Supply chain risk identification using AI-powered topic modelling: implications for demand forecasting in oil and gas companies

The oil and gas industry is highly exposed to political, economic, and supply–demand disruptions, making early risk identification critical for supply chain performance. This study applies AI-enabled topic modelling to extract risk indicators from large unstructured datasets and integrates them into gasoline demand forecasting models, showing how qualitative risk signals can improve uncertainty management in supply chains. Building on these insights, the paper proposes a generalisable three-stage methodological framework that formalises the pipeline from automated text-based risk identification through forecast validation to short-term operational adjustment. In this context, short-term operational adjustments include tactical changes to procurement, inventory levels, production scheduling, logistics routeing, and short-term hedging decisions made in response to risk-aware demand and volatility forecasts. The framework’s principal methodological contribution lies in operationalising AI-enabled text analytics within real-world supply chain planning processes through a formally specified decision-flow model. Results from the empirical validation on a 15-year gasoline demand dataset show that incorporating text-derived risk indicators, selected via rigorous feature selection, improves both point-forecast accuracy and density-forecast calibration relative to naive and seasonal baselines, offering transferable guidance for policymakers and industry stakeholders seeking to strengthen energy system resilience.