Academic research has barely touched NLP for bank marketing: only 8 of 109 peer‑reviewed studies directly examine it, and the literature clusters on retention rather than acquisition, personalization or external-data integration—pointing to large unrealised economic and managerial opportunities.

Main Finding

A systematic review (PRISMA protocol) of 109 peer‑reviewed articles (2014–2024) finds that NLP applications in bank marketing are severely under‑studied: only 8 papers directly examine NLP in bank marketing, while 74 study NLP in marketing broadly and 27 study marketing in banking without NLP. Existing work concentrates on customer retention; customer acquisition, personalized engagement, use of external text data, and operational excellence remain important gaps.

Key Points

• Scope and sample: 109 peer‑reviewed journal articles published 2014–2024 were screened and reviewed.
• Coverage breakdown: 8 papers on NLP in bank marketing; 74 on NLP in marketing (general); 27 on broader marketing in banking (non‑NLP).
• Thematic concentration: literature is clustered around customer retention (e.g., churn prediction, complaint handling, relationship management).
• Under‑explored areas: customer acquisition, personalization at scale, external text sources (social media, news, public reviews), operational process improvement, and cross‑channel integration.
• Novelty of the study: one of the first systematic reviews focused specifically on NLP in bank marketing; organizes findings along the customer journey and the marketing mix to provide a practical taxonomy.
• Visualization/analysis tools: sentence‑transformer embeddings to encode article text and UMAP (Uniform Manifold Approximation and Projection) to map the thematic landscape and identify sparsely populated research areas.

Data & Methods

• Literature protocol: Preferred Reporting Items for Systematic Reviews and Meta‑Analyses (PRISMA).
• Selection: peer‑reviewed journal articles between 2014 and 2024; final sample n = 109.
• Analytical approach:
  • Structured review of two overlapping domains: analytical marketing in banking and NLP applications in marketing.
  • NLP embedding technique: sentence‑transformer models to produce dense vector representations of article abstracts/text.
  • Dimensionality reduction/visualization: UMAP to project embeddings into a low‑dimensional thematic map, enabling cluster identification and gap detection.
• Outcome measures: topical coverage by customer journey stage (acquisition, engagement, retention, operational excellence) and by elements of the marketing mix.

Implications for AI Economics

• Market potential and diffusion
  • Limited applied research on NLP for acquisition and personalization suggests unrealized value in banking: NLP could enable more efficient, targeted customer acquisition and cross‑sell, potentially lowering customer‑finding costs and increasing lifetime value.
  • The concentration on retention reflects earlier, lower‑risk adoption; moving into acquisition/personalization could accelerate competitive dynamics and spur investment in NLP capabilities.
• Productivity and organizational complementarities
  • Realizing NLP value requires organizational investments (data pipelines, model deployment, integration with CRM), implying complementarity between AI tools and managerial/IT capabilities. Returns depend on these complementarities—an important consideration for adoption models and firm‑level productivity studies.
• Data externalities, privacy, and regulation
  • Under‑use of external text sources (social media, news, reviews) may owe to privacy, legal/regulatory uncertainty, or integration costs. Economic models should account for data access frictions, platform controls, and regulatory constraints when estimating NLP’s ROI in banking.
• Competition, differentiation, and market structure
  • Banks that operationalize NLP for personalization and acquisition can differentiate services, increasing switching costs and possibly market concentration if these advantages scale. Antitrust and market structure analyses should monitor NLP‑driven segmentation and incumbency advantages.
• Labor and skill composition
  • Greater NLP adoption could shift demand toward analytics, data engineering, and model‑monitoring skills while reducing routine marketing tasks. This has implications for wage structure and re‑skilling needs within financial services.
• Measurement and causal inference needs
  • The literature’s descriptive/engineering tilt highlights a gap in causal, field‑experimental evidence on NLP interventions’ effects on customer behavior and firm profits. AI economics should prioritize experimental or quasi‑experimental designs to estimate treatment effects, cost‑benefit, and general equilibrium impacts.
• Research & policy priorities
  • Empirical work quantifying the welfare impacts (consumer surplus, privacy costs), the distribution of gains across firms and consumers, and regulatory trade‑offs is needed.
  • Studies that incorporate externalities (data spillovers), platform interactions (e.g., social media data access), and dynamic effects on competition will sharpen policy guidance.

Suggested immediate research directions for AI economists: - Field experiments in banks testing NLP‑driven acquisition/personalization versus standard approaches, measuring CAC, LTV, retention, and consumer welfare. - Structural/empirical models of adoption decisions that account for data access costs and organizational complementarities. - Analyses of privacy regulation impacts on the availability and economic value of external text data for marketing. - Macro‑level studies on how scaling NLP in banking affects market structure, employment, and productivity.