Cost overruns and delays remain persistent challenges in construction projects. Traditional estimation methods struggle to capture uncertainty and nonlinear interactions among risk factors, leading the industry to adopt artificial intelligence (AI) for predictive management. In this study, a systematic literature review (SLR) of 44 documents is conducted to identify the best-performing AI models for forecasting cost overruns and delays across project lifecycle phases. The review reveals that model applicability is dictated by data maturity: Regression models, particularly decision tree ensembles and hybrid models, are most effective during Feasibility and Design, respectively, where data availability is limited. Classification models do not appear until the Bidding–Contracting stage, when risk definitions become more granular. The Construction phase concentrates the highest model diversity and is dominated by robust neural networks, decision tree ensembles, and hybrid architectures capable of processing dynamic execution data. Post-construction analyses rely primarily on Bayesian networks and hybrid models to capture causal dependencies and support learning for future cycles. Findings indicate a clear industry shift toward hybrid and ensemble solutions and highlight the need for multi-objective approaches that jointly model cost and schedule outcomes. This study contributes a phase-based framework to guide practitioners in selecting AI techniques aligned with project data maturity.