1. Local government HR: the ideal pilot case
The need. Six people in the HR department of a French municipality spend 40% of their time searching for information in statutes, agreements, DGCL/DGAFP circulars (national HR oversight bodies), and internal memos.
What works. An internal assistant, scoped strictly to the HR team, answers technical questions with citations from the source document. Measured gain: −55% search time.
What fails. Trying to load individual employee files into the same knowledge base. Too sensitive — these need to sit in a dedicated assistant with separate GDPR framing.
More details in A city hall simplifies its HR function.
2. Urban planning: the eligibility check
The need. The urban planning department receives 50 requests per week, 70% of which ask the same question: “can I do this on my plot?”
What works. An assistant fed with the local zoning plan (PLU), zone regulations, and the latest municipal orders. The resident enters their address, the assistant returns the applicable zone, the main constraints, and the required documents.
What fails. Trying to replace the case officer. The assistant provides orientation, never a final ruling. Every request then goes to a human officer who validates, completes, or contradicts.
3. Civil registry: the document checklist
The need. A classic loop: a resident comes in, is told they’re missing documents, comes back. Multiply by 200 per month.
What works. A 24/7 assistant that lists the exact documents required for each situation (nationality, marital status, reason for the request). It prepares the appointment, it doesn’t replace it.
What fails. Trying to process the request online without human review. Civil registry acts still require a public servant’s signature.
4. Public procurement: internal knowledge management
The need. The procurement officer needs to draft tender specifications based on precedents. Fifteen years of archives, poorly indexed.
What works. An internal assistant fed with the CCTP/CCAP (technical and administrative specifications) from the last five years plus recent case law. Drafts 80% of a new specification from a similar past one.
What fails. Assuming the assistant knows real-time case law. It only knows what you loaded into it. External legal monitoring remains a separate job — see also our AI glossary.
5. Libraries and culture: reading recommendations
The need. Librarians spend a significant share of their time guiding patrons (“do you have something like…”).
What works. An assistant fed with the collection catalog and the librarians’ professional notes. Personalized answers with citations from detailed catalog records.
What fails. Trying to replace human advice. Part of a library’s appeal is the relationship with its staff. The assistant complements them, on shifts when no librarian is available.
Cross-cutting takeaways, after eighteen months in local government
These results assume rigorous upstream framing: clear scope, controlled ingestion, systematic human escalation, regular audit. Without these guardrails, the project drifts within eight months — that’s what separates a successful deployment from a press release.
Step-by-step in Practical guide: deploying AI in your local government.
Twenty minutes to identify the most profitable pilot case in your local government. We look at the organization, the volumes, and scope the GDPR framing.
Book a demo→