AI in Public Administration

Some 660,000 people live in Düsseldorf. They register vehicles, apply for documents, look for available appointments, ask which department is responsible or need to know which paperwork to bring. Many of these inquiries recur. Yet each one matters to the person making it. Residents expect an administration that is accessible, communicates clearly and provides guidance even when the relevant department is not immediately available. 

Düsseldorf was well placed to meet those expectations. The city was open to AI, had already gathered initial ideas across several departments and had identified relevant entry points within its existing digital strategy. The next step was clear: individual initiatives needed to be brought together within a common framework – one that could turn ideas into practical applications while establishing clear guardrails. 

Working with MHP, the City of Düsseldorf developed an AI strategy in less than six months, focusing on citizen services and process support. The objective was deliberately pragmatic: quickly assess the city’s current level of AI maturity, prioritize relevant use cases and prepare an initial pilot that could provide a robust foundation for further applications. By combining strategy, governance and delivery, Düsseldorf created the conditions for AI to generate real value in public administration and perform reliably in day-to-day use. 

The clearest proof is already live: an AI-enabled voicebot for vehicle registration that answers standard inquiries around the clock and noticeably reduces the burden on staff. 

Düsseldorf was not starting from scratch. Several departments had already explored how AI could ease administrative workloads, while initial pilots and ideas demonstrated a clear appetite for innovation. At the same time, a familiar pattern had emerged: when each department works independently, parallel approaches quickly proliferate. Knowledge remains siloed, similar problems are solved more than once, and successful solutions become harder to replicate elsewhere.  

For the city, the next step was therefore one of organizational maturity. AI needed to be embedded within the existing digital strategy and used to shape specific areas of action. Two priorities stood out: improving services for residents and providing better support for internal processes. 

The timetable was tight. Between January and May 2025, MHP was to help the city develop a strategic framework, evaluate potential use cases and prepare an initial application for piloting. A lengthy strategy exercise would have added little at this stage. Düsseldorf needed to identify quickly where AI could make a tangible difference in everyday work. 

Municipalities are also under growing pressure to make digital services available more quickly and meet the requirements of Germany's Online Access Act (Onlinezugangsgesetz). Residents expect straightforward digital access, while data protection, technical reliability and public trust set a high bar for implementation. 

This presented the city with a clear choice. It could continue to allow individual initiatives to develop and attempt to align them at a later stage. That path would have required less coordination in the short term, but it would have made it considerably harder to transfer successful approaches over time. 

Düsseldorf chose the more ambitious route: a shared framework that considered value creation, governance, and reusability together from the outset.

Embedding AI in public administration under real-world conditions 

Municipal AI applications must integrate with existing IT environments, meet legal requirements and gain the acceptance of both employees and residents. 

The central challenge of municipal AI lies not only in the technology, but also in achieving a viable implementation. What matters is how an initial concept becomes a dependable service that integrates with existing structures and performs consistently in everyday use. 

This is where MHP applied a structured methodology that reduced complexity and created a clear path from concept to implementation. 

1. Assessing AI maturity
   MHP began with a structured assessment across selected departments. Through interviews and workshops, the team analyzed how AI was already being used, which initiatives were underway and where concrete action was required. Based on the findings, MHP developed targeted recommendations.
2. Prioritizing use cases
   An opportunity assessment followed. Working with the relevant departments, MHP collected, structured and prioritized potential use cases. Each idea was assessed against three central questions:
   1. Could it be tested in the short term?
   2. Would it deliver a visible benefit to residents?
   3. Could it provide a foundation for further applications?
3. Selecting a pilot with visible impact
   This assessment brought vehicle registration into focus. Düsseldorf’s central information line received a particularly high volume of inquiries on the subject, many of them concerning the same recurring questions. 
   How do I register a vehicle? Which documents do I need? What are the rules for deregistration? Where can I find the appropriate appointment?
4. Designing and integrating the voicebot
   Delivery followed a pragmatic model. MHP led the concept, content and service design. An integration partner connected the solution to the existing phone system, while a platform provider supplied the sandbox environment. This made it possible to build and test the voicebot in a controlled environment before connecting it to the live telephone system. The service went live in October 2025. 

MHP prepared the subject-matter content, defined the voicebot's persona, designed the dialog flows, developed guardrails for the underlying large language model and conducted testing and reporting. The bot was designed to answer frequently asked questions reliably, recognize when it had reached its limits and hand complex inquiries over to a human agent.

The voicebot now answers standard questions about vehicle registration around the clock. Residents can obtain information directly, regardless of office hours and without waiting on hold. When an inquiry is too complex, the caller is transferred to a service agent. 

The impact is clear: the voicebot handles around 50 percent of inquiries without any transfer. Düsseldorf’s central information line is relieved of routine call volume, staff gain more time for cases requiring human judgment, and residents receive guidance more quickly. 

Internally, Düsseldorf positioned this AI use case as practical support for daily work. By intercepting repetitive inquiries, the solution gives employees greater capacity to focus on more complex cases. 

Externally, the city introduced the voicebot through clear and transparent communications, explaining that the service would continue to be refined. After some initial hesitation, acceptance increased. Residents became more willing to try the bot, remain in the automated interaction for longer, and seek direct human assistance less frequently. 

The experience gained from the pilot provides a foundation for assessing further conversational-AI applications. In any future implementation, the interaction between several factors will remain critical: content quality, data protection, IT integration and user acceptance must all work together. 

Project facts 

• Timeframe: AI strategy development, maturity assessment, and use case prioritization took place from January to May 2025 
• Pilot: AI-supported voicebot for vehicle registration 
• Go-live: October 2025 
• Impact: Around 50 percent of inquiries resolved without transfer 
• Service effect: 24/7 information on common vehicle registration questions 

Responsible partner: Tolga Talas 
Project lead: Niklas Kroll 

Düsseldorf's approach addresses precisely the point many administrations have now reached: initial AI ideas exist, but the real work begins when those ideas have to be turned into reliable services. 

Düsseldorf demonstrates one viable route: understand the organization’s AI maturity, prioritize potential use cases and select an area in which the impact can become visible quickly. Governance and acceptance management should run in parallel from the outset. The potential to transfer a pilot to other areas should be considered at an early stage. 

Getting started does not require a large-scale program. A focused assessment and one high-visibility pilot can provide a robust test of value, acceptance and potential for wider adoption. 

Düsseldorf also strengthened its position in the Smart City Index, rising from 17th place in 2024 to sixth in 2025. Its AI initiatives form part of a broader digital transformation agenda and reinforce the city’s profile as a forward-thinking municipality. 

Smart City Index 2025: Düsseldorf climbed 11 places, from 17th in 2024 to sixth in 2025 – driven in part by an effective AI strategy. Source: Bitkom 

From strategy to delivery with MHP 

MHP helped Düsseldorf turn its openness to AI into a robust implementation pathway. This work encompassed the maturity assessment, structured use case prioritization, governance and risk assessment, stakeholder management, and the conceptual design of the voicebot pilot. 

MHP’s contribution was to bridge the gap between strategy and delivery. The city gained a solid working foundation for implementing specific applications. 

MHP also brought an industrial approach to scaling: structuring complexity, equipping decision-makers, measuring impact and designing successful solutions so that they can be reused and extended across other areas. 

Lessons from Düsseldorf for the public sector

1. Put governance first
   AI in public administration requires clear guardrails. Data protection, the EU AI Act, clear accountability and user-acceptance considerations should all be addressed early. This reduces the risk of costly corrections at a later stage.
2. The right pilot builds trust
   A pilot has the greatest impact when it solves a genuine everyday problem. Vehicle registration was selected deliberately: it involved a high volume of inquiries, recurring questions, and a visible benefit for residents.
3. Legacy systems are not future-proof
   IT systems that have evolved organically over time can constrain the integration and scaling of new AI applications. Dependencies, interfaces and modernization requirements should therefore be identified and made transparent during the pilot phase.
4. Build for scale and reuse from day one
   Shared standards, reusability and a common target architecture lay the groundwork for extending AI into other areas. From the outset, Düsseldorf treated the voicebot as part of a broader strategic vision.