Services

Libraries serve more channels and expectations with limited teams. The result is constant tension: high response times, repetitive queries consuming capacity, users who cannot "reach" the right resources, and underutilization of digital collections. This service turns that pressure into an opportunity: a service front that operates 24/7, with institutional quality and human escalation when needed.

1. 24/7 support (FAQ + procedures + services): resolves frequently asked questions and guides processes: hours, membership, loans, renewals, fines, rooms, remote access, digital resources, events, and help paths.
2. Reference and recommendation agent: supports brief reference interviews, suggests search paths and resources (catalog, repository, subject guides, databases), and explains its recommendations.
3. Research co-pilot: assists with formulating search strategies, iterative refinement, and organizing findings. Prioritizes evidence and avoids "inventing" sources.

We define during the diagnostic phase which capability to implement first. Further capabilities can be added in subsequent phases.

1. Assistant deployed on web (and optionally on other channels).
2. Curated knowledge base with owners, versioning, and update cycle.
3. Conversational flows with human escalation, institutional messages, and safety paths.
4. Traceable responses: links to official sources and flags when something requires human validation.
5. Analytics panel: most consulted topics, information gaps, resolution rate, and improvement opportunities.
6. Operations manual + training: how to update content, review quality, and adjust rules.

• Human-in-the-loop for sensitive cases (personal data, legal matters, wellbeing, formal complaints).
• Anti-hallucination policy: the assistant cites sources or declares uncertainty; it does not "fill in" with assumptions.
• Data minimization, role-based permissions, and retention guidelines.
• Interaction logs and periodic review (quality, bias, accessibility, and security).

• Resolution rate without human (containment) and successful escalation rate.
• Average response time (before/after).
• User satisfaction (CSAT/NPS) and perceived quality.
• Reduction in repetitive tickets and front-desk load.
• Information gaps detected (input for improving guides and signage).

Cataloging underpins everything else: discovery, analytics, access, and preservation. But it is also one of the most time-intensive processes and most susceptible to rework from inconsistencies, duplicates, or lack of normalization. This service accelerates the workflow without sacrificing quality or professional judgment: AI proposes, the team decides.

• Assisted metadata extraction from reliable sources (cover, TOC, DOI/identifiers, repositories, publishers).
• Description suggestions (abstracts, notes, audiences, keywords) aligned to local policies.
• Normalization and authority control: names, series, subjects, and terms; detection of variants and inconsistencies.
• Metadata enrichment to improve discoverability: more consistent subjects, improved descriptors, complete identifiers.
• Assisted classification/indexing and quality alerts (missing fields, duplicates, conflicts).

The exact scope is defined during the diagnostic: document types, standards, languages, sources, and library management system/repository.

We start from a baseline (actual times per material type) and design the workflow to maximize savings where reasonable: smart pre-filling, reusable suggestions, normalization rules, and rework reduction. We then test on a representative pilot and adjust until the agreed target is reached.

1. "Before/after" process map and operations guide.
2. Field templates and normalization rules agreed with the cataloging team.
3. Pipeline implemented with traceability of suggestions and changes.
4. Quality checklist and basic productivity and consistency dashboard.
5. Technical training for the core team + operational onboarding.

• Average time per record (by material type).
• % of suggested fields accepted vs. corrected (assistance quality).
• Rework rate and detected duplicates.
• Authority and subject consistency (before/after).

Users expect to search the way they speak, not the way libraries catalog. The gap between natural language and controlled terms creates frustration, low collection use, and the perception that "the library doesn't have what I need" — when it does. This service closes that gap with semantic and conversational search.

• Semantic search over catalog, institutional repository, resource guides, and digital collections.
• Conversational interface: the user expresses their need in natural language and receives relevant results with explanations.
• Query expansion: synonyms, related terms, thesauri, and contextual suggestions.
• Dynamic filters: by collection, material type, date, language, and availability.
• Result traceability: why each resource is recommended.

1. Semantic search engine implemented and integrated with agreed sources.
2. Interface configured with institutional identity.
3. Technical documentation of architecture and maintenance guide.
4. Analytics dashboard: most searched terms, collection gaps, click-through rate, and satisfaction.
5. Training for technical and reference teams.

• Zero-results search rate (before/after).
• Clicks on results and resource access rate.
• User satisfaction with results (brief survey).
• Digital collection usage before/after implementation.

Libraries generate valuable data that rarely becomes decisions. Collection development is based on intuition, usage reports arrive late and in hard-to-interpret formats, and institutional impact remains invisible. This service transforms scattered data into actionable intelligence for leadership, services, and planning.

• Institutional KPI dashboard: collection use, loans, remote access, events, and satisfaction.
• Collection development analytics: most and least used titles, thematic gaps, duplicates, and candidates for withdrawal or acquisition.
• Impact reports: indicators to demonstrate library value to leadership and funding bodies.
• Automatic alerts: usage thresholds, expirations, anomalous trends.
• Integration with existing data sources (ILS, repository, surveys, commercial databases).

1. Operational dashboard implemented with role-based access (leadership, team, reports).
2. Initial set of KPIs agreed with leadership.
3. Documentation of data sources, update frequency, and owners.
4. Metrics interpretation and usage guide.
5. Training for the team that will administer and use the system.

• % of collection decisions supported by data (vs. intuition).
• Time to produce usage reports (before/after).
• Leadership satisfaction with information available for planning.
• Dashboard usage by the team (frequency and queries generated).

Implementing AI without governance means taking unnecessary risks: compromised privacy, undetected biases, staff without criteria to evaluate tools, and decisions made without traceability. Governance is not bureaucracy — it is the foundation for AI to function reliably, durably, and responsibly in your institution.

• Institutional AI use policy: principles, scope, restrictions, and update mechanisms.
• Roles and responsibilities map: who can use which tools, with what data, and under what conditions.
• Privacy Impact Assessment (PIA) for existing or planned AI services.
• Risk matrix: identification, probability, impact, and controls by risk type (privacy, bias, vendor dependency, misinformation).
• AI vendor evaluation checklist: critical questions before contracting or renewing.
• Critical AI training plan: internal diploma by levels (awareness, responsible use, oversight).

1. Institutional AI policy (version 1.0, with annual review cycle).
2. Privacy impact assessment completed for current or in-progress services.
3. Risk matrix with assigned controls and owners.
4. Vendor checklist (ready to use in current and future evaluations).
5. Critical AI training plan for the entire team.
6. Kickoff workshop (3–4 h) with the full team: fundamentals, ethics, and responsible use.

• % of staff trained in critical AI (basic and advanced levels).
• AI policy approved and communicated institutionally.
• Number of privacy incidents or inappropriate AI use (baseline and tracking).
• Vendor evaluations completed before contracting new tools.