Sustainability impact assessment is fragmented but dominated by multi-criteria decision-making; the authors harmonise indicator sets and introduce a normalised weighting to create an early-stage benchmark. Their synthesis paves the way for standardised SIA datasets and hybrid MCDM–AI tools that could power ML-driven firm-level impact estimation and decision support.

Main Finding

The paper systematically reviews methodological developments in corporate Sustainability Impact Assessment (SIA) using a Triple Bottom Line (TBL) lens and finds that research is methodologically fragmented but dominated by multi-criteria decision-making (MCDM) approaches (≈55% of studies). Studies concentrate on supply-chain contexts, show growing thematic convergence with sustainability disclosure frameworks, and exhibit heterogeneous indicator sets. By synthesising indicators and introducing a study-specific normalised weighting approach, the review offers a preliminary integrative reference to simplify early-stage SIA and improve cross-study comparability.

Key Points

• Scope and corpus
  • Bibliometric mapping: 303 articles.
  • In-depth qualitative synthesis (PRISMA-based): 67 peer-reviewed studies from 2015–2025.
• Methodological landscape
  • MCDM approaches dominate (~55% of reviewed studies).
  • Methodological heterogeneity: many distinct indicator sets, weighting schemes, and procedural steps.
  • Limited methodological integration between SIA methods and sustainability disclosure frameworks despite thematic convergence.
• Contexts and themes
  • Strong emphasis on supply-chain contexts.
  • Indicator groups with highest stakeholder-assigned weights (and cross-TBL span):
    • Greenhouse gas emissions / climate change
    • Pollution and waste management
    • Green management and circularity
    • Operational and process efficiency
  • Stakeholder-prioritised weights are distributed almost evenly across environmental, social, and economic pillars, with a slight tilt toward economic indicators.
• Contribution
  • Consolidates heterogeneous indicator sets into a harmonised inventory.
  • Introduces a study-specific normalised weighting approach to enable early-stage, cross-study comparability.
  • Uses the Theory–Context–Characteristics–Methodology (TCCM) framework for structured analysis.

Data & Methods

• Mixed-method review combining:
  • Bibliometric analysis (303-article mapping) to identify publication patterns, thematic clusters, and methodological prevalence.
  • PRISMA-guided qualitative systematic literature review for in-depth synthesis of 67 recent peer-reviewed studies (2015–2025).
• Analytical framework: Theory–Context–Characteristics–Methodology (TCCM) to organise and compare theoretical foundations, sectoral/operational contexts, study characteristics (indicators, weights), and methodological choices (MCDM, qualitative, hybrid, etc.).
• Indicator synthesis:
  • Extraction and consolidation of indicator groups across studies.
  • Development of a normalised, study-specific weighting procedure to make heterogeneous weights comparable across papers and derive a stakeholder-weighted ranking of indicator groups.

Implications for AI Economics

• Dataset standardisation and benchmarking
  • The fragmented indicator space and heterogenous weights underscore the need for standardised, labelled SIA datasets. Such datasets would enable rigorous ML/AI model training, validation, and benchmarking for economic impact estimation and policy simulation.
• Feature engineering & priors
  • The consolidated indicator groups and normalised weighting approach can serve as structured feature sets and priors for predictive models (e.g., for estimating firm-level sustainability impact on financial or social outcomes).
• Methodological synthesis opportunities
  • Dominance of MCDM suggests a gap for AI methods: combine MCDM with machine learning, causal inference, and optimization (e.g., hybrid models that use ML for indicator measurement/forecasting and MCDM for preference aggregation).
• Causal and econometric modelling
  • The review stresses the need to better link sustainability implementation to measurable outcomes. AI economists can focus on causal inference approaches (instrumental variables, difference-in-differences, double ML) to estimate effects of SIA-informed interventions on economic performance, externalities, and welfare.
• Supply-chain and network models
  • The supply-chain focus invites network-economics and agent-based models enhanced with AI to capture propagation of sustainability impacts, externalities, and policy interventions across interfirm networks.
• Explainability, accountability, and disclosure integration
  • Convergence with disclosure frameworks calls for AI systems that produce explainable, audit-ready SIA outputs compatible with reporting standards—important for regulation, investor decision-making, and incentive design.
• Practical tooling and decision support
  • The normalised weighting and indicator consolidation can be embedded into decision-support tools and automated SIA workflows (e.g., dashboards, policy simulators) to aid firms and regulators in early-stage assessments and scenario analysis.
• Research agenda suggestions for AI economists
  • Build and publish standard SIA datasets with harmonised indicators and weights.
  • Develop hybrid MCDM–ML pipelines and validate them against causal-impact estimands.
  • Explore transfer learning across sectors to address sparse data in firm-level SIA.
  • Design interpretable AI models to map sustainability actions to economic outcomes for investors and policy-makers.

If you want, I can: (a) turn the consolidated indicator groups and the normalised weighting procedure into a concrete dataset schema or feature list for ML; or (b) outline a prototypical hybrid MCDM + causal ML pipeline for estimating firm-level sustainability impacts. Which would be more useful?