A review of 33 studies finds that firms’ AI capabilities — a bundled set of tangible, human and intangible resources — help convert AI investments into innovation and performance gains, but the evidence is fragmented across contexts and weak on causality, temporal evolution and the risks for resource-constrained organizations.

Main Finding

AI capability has emerged as a distinct, higher‑order organizational capability (building on IT capability research) that converts AI technology into sustained value only when bundled with complementary tangible, human, and intangible resources. The literature converges on resource‑based and dynamic‑capabilities logics but is fragmented across theories and application contexts; empirical work shows strong heterogeneity in how AI capabilities create value (e.g., green innovation, supply‑chain resilience, public governance), and important gaps remain around temporal dynamics, adverse effects, SME pathways, and risk governance.

Key Points

• Conceptual evolution
  • Originated from IT capability and big‑data analytics literatures; Mikalef & Gupta (2021) provided a widely used definition: AI capability = firm ability to select, orchestrate, and leverage AI‑specific resources.
  • Classic taxonomy: three higher‑order dimensions (tangible, human, intangible) as formative elements of AI capability.
  • Recent work disaggregates AI capability by technology type (e.g., generative AI, XAI) and by context (development vs implementation; sectoral variants).
• Theoretical foundations
  • Primary anchors: Resource‑Based View (RBV) and Dynamic Capabilities Theory.
  • Complementary lenses: Technology–Organization–Environment (TOE) framework and Signaling Theory.
  • RBV explains sources of sustained advantage (scarcity, inimitability of bundled resources); dynamic capabilities explain sensing, seizing, and reconfiguring enabled or amplified by AI.
• Empirical patterns and scenario heterogeneity
  • AI capabilities enable both exploratory and exploitative green innovation, improve ESG performance, and support digital + green transformations.
  • Context matters: e.g., explainable AI reduces user resistance in decision support; generative AI interacts with organizational slack; public agencies show strong prototyping but weak scaling (implementation gaps).
  • AI capability often acts as mediator/moderator between strategic intent (e.g., green strategy) and outcomes (e.g., circular‑economy activities).
• Major limitations identified
  • Fragmentation across theories and scenarios; lack of integrated nomological framework beyond RBV/dynamic capabilities.
  • Overreliance on cross‑sectional designs; limited longitudinal, experimental, or causal inference studies.
  • Underexplored “dark side” (negative externalities, algorithmic harms), temporal evolution of capabilities, and capability‑building paths for SMEs and resource‑constrained organizations.
  • Sparse attention to risk governance, accountability, and regulatory boundary conditions.
• Representative empirical findings cited
  • Mikalef & Gupta (2021): tripartite construct validated with survey of 143 U.S. tech executives; positive effects on innovativeness and performance.
  • Olan et al. (2025): XAI reduces supply‑chain managers’ resistance by ~31% vs black‑box systems in experiment (n=89).
  • Zahoor et al. (2025), Kurrahman et al. (2026), Qiao et al. (2026): show mediating/moderating roles of AI capability in green/circular transformation and interactions with organizational slack and strategic orientation.

Data & Methods

• Literature search
  • Databases: Web of Science Core Collection and Scopus.
  • Search terms: “AI capability,” “artificial intelligence capability,” “AI capabilities.”
  • Time window: 2021–2026; limited to academic journal articles.
  • Initial hits: ~50 articles; screening and quality filtering reduced sample to 33 core, peer‑reviewed studies.
• Inclusion/exclusion and screening
  • Excluded reviews, book reviews, conference abstracts; excluded purely technical/engineering papers that do not treat AI capability at organizational level.
  • Removed duplicates, papers with missing data or flawed research designs; prioritized high‑impact journals and rigorous empirical work.
• Analytic approach
  • Systematic review organized along “conceptual evolution → theoretical foundations → application scenarios → research outlook.”
  • Coded studies for conceptualization, measurement approaches, theoretical framing, empirical design, application domain, mechanisms, and boundary conditions.
  • Grouped empirical work into eight application domains (examples below) to map heterogeneity in value realization.
• Application domains synthesized (examples from the review)
  • Green innovation & sustainable management
  • Green supply chains & circular economy
  • Supply‑chain operations & resilience
  • Public governance / public sector AI capability
  • Business model innovation and organizational transformation
  • Higher education and service sectors
  • (Review notes a total of eight domains; the paper synthesizes findings across these sectors.)

Implications for AI Economics

• Conceptualizing AI capability in economic models
  • Treat AI capability as firm‑specific productive capital (distinct from physical AI hardware/software): a bundled capital stock combining tangible assets (software, data infrastructure), complementary human capital (AI expertise), and organizational capital (processes, culture).
  • This stock is inherently heterogeneous, path‑dependent, and costly to imitate—key for models of firm heterogeneity and productivity dispersion.
• Empirical strategy suggestions for economists
  • Move beyond cross‑sectional estimations: use panel/longitudinal firm‑level data, difference‑in‑differences, IV, and structural models to estimate causal effects of AI capability accumulation on TFP, innovation, and employment.
  • Instrument and measure complementarities: explicitly model interactions between AI capability and data endowments, human capital, organizational capital, and market structure to test non‑linear returns and thresholds.
  • Identify and measure “capability formation” investments (training, data governance, process redesign) rather than treating AI adoption as a one‑off treatment.
• Market structure, competition, and inequality
  • If AI capability is costly and indivisible, expect rising concentration: larger incumbents with data and organizational complements may generate persistent rents—test implications for markups, entry, and industry concentration.
  • Distributional effects: model labor reallocation and wage dispersion arising from heterogenous AI capability adoption—distinguish task displacement vs productivity‑complementarity effects.
• Policy and regulation implications
  • SME gaps: evidence shows resource‑constrained firms struggle to translate AI into capability—policy should target complementary investments (training, data ecosystems, subsidized platforms) rather than only subsidizing AI hardware.
  • Risk governance & XAI: explainability and accountability are economically consequential (affect adoption, trust, and transaction costs). Regulation that improves verifiability may lower adoption frictions and impact market outcomes.
  • Green transition: AI capability can accelerate decarbonization when paired with green strategy—economists should quantify its role in reducing firm‑level carbon intensity and model policy interactions (carbon pricing, green subsidies).
• Research agenda for AI economics (practical directions)
  • Estimate causal impacts of AI capability on firm productivity, innovation output, and markups using richer microdata (administrative, patent, platform logs).
  • Model capability accumulation as an endogenous investment problem under uncertainty and imitation costs; study dynamics of capability diffusion and lock‑in.
  • Quantify complementarities and thresholds (when AI adoption yields superlinear returns) and heterogeneity by firm size, sector, and market position.
  • Measure externalities (algorithmic harms, systemic risk) and incorporate them into welfare analyses and optimal policy design.
  • Study effects on market structure and competition: does AI capability amplify incumbent advantages and what antitrust or industrial policies mitigate negative outcomes?
  • Incorporate environmental externalities: evaluate how AI capability mediates firms’ responses to green regulation and climate policy, and its net effect on emissions trajectories.

Brief takeaway: for economists, the reviewed literature frames AI capability as a firm‑level, persistent, and heterogeneous asset that reshapes productivity, innovation, and market structure—but its empirical measurement and causal identification remain underdeveloped. Incorporating AI capability as a dynamic, complement‑rich capital in applied micro and macro models should be a priority.