Autonomous Permit Processing for City Government
How KXN deployed a GDPR-compliant agentic AI system for a major European city government, reducing building permit processing from 47 days to 13 days and eliminating a 12,000-permit backlog.
The Challenge
A major European city government's planning department was drowning. Building permit applications — for residential extensions, commercial developments, and infrastructure projects — required sequential manual review across 14 separate departments. Zoning. Structural. Environmental. Fire safety. Heritage. Traffic. Each department had its own inbox, its own backlog, its own processing timeline.
The result: a 47-day average approval cycle and a backlog of 12,000 outstanding permits growing by 300 applications per month.
Key Constraints
- Sovereignty: All processing must occur within the EU; no data may leave the jurisdiction
- GDPR Compliance: Applicant data requires strict access controls and full audit logging
- Legacy Integration: Must connect to a 15-year-old permitting database with no public API
- Transparency: Citizens and developers must be able to see the exact status of their application at every stage
The Solution: Multi-Department Orchestration Agent
KXN designed a BPMN-based agentic orchestration layer that wraps the existing 14-department approval workflow without replacing any department's internal tools or processes.
Architecture Overview
-
Ingestion Agent: Receives permit applications via a web portal and classifies them into one of 12 permit types, extracting structured data from submitted PDFs and site plans.
-
Parallel Routing: Rather than sequential sign-off, the orchestrator dispatches review tasks to all relevant departments simultaneously. A zoning review and a fire safety review no longer have to wait for each other.
-
Specialist AI Reviewers: Three AI agents pre-screen applications before human review:
- Zoning Compliance Agent — cross-references proposed plans against the city's GIS zoning map
- Structural Code Agent — validates building dimensions and materials against current EN Eurocodes
- Environmental Impact Agent — flags applications within protected zones or flood risk areas
-
Exception Routing: Applications that fail any pre-screen check are routed immediately to a human reviewer with a structured exception report. Only 12% of applications require human review.
-
Citizen Portal: A real-time status dashboard showing applicants exactly which departments have approved, which are pending, and the estimated completion date.
Technology Stack
- Orchestration: LangGraph for stateful, multi-step agent workflows with full checkpointing
- Workflow Engine: BPMN process definitions for the 14-department approval graph
- Data Layer: PostgreSQL with row-level security for GDPR-compliant data isolation
- Infrastructure: Azure Government Cloud (EU sovereign region) for data residency compliance
- Legacy Bridge: Custom ETL connector to the existing Oracle-based permitting database
The Impact
Deployed over 18 months, the system went live with a phased rollout — starting with residential extensions (the highest-volume permit type) before expanding to all 12 permit categories.
By the Numbers
- 47 days → 13 days: Average permit processing time reduced by 73%
- 12,000 permits cleared from the existing backlog within 6 months of full deployment
- 88% of applications processed without requiring human review
- €4.2M annual savings in staff time reallocated from manual routing to complex case work
- NPS improved from 18 to 67 among developers and architects using the portal
Governance and Compliance
Every agent decision is logged to an immutable audit ledger. Citizens can submit a data subject access request and receive a complete record of every data processing event associated with their application — in full GDPR compliance.
The system has been audited by the city's data protection officer and received formal approval for processing special category data (heritage-listed properties).
"We eliminated a backlog that had accumulated for three years in under six months. Developers now get decisions in under two weeks — that's transformational for investment attraction." — Director of Digital Transformation, European City Government
Lessons Learned
The biggest implementation challenge was not technical — it was organisational. Each of the 14 departments had grown accustomed to controlling their own queue. The key unlock was designing the system so that department workload decreased (only complex cases arrive at reviewers' desks) rather than departments feeling their role was being automated away.
Change management, not AI architecture, determined the project timeline.