AI as the key to administrative modernization: A practical step-by-step model for public administration
Introduction
The debate on administrative modernization is gaining momentum and importance. This can be clearly seen in the establishment of the BMDS, in analyses by the National Regulatory Control Council (NKR) on speeding up procedures and - most recently - in a new book by Patrick Bernau entitled "Bürokratische Republik Deutschland". The debate is not just about pure efficiency gains - it is about the acute shortage of skilled workers, a profound cultural change and nothing less than regaining the state's ability to act.
Artificial intelligence (AI) is now an important piece of the puzzle in this growing debate. Especially with regard to internal efficiencies and individualized communication with citizens, (almost) everyone involved believes that AI has enormous potential. It is clear that the state must take advantage of the possibilities offered by AI and keep up with the much older (and even more complex) digitization of administration.
This article is dedicated to the concrete practical implementation of AI in the public sector and offers a step-by-step model based on our technical and organizational experience that can be read as a step-by-step guide. For over 5 years, we at Merantix Momentum have specialized in supporting our clients on their AI journey and have helped shape some of the most exciting AI initiatives for the German public sector in recent years. In this first blog post, we will only focus on the most important technical solution components and use cases. In another blog post to follow, we will look at the structural, organizational and human challenges of AI change.
Greatest added value for the administration
The implementation of the following steps does not necessarily have to take place in this order, but can be parallelized. The sequence corresponds to the typical steps of the AI maturity levels that we see at our customers (public and private sector). It is also a sensible sequence from a change management perspective, because the intervention in the core processes and workflows increases with each step.
Stage 1: Introduce LLMs and Enterprise GPTs
The first step is to introduce legal, secure access to a large language model for each employee that is adapted to the administration. Experts speak of Enterprise GPTs, you could also say: ChatGPT for administration. There are various reasons why the rollout is easier compared to other AI products: The main reason is that there are already solutions developed specifically by and for the administration.
The goals are manifold: many small micro-facilitations result in a large efficiency gain, every employee should have their first contact with AI, reduce fears and recognize the added value, the shadow use of ChatGPT is minimized. A successful rollout focused on added value and user acceptance forms the technical, but above all organizational and structural foundation for all further AI stages.
There are already many successful examples and experience reports: F13 from Baden-Württemberg, LLMoin from Hamburg, NRW Genius, MUCGPT, KIPITZ. We helped develop LLMoin, which is already in use in five federal states, as AI product managers.
Stage 2: Connect AI research assistants and data records
As soon as an LLM is introduced, requests such as "Can we link the search to our internal data?" or "Can I develop an assistant that checks my reports for gender equality?" are automatically made. Since at least 40% of LLM usage is for information searches and document processing, the use cases for internal data integration are quickly obvious. A RAG (Retrieval Augmented Generation) research assistant can quickly add value, especially for complex research tasks, legal work and for offices with a lot of contact with citizens. These systems often replace outdated CMS systems or drives based on keyword searches.
Concrete examples that we have already implemented at MXM include research assistants (for precise law searches), connecting internal wikis, mini chatbots, automating text checks and spelling corrections, converting texts into simpler language and answering citizens' queries automatically.
Level 3: Agent workflow automation
This is where we start to automate the core processes of administration and thus leverage the real potential. We are not aware of any administration that is already implementing a strategic approach here. However, there are already isolated projects that use agent-based systems.
In the coming years, AI-centered workflow automation tools (UIPath, Zapier, n8n, or in-house developments) will find their way into administrative processes and take over a wide variety of routine tasks: e.g. checking forms for completeness, processing applications, or drafting citizens' inquiries. It is important that employees always make the critical decisions, but are relieved of initial and downstream routine and hard work.
A concrete example of workflow automation: At municipal and district level, thousands of funding applications are submitted every year, which are then evaluated against local funding guidelines and summarized for the district assembly. We recently showed at an internal hackathon that we can achieve almost perfect results by splitting the complex distribution into several LLM calls, which then only need to be checked by the experts.
Stage 4: Digitize specialist processes in their entirety and back them up with AI
This is the holy grail of administrative modernization! The day-to-day work of the employee at level 4 looks very different: Her entire work process is carried out on a standardized, digital solution, applications are received automatically and digitally, subsequent claims are processed automatically via an interface with the applicants, and at each step of the process, various AI agents provide support in processing applications and matters according to the HITL (Human in the Loop) principle.
Although as an organization you should go through the first stages from the bottom up, the direct, holistic digitalization and AI-fication of an entire specialist procedure - from the application to the draft administrative file - has a certain "greenfield" advantage. The vision currently seems like a distant dream, but it is not! There are already initial experiences and experiments on this topic.
- DiPlanung is a solution from Hamburg for complete and integrative process management in urban land-use planning, regional planning and planning approval.
- The BMI (and now the BMDS) are working on the development of AI components for planning and approval processes.
- for.ml is a start-up from Munich that automates some of the steps involved in processing housing benefit applications and is already in operation in several cities.
- The start-up Rule Mapping combines AI and rule logic to pave the way for legally compliant decisions by digital audit agents.
- A pilot study by the DZSF is investigating how AI-supported objection management could be used to handle objections in planning processes.
- The study "AI in specialist procedures: The key to modern administration" by Possible and Bundesdruckerei shows how scalable AI solutions could be across specialist procedures.
Closing words
We hope that this step-by-step guide with many references and examples has been helpful and can provide a technical perspective on the possible AI transformation in administration. It is important to emphasize: AI is not a stand-alone tool, not an end in itself and not a simple answer to problems that have evolved over decades. The risks and complex challenges involved in implementing Level 3 and Level 4 should not be underestimated: Data protection, quality assurance, bias mitigation, traceability, liability and co-determination must be part of the overall solution and change process. This is precisely why it is important to start the AI cultural change yesterday with level 1 solutions and thus lay the foundations for being able to digitize and "AI-fy" specialist processes holistically in a few years' time.
In our next blog post, we will look at the organizational drivers of AI change.
As an innovation driver and partner for piloting, scaling and productive introduction of AI solutions in administration, Merantix Momentum accompanies public organizations on the path from initial piloting to the holistic digitization of complex specialist processes. Talk to us at the SCCON!
Subscribe to the Merantix Momentum Newsletter now.
More articles
.png)
.png)
AI as the key to administrative modernization: A practical step-by-step model for public administration
Introduction
The debate on administrative modernization is gaining momentum and importance. This can be clearly seen in the establishment of the BMDS, in analyses by the National Regulatory Control Council (NKR) on speeding up procedures and - most recently - in a new book by Patrick Bernau entitled "Bürokratische Republik Deutschland". The debate is not just about pure efficiency gains - it is about the acute shortage of skilled workers, a profound cultural change and nothing less than regaining the state's ability to act.
Artificial intelligence (AI) is now an important piece of the puzzle in this growing debate. Especially with regard to internal efficiencies and individualized communication with citizens, (almost) everyone involved believes that AI has enormous potential. It is clear that the state must take advantage of the possibilities offered by AI and keep up with the much older (and even more complex) digitization of administration.
This article is dedicated to the concrete practical implementation of AI in the public sector and offers a step-by-step model based on our technical and organizational experience that can be read as a step-by-step guide. For over 5 years, we at Merantix Momentum have specialized in supporting our clients on their AI journey and have helped shape some of the most exciting AI initiatives for the German public sector in recent years. In this first blog post, we will only focus on the most important technical solution components and use cases. In another blog post to follow, we will look at the structural, organizational and human challenges of AI change.
Greatest added value for the administration
The implementation of the following steps does not necessarily have to take place in this order, but can be parallelized. The sequence corresponds to the typical steps of the AI maturity levels that we see at our customers (public and private sector). It is also a sensible sequence from a change management perspective, because the intervention in the core processes and workflows increases with each step.
Stage 1: Introduce LLMs and Enterprise GPTs
The first step is to introduce legal, secure access to a large language model for each employee that is adapted to the administration. Experts speak of Enterprise GPTs, you could also say: ChatGPT for administration. There are various reasons why the rollout is easier compared to other AI products: The main reason is that there are already solutions developed specifically by and for the administration.
The goals are manifold: many small micro-facilitations result in a large efficiency gain, every employee should have their first contact with AI, reduce fears and recognize the added value, the shadow use of ChatGPT is minimized. A successful rollout focused on added value and user acceptance forms the technical, but above all organizational and structural foundation for all further AI stages.
There are already many successful examples and experience reports: F13 from Baden-Württemberg, LLMoin from Hamburg, NRW Genius, MUCGPT, KIPITZ. We helped develop LLMoin, which is already in use in five federal states, as AI product managers.
Stage 2: Connect AI research assistants and data records
As soon as an LLM is introduced, requests such as "Can we link the search to our internal data?" or "Can I develop an assistant that checks my reports for gender equality?" are automatically made. Since at least 40% of LLM usage is for information searches and document processing, the use cases for internal data integration are quickly obvious. A RAG (Retrieval Augmented Generation) research assistant can quickly add value, especially for complex research tasks, legal work and for offices with a lot of contact with citizens. These systems often replace outdated CMS systems or drives based on keyword searches.
Concrete examples that we have already implemented at MXM include research assistants (for precise law searches), connecting internal wikis, mini chatbots, automating text checks and spelling corrections, converting texts into simpler language and answering citizens' queries automatically.
Level 3: Agent workflow automation
This is where we start to automate the core processes of administration and thus leverage the real potential. We are not aware of any administration that is already implementing a strategic approach here. However, there are already isolated projects that use agent-based systems.
In the coming years, AI-centered workflow automation tools (UIPath, Zapier, n8n, or in-house developments) will find their way into administrative processes and take over a wide variety of routine tasks: e.g. checking forms for completeness, processing applications, or drafting citizens' inquiries. It is important that employees always make the critical decisions, but are relieved of initial and downstream routine and hard work.
A concrete example of workflow automation: At municipal and district level, thousands of funding applications are submitted every year, which are then evaluated against local funding guidelines and summarized for the district assembly. We recently showed at an internal hackathon that we can achieve almost perfect results by splitting the complex distribution into several LLM calls, which then only need to be checked by the experts.
Stage 4: Digitize specialist processes in their entirety and back them up with AI
This is the holy grail of administrative modernization! The day-to-day work of the employee at level 4 looks very different: Her entire work process is carried out on a standardized, digital solution, applications are received automatically and digitally, subsequent claims are processed automatically via an interface with the applicants, and at each step of the process, various AI agents provide support in processing applications and matters according to the HITL (Human in the Loop) principle.
Although as an organization you should go through the first stages from the bottom up, the direct, holistic digitalization and AI-fication of an entire specialist procedure - from the application to the draft administrative file - has a certain "greenfield" advantage. The vision currently seems like a distant dream, but it is not! There are already initial experiences and experiments on this topic.
- DiPlanung is a solution from Hamburg for complete and integrative process management in urban land-use planning, regional planning and planning approval.
- The BMI (and now the BMDS) are working on the development of AI components for planning and approval processes.
- for.ml is a start-up from Munich that automates some of the steps involved in processing housing benefit applications and is already in operation in several cities.
- The start-up Rule Mapping combines AI and rule logic to pave the way for legally compliant decisions by digital audit agents.
- A pilot study by the DZSF is investigating how AI-supported objection management could be used to handle objections in planning processes.
- The study "AI in specialist procedures: The key to modern administration" by Possible and Bundesdruckerei shows how scalable AI solutions could be across specialist procedures.
Closing words
We hope that this step-by-step guide with many references and examples has been helpful and can provide a technical perspective on the possible AI transformation in administration. It is important to emphasize: AI is not a stand-alone tool, not an end in itself and not a simple answer to problems that have evolved over decades. The risks and complex challenges involved in implementing Level 3 and Level 4 should not be underestimated: Data protection, quality assurance, bias mitigation, traceability, liability and co-determination must be part of the overall solution and change process. This is precisely why it is important to start the AI cultural change yesterday with level 1 solutions and thus lay the foundations for being able to digitize and "AI-fy" specialist processes holistically in a few years' time.
In our next blog post, we will look at the organizational drivers of AI change.
As an innovation driver and partner for piloting, scaling and productive introduction of AI solutions in administration, Merantix Momentum accompanies public organizations on the path from initial piloting to the holistic digitization of complex specialist processes. Talk to us at the SCCON!
.png)
.png)