Academic interest in AI for Islamic finance has surged since 2020, yet no published study to 2025 applies large language models, RAG, or the AAOIFI standards as AI knowledge-bases — exposing a clear research and practical-compliance gap.

Main Finding

This systematic review (2018–2025, Scopus) maps AI, ML, and LLM research in Islamic finance and identifies a fast-growing but narrowly focused field. Although AI/NLP/ML methods are increasingly applied across Islamic banking, social finance, capital markets, and governance, there is a complete absence (in the 45 peer‑reviewed articles retained) of studies that: (a) apply large language models (LLMs) or generative AI/RAG architectures to Islamic finance and (b) leverage the AAOIFI corpus (100+ Shariah, accounting, auditing, governance standards) as a knowledge base. The authors flag the “LLM–RAG–AAOIFI” nexus as the most consequential research gap and propose 15 future research questions (including three novel, domain‑directed directions).

Key Points

• Scope and scale
  • Initial Scopus hits: 868; final PRISMA‑screened corpus: 45 peer‑reviewed articles (2018–2025).
  • Authors: 142 researchers across 14 countries.
  • Rapid recent growth: 21 of the 45 articles published in 2025 alone.
  • Total citations at extraction: 407 (mean ≈ 9.04 per article); most‑cited paper (Syed et al., 2020) has 114 citations but uses classical NLP rather than LLMs.
• Geographic patterns
  • Indonesia leads publication volume; Bahrain and Jordan notable for citation impact; Saudi Arabia a central collaborator.
• Thematic streams (from content analysis)
• AI and ML in Islamic banking and risk management (e.g., credit scoring, efficiency analysis).
• AI, NLP, and text analytics in Islamic social finance (e.g., zakat, Qardh‑Al‑Hasan programmes).
• ML and deep learning applications in Islamic capital markets (e.g., sukuk analytics, trading).
• Ethical AI governance in Islamic FinTech (e.g., cyber governance, Shariah governance implications).
• Critical gap
  • No retained study directly applies LLMs, generative AI, or retrieval‑augmented generation (RAG).
  • No study indexes, embeds, retrieves from, or queries the AAOIFI standards corpus computationally.
• Practical examples from corpus
  • Murabaha business‑process screening against Quran/Hadith (text mining).
  • ML‑based P2P lending model for SMEs.
  • Zakat administration knowledge discovery (YouTube/Zoom text mining).
  • Cyber‑governance and AI influence studies.

Data & Methods

• Search and selection
  • Database: Scopus only.
  • Time window: PUBYEAR > 2017 AND PUBYEAR < 2026 (i.e., 2018–2025).
  • Boolean TITLE‑ABS‑KEY query combined AI/LLM/ML terms with Islamic finance terms (explicitly included "large language model", LLM, "generative AI", ChatGPT, RAG‑relevant language).
  • PRISMA 2020 three‑stage selection: Identification → Screening → Eligibility; reasons for exclusion documented (book chapters, conference papers, reviews, non‑English, out‑of‑range, irrelevant methods/topics).
  • Flow: 868 → (remove non‑journal items, duplicates, non‑English, etc.) → 418 → (title/abstract screening) → 79 → (eligibility full‑text) → 45 included.
• Analysis tools and approach
  • Bibliometric mapping with VOSviewer: keyword co‑occurrence, country co‑authorship networks.
  • Full‑text content analysis to define thematic streams and research gaps.
  • Citation analyses (global citations, citations/year, normalized citations).
• Key data/metrics reported
  • 45 articles, 142 authors, 14 countries.
  • Mean collaboration index ≈ 4 co‑authors/document.
  • Corpus accumulated 407 citations (mean ≈ 9.04); publication year citation breakdown provided.

Implications for AI Economics

Opportunities - Compliance automation and cost reduction - RAG + LLMs grounded in AAOIFI could automate Shariah‑compliance screening (contracts, sukuk structuring, fatwa retrieval), reducing transaction and compliance costs and accelerating product innovation. - Domain‑specific LLM markets - Demand for LLMs fine‑tuned on AAOIFI standards, fatwas, jurisprudential corpora, and localized regulatory texts (regional adaptations) — economic opportunity for fintech vendors and open‑data initiatives. - Efficiency and inclusion - Scalable AI tools (zakat beneficiary identification, P2P credit scoring) could expand financial inclusion and reduce frictions in Islamic microfinance. - New datasets and benchmarks - AAOIFI and associated Shariah corpora can become high‑value domain datasets for model training, benchmarking, and reproducible research.

Risks and research needs - Governance and interpretability - Shariah compliance requires transparent, interpretable decisions and human oversight (Shariah boards). LLM opacity raises risks of incorrect compliance verdicts and reputational/regulatory damage. - Validity and standards - Need formal metrics and evaluative frameworks for "Shariah‑compliance accuracy" (not standard in mainstream ML), plus gold‑standard annotated corpora. - Market structure and labor impacts - Automation may displace or transform roles of Shariah advisors and compliance staff — economic implications for labor markets and service pricing. - Regulatory and ethical concerns - Potential for model bias, misuse, and adversarial manipulation of RAG retrieval sources; cross‑jurisdictional differences in Shariah interpretation could create regulatory arbitrage.

Policy and research actions recommended - Open, audited AAOIFI‑aligned datasets: encourage AAOIFI or consortia to publish machine‑readable standards and annotations to enable safe model development and benchmarking. - Human‑in‑the‑loop design: mandate Shariah‑board verification layers and explainability requirements for deployed LLM/RAG systems. - Benchmarks and evaluation metrics: create standardized tests for Shariah compliance, faithfulness of retrieval, and downstream economic impact. - Interdisciplinary studies: combine economics, Islamic jurisprudence, NLP, and regulation to evaluate market effects (pricing, competition, inclusion).

Concrete future research directions (illustrative, drawn from authors’ emphasis) 1. Build and evaluate LLMs fine‑tuned on AAOIFI standards + fatwa corpora; measure Shariah compliance fidelity using expert annotations. 2. Design RAG‑powered compliance assistants that index AAOIFI + national regulations to support sukuk issuance, contract drafting, and automated fatwa retrieval with human oversight. 3. Quantify the economic impacts of domain LLM adoption in Islamic finance (costs saved, effects on product innovation, labor displacement among Shariah advisors, and market concentration effects).

Summary takeaway The field has rapid bibliometric growth and several promising AI applications in Islamic finance, but misses an urgent frontier: applying LLMs/RAG to the AAOIFI‑anchored compliance problem. Filling this LLM–RAG–AAOIFI gap could generate substantial economic value and innovation — provided governance, interpretability, and Shariah oversight are built into model design, evaluation, and deployment.