An Automating Tendering Performance Interpretation via Process Mining and an LLM-Based Agent

Digital transformation in procurement, particularly e-tendering, is generating rich transaction data that can support evidence-based performance evaluation and decision-making. Process mining provides a data-driven methodology for process execution; however, its outputs, including process models and activity performance statistics, frequently pose interpretative challenges for non-expert users and necessitate manual analysis. This challenge can slow decision-making and limit the organizational adoption of process-mining-based performance evaluations. This study presents an end-to-end framework that integrates event-log-based process mining with an LLM-based AI agent to automate the interpretation of descriptive tender performance by translating process-mining evidence into decision-oriented narratives aligned with predefined analytical objectives. In a case study, the agent used a DFG artifact to extract 16 activity labels and their average durations with 100% accuracy, validated against process-mining activity statistics. The agent further produced descriptive narratives that summarize activity performance, identify bottlenecks and dominant delay paths, and prioritize improvement recommendations. An expert review confirmed that the interpretations and suggestions make sense, fit the situation, and can be used for tendering. Overall, the proposed framework operationalizes an evidence-grounded interpretation layer that reduces reliance on manual interpretation, improves the usability of process-mining outputs for procurement decision-makers, and advances automation within the descriptive analytics layer of the augmented Business Process Management (BPM) pyramid.