AI/NLP-Based Predictive Project Risk Analytics: The Contract Risk Analyzer for Energy EPC Contracts
DOI:
https://doi.org/10.61336/cjmr.1604.36Keywords:
Contract Analytics, Text Mining, Natural Language Processing, Hybrid Ensemble Learning, B2B Tendering, Risk Intelligence, Energy EPC ProjectsAbstract
In business-to-business energy markets, much of the value of a project is won or lost at the tender stage, where long engineering, procurement and construction (EPC) contracts allocate cost, schedule and liability risk through dense legal text. This article develops and evaluates the Contract Risk Analyzer, a design-science artefact that converts contract sentences into a prioritized and explainable risk register. A corpus of 1,304 cleaned sentences drawn from four FIDIC contract books was labelled by sentence type (five classes) and responsible party (three classes). Twenty individual classifiers built on bag-of-words, TF-IDF, character n-gram and latent semantic representations were benchmarked against eight hybrid ensembles using an 80/20 hold-out design and five-fold cross-validation. The deployed stacking meta-learner achieved 89.7% accuracy, 88.7% macro-F1 and 98.9% ROC-AUC for sentence type, and an optimized soft-voting ensemble achieved 97.4% accuracy and 98.1% macro-F1 for responsible party. The best hybrid exceeded the best individual model by 4.7 macro-F1 points. Classifier outputs were combined with onerous-term rules and a 3,000-record quantified risk library and applied to a 176-page EPC works contract containing 2,240 sentences. The system registered 371 risk-bearing findings across 24 categories, including 102 Critical and 165 High findings, produced a Contract Risk Index of 79.2% and allocated 71.4% of findings to the contractor. A composite predictability index routed 30.5% of findings to expert review. The results show how text analytics turns contract documents into decision-ready intelligence for bid/no-bid, pricing and negotiation decisions, and they offer a reproducible template for hybrid text classification in marketing research.
Downloads
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Canadian Journal of Marketing Research

This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.

