AI Agent Operational Lift for San Diego County Library in San Diego, California

Why AI matters at this scale

San Diego County Library (SDCL) operates as a mid-sized public library system with 201-500 employees serving a diverse population across urban, suburban, and rural branches. With an estimated annual budget around $45 million, SDCL sits in a unique position: large enough to generate significant data from millions of annual circulations and patron interactions, yet constrained by public-sector procurement cycles and limited dedicated technology staff. AI adoption here isn't about flashy innovation—it's about doing more with fixed resources, improving equity of access, and freeing skilled librarians from repetitive tasks to focus on community engagement. The library's core assets—vast metadata, circulation histories, and digital collections—are fuel for machine learning models that can transform patron experiences without requiring massive new investments.

1. Reinventing discovery with conversational AI

The highest-ROI opportunity lies in replacing the traditional OPAC (Online Public Access Catalog) keyword search with a conversational interface. Patrons often struggle to translate their needs into Library of Congress subject headings. An LLM-powered search layer, fine-tuned on the library's catalog and community interests, allows queries like "I want a gripping mystery set in a small town, similar to Louise Penny" and returns actual available titles with real-time hold status. This reduces staff time spent on reader's advisory and increases circulation of mid-list titles. Deployment can start as a vendor-provided widget integrated into the existing Bibliocommons or Sirsi Dynix interface, with a modest $15-25k annual cost that can be offset by measurable increases in digital and physical checkouts.

2. Automating special collections digitization

SDCL holds unique local history materials—photographs, maps, and oral histories—that remain undiscoverable in unstructured digital folders. Computer vision APIs and open-source NLP models can auto-generate descriptive metadata, transcribe handwritten documents, and tag images with locations and dates. A pilot focusing on 10,000 items could reduce cataloging backlogs by 70% and make these collections searchable online for the first time. The ROI is both mission-driven (preserving local heritage) and operational (reducing manual metadata entry by specialized archivists). Risks include model accuracy on historical cursive and the need for human-in-the-loop review, but even semi-automated workflows yield 5x efficiency gains.

3. Predictive analytics for materials management

Libraries waste budget on bestsellers with long hold lists while under-investing in niche topics with steady demand. By training a model on five years of circulation data, demographic overlays, and seasonal trends, SDCL can forecast demand at the branch level and optimize purchasing. This reduces the "turnaway rate" (patrons leaving empty-handed) and improves the return on the $3-5 million annual materials budget. Implementation requires clean data pipelines from the ILS (Integrated Library System) and a dashboard for collection development librarians—achievable with a part-time data analyst and Power BI licenses already in the Microsoft 365 Government stack.

Deployment risks specific to this size band

Mid-sized public libraries face acute risks around vendor lock-in and sustainability. A flashy AI chatbot from a startup may disappear in two years, leaving no institutional knowledge. SDCL should prioritize solutions built on open-source models or offered by established library vendors with state procurement contracts. Data privacy is paramount—patron borrowing records are protected by California law, so any cloud-based AI must be deployed in a government community cloud (GCC) environment with no data used for model training. Finally, staff buy-in is critical; without a change management plan that positions AI as an assistant rather than a threat, even the best technology will be underutilized. Start with a transparent pilot, measure patron satisfaction and staff time saved, and scale based on evidence.