Identified needs
Lanbide, the Basque Employment Service, set itself the goal of ensuring transparent, ethical and secure use of Artificial Intelligence (AI) in its business processes and decision-making. In a context of increasing regulation and a citizenry growing ever more demanding in terms of digital trust, the organisation required a comprehensive approach to ensure regulatory compliance and consolidate credibility in the use of AI in the public sector.
Before expanding its scope towards artificial intelligence, Lanbide had already developed a solid foundation in data governance, implemented in Anjana Data Platform, with the following key capabilities:
- Role and permission-based operating model — Clear definition of responsibilities through validation workflows aligned with RACI matrices and the DAMA-DMBOK2 philosophy, ensuring custody and traceability of data and information assets.
- Data and information asset catalogue — Structured inventory covering datasets, reports, indicators, dimensions, business rules and quality, providing an organised framework for metadata management and governance.
- Unified business glossary — A common language within the organisation that facilitates understanding, interpretation and use of data across different areas of the entity.
- Data traceability and lineage — Mechanisms to track the journey of data from its origin to its final consumption, ensuring transparency, auditability and trust in the information used.
The main challenges identified in the area of AI governance were:
- Regulatory compliance and strategic alignment — Adapting to the EU AI Regulation (AI Act) and the Ethical Data Manifesto for the Basque public sector (CAE).
- Algorithmic risk management and mitigation — Identifying, assessing and controlling risks associated with the use of AI models in decisions affecting citizens.
- Transparency and accountability — Ensuring that the use of AI is explainable, auditable and traceable to citizens and regulatory bodies.
- Ethics and fundamental rights — Ensuring that AI systems respect people's rights and do not generate bias or discrimination in automated decisions.
Current and future architecture
Currently, Anjana Data Platform instances at Lanbide are deployed in On-Premise mode on EJIE infrastructure (Basque Government IT Society), integrated with ODS and DWH on Oracle and with Hive views on EJIE's BDaaS platform (Cloudera). EJIE is migrating its BDaaS architecture to AWS (BatData), and plugins for integration with Anjana Data Platform are already under development.
Use case
Lanbide aims to lead AI governance in the public sector with an innovative approach that goes beyond traditional practices.
Pillar 01
Governed AI inventory
Integrated with the data governance platform for precise control of models and data.
Pillar 02
Continuous updates
Going beyond the CAE catalogue, which lacks proactive maintenance mechanisms.
Pillar 03
Advanced search and traceability
Structured navigation with detailed filters across models, datasets and reports for effective oversight.
Pillar 04
Structured operating model
Definition of roles and validation workflows to ensure quality and regulatory compliance throughout the model lifecycle.
Goals
To address these challenges, Lanbide has defined the following strategic objectives:
- Implement a robust AI governance framework, aligned with the principles of the Ethical Data Manifesto for the Basque public sector (CAE).
- Deploy a governed inventory of algorithmic models and AI systems, ensuring traceability, auditing and oversight.
- Support the public register of algorithms and AI systems, in line with the Basque Parliament initiative.
- Ensure compliance with the EU AI Regulation (AI Act), assessing existing models and adapting the necessary processes.
- Drive AI awareness and training within public administration, promoting a culture of responsible governance.
- Serve as a reference for other public administrations, sharing Lanbide's experience and methodology in AI governance.
Benefits achieved
The Governed Inventory of Algorithms and AI Systems at Lanbide delivers key improvements over the CAE Catalogue:
Standardisation and homogenisation
Structured templates with validations, avoiding inconsistencies and ensuring regulatory alignment. Pre-configured metadata versus the CAE's free-text PDF format.
Always up-to-date information
Continuous update mechanism to ensure accurate, real-time data. Unlike the CAE Catalogue, it avoids outdated model owners and stale information.
Advanced search
Filtering by any attribute and navigation from datasets, reports or business terms. Far more powerful than the CAE Catalogue.
Operating model with governance
Roles, validation workflows and traceability to ensure effective oversight of AI models.
Comprehensive governance of model and data
Not only are AI models governed, but also the data that feeds them, helping to detect and mitigate biases from the source.
Challenges overcome
Implementing the Governed Inventory required overcoming a range of both technical and organisational challenges. Key to success: collaboration with certified partners such as Telefónica and Desidedatum, and advisory support from Gartner (a platform recognised in the Magic Quadrant for Data & Analytics Governance Platforms).
Fragmentation and heterogeneity of documentation
The organisation moved from unstructured, hard-to-maintain formats to a standardised system with templates and validations that guarantee inventory consistency.
Outdated information
A continuous update mechanism was established to ensure accurate, real-time data, eliminating the obsolescence of model information.
Difficulty in traceability and auditing
An operating model with roles, validation workflows and traceability was implemented, ensuring effective oversight and the ability to audit any algorithmic decision.
Regulatory compliance and adaptation to the AI Act
A thorough analysis of existing models was carried out to align them with European regulations, establishing risk classifications and appropriate controls in accordance with the AI Regulation requirements.
Internal awareness and training
Staff were trained in the responsible use of AI, fostering a governance culture that makes every team member a guardian of ethical and regulatory principles.