AI systems and vendor consolidation have turned parts of the academic library into a franchise model: a few suppliers control roughly 84% of ARL market share while automated cataloguing performs poorly (≈26% F1), leaving librarians accountable for opaque, vendor-controlled systems that prioritize cost-cutting over information access.

Main Finding

Vendor consolidation combined with AI deployment in academic library systems creates a "franchise" model: libraries operate vendor-controlled, cloud-hosted discovery and management platforms (the franchisor) and remain publicly accountable for service quality (the franchisee) while lacking substantive control. AI is deployed to reduce vendor labour costs, but its performance is insufficient for core professional tasks (e.g., cataloguing), and vendors externalise risk and remediation to subscribing institutions. This dynamic extracts value from professional labour, undermines professional jurisdiction, and generates information-quality externalities with systemic economic and democratic consequences.

Key Points

• Franchise mapping: Vendors (e.g., Ex Libris, Clarivate, OCLC) act as franchisors by owning proprietary SaaS platforms (Alma, WorldShare, Primo), setting defaults, roadmaps, and limiting local customization. Libraries act as franchisees paying recurring fees, operating identical systems, and absorbing user-facing failures.
• Market concentration: At peak concentration reported, vendor platforms controlled roughly 84% of ARL member institutions’ market share (Breeding, 2020), producing high switching costs and lock‑in.
• AI performance shortcomings: Multiple studies show AI cataloguing/metadata tools perform well below production requirements (paper reports ~26% F1 accuracy for subject classification in cited experiments), with independent reviews and PCC survey signalling institutional scepticism.
• Verification crisis (case study): A 2025 university case found 84% of a journal article’s 32 references fabricated; the indexed journal nonetheless carried institutional markers (DOI, ISSN, CrossRef). Vendor (Ex Libris) disclaimed proactive quality responsibility and recommended libraries manage remediation locally (Help Ticket #552419, Oct 8, 2025).
• Algorithmic content control: Documented instances of vendor-mediated content filtering/blocking for politically sensitive topics (examples include Gaza War and Tulsa race massacre) illustrate opaque moderation at scale.
• Professional impact: Platform vendors' automated processes and opaque defaults displace traditional professional jurisdiction (cataloguing, collection development, quality verification), creating "platform professionalism" where corporate tech mediates professional practice and accountability.
• Resistance & remedies: Proposed responses include collective organising, development of alternative (often open‑source or consortial) infrastructure, and reframing professional standards to address algorithmic mediation.

Data & Methods

• Methodological approach: Critical qualitative synthesis combining:
  • Empirical studies on AI accuracy for cataloguing and metadata (e.g., Potter & Saccucci 2024; Golub et al. 2024; Taniguchi 2024).
  • Longitudinal market concentration analyses (Breeding, 2020).
  • Documented cases of content filtering and verification failures (e.g., vendor support ticket #552419; reports of blocked topics).
  • Theoretical lenses from platform capitalism, sociology of professions, and critical information science.
• Evidence types: mixed—quantitative market-share and AI-evaluation statistics, qualitative case study of a verification/metadata failure, vendor support communications, and secondary literature review.
• Scope: Primarily U.S. academic library context with transnational relevance and corroborating international examples (European AI cataloguing, Australian open‑source adoption, Koha in lower‑income countries).
• Limitations noted: reliance on documented incidents and secondary empirical studies rather than large-scale randomized experiments; some accuracy metrics aggregated from different methodologies.

Implications for AI Economics

• Market structure and market power
  • Consolidation + SaaS delivery creates strong lock‑in, high switching costs, and persistent vendor rents. This resembles classic platform economics where incumbents extract recurring fees and data rents.
  • The 84% market-share concentration figure implies limited competitive pressure to improve AI quality or transparency.
• Labor and productivity dynamics
  • Vendors deploy AI to reduce vendors’ labour costs (automated cataloguing, indexing) but performance shortfalls transfer remediation labour to library staff. This reallocates labour effort from skilled, autonomous tasks to reactive, low‑visibility remediation—reducing incentives to invest in professional skills and shifting costs to institutions.
  • The franchise arrangement captures surplus created by professional labour: vendors monetize standardized outputs and usage data while diminishing the human capital base that produces high-quality metadata.
• Quality externalities and market failure
  • Poor AI accuracy (e.g., ~26% for subject classification) and opaque indexing produce negative externalities: misinformation, fabricated citations, and hidden content filtering that degrade trust and impose reputational costs on subscribing institutions.
  • Vendors’ disclaimers of responsibility signal an accountability gap: private platforms can impose social costs (misinformation, censorship risks) without internalizing them—suggesting a role for regulation or collective governance.
• Data rents and surveillance revenues
  • Platforms can turn researcher and patron interactions into prediction/analytics products (surveillance publishing), creating new monetizable streams and privacy trade‑offs. This reorients platform incentives toward data extraction rather than information quality.
• Innovation and ecosystem effects
  • Lock‑in reduces entry and incentives for interoperable, quality‑focused innovation. Open-source and consortial alternatives (where they exist) counteract vendor rents but face funding and scaling challenges.
• Policy and institutional responses (economic levers)
  • Antitrust and competition policy could mitigate concentration and lower switching costs.
  • Public investment in open, interoperable infrastructure (treated as public goods) would address underprovision of verification and quality assurance.
  • Collective procurement, shared governance, and pooled verification services (consortia) can internalize quality control and reduce vendor externalities.
  • Procurement rules requiring transparency of algorithmic processes, performance metrics, and liability allocation would realign incentives.
• Long-run macro effects
  • If unchecked, the franchise dynamic may depress human capital in professional information work, shift costs to public institutions, and concentrate economic value in a small number of platform firms—reinforcing broader patterns of platform capitalism and surveillance revenues.

Summary takeaway for AI economists: this paper identifies a concrete institutionalized mechanism—vendor-franchise dynamics in libraries—through which AI deployment and platform consolidation generate rent extraction, accountability externalities, degraded information public goods, and labour‑market distortions. Remedies involve competition policy, public/institutional investment in alternative infrastructure, and governance models that internalize quality and accountability.