How is the Haute-Garonne Department of
s using AI
to support its employees?

With nearly 7,000 employees, 175 job categories, and approximately 200 locations, the Haute-Garonne Departmental Council (CD31) has chosen to establish a structured approach to AI before expanding its use cases. This involves developing a strategic framework, implementing a mandatory AI certification program, establishing cross-functional oversight, and deploying AI agents on the sovereign Wikit Semantics platform—all designed to serve employees rather than users.

In this webinar, Jean-François Bonnans, a project manager in the Department’s Information Systems and Digital Technology Division, discusses the approach, real-world use cases, and lessons learned from the field, alongside Yvan Levacher, Head of Sales at Wikit.

Key Takeaways from This Webinar

• An AI project that is 80% organizational: cross-functional management (DSIN, senior management, legal, HR, DPO, CISO, environmental, and communications) rather than a purely technical project.
• A framework before the tools: an AI doctrine, best practices for use, and an AI Pass—a mandatory common training foundation for all employees.
• A clear ethical guideline: no AI is directly exposed to users; all AI agents serve the Department's staff.
• 3 AI agents in production: first-level IT support, assistance to receptionists in the transportation department, and legal monitoring of public procurement.
• A 96–98% success rate for the automatic extraction of data from annual performance reviews of middle school staff.
• A single platform for standardizing AI use cases and avoiding a proliferation of incompatible tools.
• 3 key tips: help agents adapt to the new environment, ensure data quality, and promote each new tool.

The department's approach: establishing a framework before deploying AI

Driven both by the IT Department’s monitoring efforts since 2022 and by a policy initiative from the Department’s president, the initiative is built on four pillars:

• Cross-functional leadership: The DSIN takes the lead, but with numerous partners—including the Environmental Division—to ensure alignment with the Department’s ecological transition commitments.
• A framework and best practices: a strategic framework for how to use and monitor AI, broken down into concrete guidelines (training, verifying AI results, etc.).
• The AI Pass: a common, mandatory training foundation for understanding what AI can and cannot do, supplemented by optional modules developed in collaboration with the CNFPT.
• An ethical approach: elected officials have ruled out exposing AI directly to users, as the Department works extensively with groups for whom human contact remains essential.

This isn't a technical project. It's really 80% organizational.

Choosing a Platform: 4 Non-Negotiable Criteria

In selecting its platform, the Haute-Garonne Department relied on principles derived from legal doctrine:

1. Sovereignty: Only French solutions tailored to the public sector.
2. Autonomy: Business administrators must be able to update the knowledge bases themselves.
3. Integration with existing systems: the EasyVista ITSM tool and the SharePoint environment.
4. Security: Data and Model Hosting, No Reuse

The Wikit Semantics platform was thoroughly reviewed and approved by the Department's DPO and CISO prior to its deployment.

AI Use Cases in the Haute-Garonne Department

The IT department identified the first use cases based on three criteria: high demand on staff, an existing documentation base, and repetitive questions. Then, as word of the initiative spread internally, several departments approached the IT department on their own to discuss their needs.

AI Use Cases in Production

• IT Support Agent: Using nearly 800 troubleshooting guides, the agent helps other agents resolve their incidents. If necessary, the agent creates a ticket in EasyVista and attaches the conversation history to it.
• Transportation receptionist: He helps summer temporary workers answer the many questions about school transportation passes.
• Legal Research Specialist: The Public Procurement Division can retrieve information from its monthly monitoring report in just a few seconds.

AI Use Cases in Construction and Exploration

• HR representatives handling work hours and travel expenses (about 40 calls per day to the teams).
• General Receptionist and Crisis Management Officer: to gain immediate access to procedures and the strategic directory.
• A data extraction tool for the annual performance reviews of 1,200 middle school staff members, built using Wikit Studio, with a success rate of 96 to 98 percent.
• A legal research tool connected to Légifrance and a personal, independent assistant for all employees.

The use cases we're implementing are modest, but they work.

Challenges and Lessons Learned by the Department

• The human factor: There is a fear of being replaced by AI. Training and cultural adaptation are the answer.
• The organizational factor: Uploading documents takes five minutes; the real work lies in the quality and structure of the data. Then, each AI agent must be promoted so that it is adopted.
• The technical aspect: Some complex formats (PDFs that combine text, diagrams, and organizational charts) need to be revised.

Unexpected result: While testing the HR agent, the Department identified inconsistencies in its work-time regulations. The AI gave incorrect answers… because the source document contained errors.

Importing data is quick. The real work lies in the quality, structure, and relevance of the data.

Key Success Factors for an AI Project in a Local Government

1. Establishing a framework before implementing tools: a guiding principle, best practices, and shared leadership among the CIO, HR, Legal, the DPO, and the CISO.
2. Educate staff: demystify AI. It helps us do what we already know how to do even better.
3. Choosing the Right Initial Use Cases: Recurring Questions, High Demand on Teams, and Existing Documentation.
4. Data management: structuring, cleaning, and removing duplicates and inconsistencies. One topic, one document, one up-to-date version.
5. Spread the word: Communicate about each new AI agent and get users on board so it doesn't just sit on the shelf.
6. Rely on a single platform: centralize AI agents to avoid a proliferation of incompatible tools.