Construction risk assessment remains anchored to simple probability–impact methods despite a post‑2010 rise in hybrid, fuzzy and AI approaches; however, validated integrated frameworks that jointly manage cost, schedule, quality and systemic cascading failures are largely absent.

Main Finding

The review of 91 peer-reviewed articles (1990–2025) shows that conventional probability-impact (P–I) approaches remain the dominant risk-assessment method in construction projects despite limitations in representing interdependent and cascading risks. Since ~2010 there has been a clear shift toward hybrid, fuzzy-logic, and AI-enhanced techniques (notably Fuzzy Set Theory, Analytic Hierarchy Process, and Monte Carlo Simulation), but important gaps persist—most notably the absence of integrated, multi-objective frameworks that jointly address cost, time, quality, and overall project performance.

Key Points

• Dataset: 91 articles from 15 leading construction/risk journals, reviewed via a structured four-stage process and synthesized through taxonomic and thematic analyses.
• Persistent dominance of the probability–impact (P–I) model, despite known shortcomings in handling risk interdependencies and cascading effects.
• Most-used alternative/advanced methods: Fuzzy Set Theory (FST), Analytic Hierarchy Process (AHP), and Monte Carlo Simulation (MCS).
• Evolution over time:
  • 1990s: basic probabilistic approaches and descriptive risk matrices.
  • Post-2010: growth in hybrid methods, fuzzy logic, and incorporation of AI/machine-learning elements.
• Geographic and publication trends analyzed (concentration in specific regions and journals noted; details in full review).
• Major gaps identified:
  • Lack of integrated frameworks that model multiple objectives simultaneously (cost, time, quality, performance).
  • Limited dynamic modelling of evolving risks and weak treatment of risk interdependencies and cascading failures.
  • Insufficient calibration and validation of AI/ML methods on large, multi-project datasets.
• Roadmap proposed: development of ML applications for predictive risk scoring, dynamic and system-level modelling, and holistic multi-objective risk assessment frameworks.

Data & Methods

• Scope: systematic taxonomic review covering 1990–2025.
• Sample: 91 peer-reviewed articles drawn from 15 leading journals in construction risk and management.
• Process: structured four-stage review (search and identification; screening/selection; taxonomic classification and mapping; thematic synthesis).
• Analytical techniques:
  • Taxonomy to classify methods by chronology, study type, authorship patterns, and focus areas.
  • Thematic analysis to extract trends, recurring themes, methodological developments, and research gaps.
• Outcome: method map showing prevalence and chronological adoption of P–I, FST, AHP, MCS, hybrid techniques, and emergent AI-enhanced approaches.

Implications for AI Economics

• Value creation and ROI:
  • AI-enhanced risk models (ML + dynamic simulation) could materially reduce cost/time overruns and quality failures, improving project-level returns and lowering financing costs.
  • Economists should quantify the marginal economic value of integrating AI risk tools versus standard P–I approaches (cost savings, schedule adherence, insurance premiums).
• Markets and diffusion:
  • Growth opportunities for AI risk-product markets (software vendors, consultancies); adoption likely uneven across regions and firm sizes—research needed on diffusion dynamics and adoption barriers.
• Labor and capital complementarity:
  • AI risk systems may change skill demands (more data/analytics roles) and alter investment in project controls and monitoring capital; study complementarities and displacement effects.
• Policy, contracts, and finance:
  • More accurate risk allocation models enabled by AI can affect contract design, insurance pricing, bond issuance, and lender covenants; regulators and standard setters may need new guidance on validated AI models.
• Research directions for AI economists:
  • Empirical ROI studies using quasi-experimental designs or randomized pilots to estimate causal effects of AI risk tools on costs, schedules, and quality.
  • Development and economic evaluation of multi-objective, dynamic risk frameworks that internalize interdependencies and cascading failures.
  • Welfare analyses of market structure changes (competition among AI vendors) and distributional effects across firms and workers.
  • Integration of probabilistic simulation (MCS), fuzzy logic, and ML for improved uncertainty quantification; economic calibration (e.g., calibration to cost/schedule loss distributions) and stress-testing.
• Data needs:
  • Large, multi-project longitudinal datasets for training/validation of ML models and for credible economic evaluation; incentives or policies may be required to improve data sharing.

If you want, I can (a) extract a prioritized list of concrete research questions for AI economists from the review, (b) map specific ML methods (e.g., gradient boosting, graph neural networks) to the identified gaps, or (c) draft an outline for an empirical study to estimate the ROI of AI-based risk modelling in construction. Which would you prefer?