978-3-032-02936-2_3

Recommended prior knowledge

Business Process Management and process modeling; conceptual modeling; an interest in AI agents and large language models.