In construction procurement, forecasts of contract award values are typically driven by the initial estimate. Rather than directly predicting the awarded amount, as performed by the baseline Random Forest model, this study introduces a deviation-based decision-support model for public procurement that predicts the difference between the estimate and the eventual award value. The dataset was obtained from the Moroccan Portal of Public Procurement. Model performance was evaluated using RMSE and MAE. The findings show that the estimate accounts for 99.7% of the predictive influence in the baseline model. However, the proposed approach provides greater predictive accuracy, recording an MAE of 113,318.82 MAD, whereas the baseline model produces an MAE of 207,986.76 MAD. Moreover, while predictions generated by the baseline model remain strongly associated with the estimate, modelling the deviation reveals the relevance of contextual characteristics, including the contracting authority’s name and geographical location, for data-driven public procurement analysis. The remainder of this paper is organised as follows. Following the introduction, the second section reviews the relevant literature, while the third section describes the research methodology and dataset. The subsequent sections report the results and discuss the strengths and limitations of the study. The final section concludes the paper and identifies directions for future academic research.