Implementing an AI Agent for Logistics Compliance

Abstract

This thesis investigates how an AI agent can be applied to logistics processes that involve un structured and variable compliance documentation. The objective is to examine whether an AI agent can support or partially automate the screening of supplier declarations, a task that cur rently requires extensive manual effort due to large amounts of declarations and inconsistent formatting. The study aims to determine the feasibility, usefulness, and organizational implica tions of introducing AI agents into a compliance-heavy workflow.

The research applies the Design Science Research Methodology in combination with a case study within an industrial logistics organization. An AI agent system model was designed using Copilot Studio to read, interpret, and classify supplier declarations based on regulatory and organiza tional requirements. The agent was integrated into a workflow and tested using real declara tions together with domain experts responsible for current manual processing. The evaluation focused on agent’s ability to analyse heterogeneous document structures and make decisions based on them.

The results show that the AI agent can reliably analyse and classify supplier declarations, includ ing cases with irregular or complex formatting in a controlled environment. Experts confirmed that the agent’s decisions and explanations were consistent with human judgement, and that the system would take less time and effort for document screening. Introducing a confi- dence-based threshold further improved perceived safety and trust in the system’s output, par ticularly given the regulatory sensitivity of the process.

The study concludes that AI agents are practical and valuable extension to existing RPA based automation solutions of the organization. By automating the initial assessment of declarations, the agent could help to reduce manual workload and enables employees to focus on deci sion-making tasks that require expertise. The findings highlight the potential of AI to address limitations of rule-based automation by providing contextual reasoning, adaptability, and trans parency. The thesis also identifies future research opportunities, including large-scale evalua tion, multi-agent architectures, and incorporating learning mechanisms to improve long-term performance and autonomy.