This article is authored by Dirk Günnewig, State Secretary of Finance of the State of North Rhine-Westphalia.

Artificial intelligence in public administration is not an optional add-on. North Rhine-Westphalia's tax administration provides an answer to this: comprehensive AI integration across 33,000 employees, combining automated risk assessment and generative AI tools with rigorous human oversight, extensive training programs, and sovereign data infrastructure.

In the 21st century, public administrations face profound structural, procedural, and demographic shifts. At the same time, they are confronted with overlapping crises and challenges with far-reaching economic and societal implications. In parallel, technological opportunities — often emerging at breathtaking speed — must be strategically identified, rigorously assessed, and responsibly leveraged. This convergence creates strong pressure to act and places significant demands on government structures, processes, and personnel. Given limited human and financial resources, rising public expectations of administrative performance, and the growing complexity of legal frameworks, the adoption of new technologies is becoming a central priority of government action.

In this context, the use of artificial intelligence (AI) is not an optional add-on but a core element of modern public administration. Its purpose is to markedly improve the quality of services for citizens and businesses, enhance the public sector’s appeal as an employer, and make government action more effective, efficient, and economical. AI can act as a strategic lever and evolve into a key technology for safeguarding the long-term viability of public administration.

The challenges of government action outlined above — and the opportunities arising from technological innovation such as AI — are particularly evident in the financial administration. 

This article focuses on the tax administration in North Rhine-Westphalia, which is responsible for the legally compliant, consistent, and timely collection of state revenues in Germany’s most populous federal state, home to 18 million people. With approximately 33,000 employees, the tax authorities of North Rhine-Westphalia are essential to ensuring the state’s fiscal capacity. Located at the heart of Europe and bordering the Netherlands and Belgium — with major cities such as Cologne, Düsseldorf, and Essen, and economically robust regions integrated into global trade — North Rhine-Westphalia holds significant geopolitical relevance and contributes substantially to Germany’s overall tax revenue.

In an increasingly digitalized society, it is essential to safeguard efficiency, service quality, employer attractiveness, and the rule of law over the long term — even under dynamic and changing conditions. The tax authorities in North Rhine-Westphalia are assuming a pioneering role in this regard: they are testing and integrating AI into concrete, scalable use cases and thus serve as a model for a modern, efficient, and future-oriented public administration in the digital age.

Strategic Frame: Objectives, Benefits, and Guiding Principles for the Use of AI

Through the use of AI, the North Rhine-Westphalian tax administration aims not only to address structural and procedural challenges, but also to transform them into drivers of modernisation and instruments of active administrative design.

The use of AI is embedded within the comprehensive and structured modernisation initiative Financial Administration for North Rhine-Westphalia. It combines a focus on efficiency, service orientation, and digital resilience. Its objectives are to optimise administrative structures, enhance processes, strengthen personnel development, accelerate digital transformation, and ultimately deliver better, faster, and more citizen-oriented services.

Within this strategic framework, AI functions as both an enabler and an accelerator of change — helping to shape a public sector that is more resilient, efficient, and people-centred.

The core objectives for the use of AI can be structured as follows:

1. Efficiency gains and workload relief: AI is to be deployed with priority in areas where repetitive routine tasks, large volumes of data, and complex administrative processes place a considerable burden on specialist staff. In tax administration in particular, the handling of numerous procedures requires qualified personnel to identify irregularities, assess risks, and make nuanced decisions. The objective is to free up human resources for demanding, non-standardisable tasks and thereby ensure processing times remain fast and reliable. This enhances both efficiency and cost-effectiveness.

2. Quality improvement: The controlled use of AI enhances the precision and consistency of administrative decision-making. Automated analyses and standardised verification mechanisms help reduce sources of error and harmonise decision-making processes. When applied systematically, AI contributes to a sustainable improvement in the quality of public services.

3. Service improvement: Digital interaction with citizens and businesses is to be accelerated and enhanced in quality through AI-supported systems. AI can expedite the preparation of tax assessments, optimise citizen services, and enable faster, more accurate responses to enquiries addressed to the tax authorities.

4. Safeguarding knowledge: AI-supported processes enable the structured capture, systematisation, and preservation of specialised knowledge. In doing so, they make an important contribution to counteracting demographically induced knowledge loss and help ensure effective knowledge transfer within the financial administration.

5. Fraud detection: AI can reliably identify irregularities within large volumes of data, thereby supporting the detection of potential fraud cases. By comparing and linking diverse data sources, it uncovers anomalies that would otherwise go undetected — or would require significant personnel resources to identify. This provides the foundation for more effective and resource-efficient risk assessment.

6. Resilience and future viability: The targeted use of AI strengthens the administration’s resilience to the dynamic challenges outlined above. It safeguards the long-term capacity to act while simultaneously enhancing innovation potential and increasing the administration’s attractiveness as an employer.

Key Elements of the AI Strategy of the North Rhine-Westphalia Finance Administration

These objectives are embedded in five strategic guiding principles that define the North Rhine-Westphalian approach to AI:

1: Responsibility, skilling, and the ‘human-in-the-loop’ principle

In the tax administration, AI is used as a supporting tool — not as a substitute for human professional expertise. Whenever AI is applied in high-volume processes, appropriate professional oversight and the ability to intervene must be ensured at all times. In a highly sensitive environment where personal tax data and economically relevant company information are processed under complex legal frameworks, one principle is non-negotiable: responsibility for administrative action remains explicitly with the qualified employees of the tax administration. To ensure optimal interaction between humans and technology, comprehensive training for staff is essential. Only those who understand what AI can — and cannot — do will be able to use it responsibly and effectively.

2: Technology openness, data security, and a multi-vendor strategy

The North Rhine-Westphalia tax authorities pursue a responsible, proven, and technology-agnostic approach to AI. A core element is a multi-vendor strategy in which various AI tools can be used — including commercial solutions as well as in-house developments such as NRW.Genius and open language models. This openness enables agile testing and further development of different technologies, always considering data protection, other legal frameworks, and efficiency potential. The protection level of the data to be processed determines which technology can be used. In areas with particularly sensitive data, regular reassessments are required — in the sense of crypto-agility — to ensure an appropriate level of protection. In contrast to classic linear planning models, the approach is deliberately cyclical, iterative, and agile, as rapid technological development makes long-term detailed planning obsolete. Each deployment is evaluated to transparently demonstrate concrete added value for service quality, administrative integrity, and legally compliant action.

3: Targeted solutions for specific challenges

AI is used as a targeted response to concrete challenges faced by the North Rhine-Westphalia tax authorities. It is not an end in itself but an instrument of responsible, needs-oriented modernisation.

4: Scalability and modernisation for the common good

The use of AI is an integral part of strategic administrative modernisation. The aim is not just to implement lighthouse projects, but to develop resilient, transferable solutions that can be applied across the financial administration — always navigating the tension between innovation, the rule of law, and a focus on the common good. Scalability is systematically considered from the outset.

5: Courage to experiment and an open culture of error

The use of AI requires the courage to tread unconventional paths and try new approaches. This also includes consistently discontinuing approaches that do not prove successful, learning from the findings, and thus using resources purposefully.

Taking into account its five strategic guidelines, the North Rhine-Westphalia tax authorities are pursuing an integrated, practical, and future-oriented approach with their AI strategy. The aim is to systematically develop both technological capability and organisational maturity.

The strategy is based on the following central fields of action:

1: Organisational and technical ‘AI readiness’:

A central prerequisite for the successful use of AI in public administration is the establishment of ‘AI readiness’ — that is, the ability of an organisation to provide suitable structures, processes, and resources not only to pilot AI projects, but to transfer them permanently and at scale into regular operations, while being prepared to seize forthcoming technological leaps as opportunities. The following measures have been implemented in the North Rhine-Westphalia tax authorities:

• Structural anchoring: Central topics such as AI and cloud have been bundled in a dedicated unit within the ministry. At the same time, an ‘AI Cloud Incubator’ was set up in the Finance Administration’s data centre (RZF). This unit enables protected testing of new technologies such as generative AI and public-cloud applications and transfers proven approaches into everyday administration.

• In-house AI hardware under construction: To strengthen digital sovereignty, safeguard data control, and ensure legal certainty with regard to the German Fiscal Administration Act and the Fiscal Code, the RZF is currently setting up an AI infrastructure in the city of Kaarst near Düsseldorf. In addition, cloud technologies are being tested, and new approaches such as confidential computing are being examined, enabling data to be encrypted during processing while ensuring compliance with tax secrecy and data protection. These measures strengthen the sustainable usability of AI in a sovereign and legally secure context.

2: Qualification and empowerment of employees

A responsible and strategic use of AI requires the targeted strengthening of employees’ professional and operational skills. The AI strategy of the North Rhine-Westphalian financial administration therefore places particular emphasis on comprehensive training and further education — aligned with Article 4 of the EU AI Act, which stipulates a qualification obligation for organisations using AI. This systematic qualification forms the foundation for informed, reflective, and responsible use of AI by all employees.

The NRW tax administration focuses on the following elements:

• A statewide, broad-based training initiative in partnership with GovTech Germany and Microsoft for all employees of North Rhine-Westphalia’s tax administration.

• A modular, manufacturer-independent learning system with digital learning content, practical examples, and interactive live formats.

• Continuous professional support for transferring learning into everyday work.

3: Governance, legal certainty, and trustworthiness

Beyond technical and organisational requirements, the introduction of AI in the sensitive area of tax administration also requires a solid legal foundation. The North Rhine-Westphalian AI strategy therefore pursues a governance structure that consistently combines the basic principles of the rule of law, trustworthiness, and transparency with capacity for innovation.

The financial administration’s AI strategy is characterised by deliberately restrictive yet innovation-friendly governance. It sets clear limits for the use of AI, creates reliable framework conditions, and at the same time allows room for open-ended testing and responsible innovation:

• Deployment only under clear framework conditions: A ministerial decree provides certainty for the 33,000 employees in the financial administration and enables safe, responsible AI use. In just under two pages, it sets binding rules on which AI tools may be used by employees of the North Rhine-Westphalia tax authorities and for which purposes. In particular, functions for research support, formulation assistance, and text analysis in non-critical business contexts without personal references are permitted. Any processing of personal data, tax-related circumstances, internal company information, or confidential content is strictly prohibited. Professional and legal responsibility expressly remains with employees. Results produced by generative AI must always be critically examined and rejected if necessary. The regulations already reflect the central requirements of the EU AI Act — in particular the principles of risk classification and human-centred control.

• Data protection and tax confidentiality: Compliance with data protection regulations and the protection of tax confidentiality have the highest priority. The handling of sensitive data is a key benchmark for any form of AI use in the tax authorities. The implementation of use cases must not jeopardize the confidentiality of personal or tax-related information at any time.

• Processing in protected environments: Generative AI systems for processing sensitive data are used exclusively in secure IT environments — for example, in specially isolated cloud segments or secure Azure tenants administered by the RZF. Other privacy-enhancing technologies such as confidential computing are also being tested. With confidential computing, data is encrypted not only during storage and transmission but also during processing. The relative anonymisation of data associated with confidential computing could also help in the future to rely on external resources for processing real tax data in the sense of an extended arm of the tax administration or its data centre. In this case, only the tax administration would hold the digital keys to access data and computations.

• Transparency and traceability: The North Rhine-Westphalia tax authorities attach particular importance to transparent decision-making channels and documented processes — especially for generative and analytical AI applications. The social acceptance of AI in administrative action depends largely on the traceability of government decisions. A central problem of analytical AI is the ‘black box’ phenomenon, i.e., the lack of transparency in the decision-making logic of complex models. In public administration, this is in tension with the obligation to justify sovereign decisions in a comprehensible manner at all times. The NRW tax authorities address this challenge by (i) using comprehensible models in sensitive contexts, (ii) documenting and tracking model versions and decisions, (iii) excluding sensitive AI models from automated individual decisions without human review (‘human-in-the-loop’), and (iv) involving data protection officers and staff representatives: all AI projects are coordinated in internal administrative committees and evaluated on an ongoing basis.

Specific Implementation Projects in the Tax Authorities of North Rhine-Westphalia

In implementing its AI strategy, the North Rhine-Westphalia tax authorities are pursuing a practical, implementation-oriented, step-by-step approach. All projects are developed iteratively, tested in scalable pilot phases, and systematically evaluated — always in compliance with legal, technical, and ethical requirements. The AI roadmap associated with the strategy focuses on four priority areas of application:

1: Infrastructure and digital sovereignty

A central strategic goal is the development of an AI-capable, sovereign IT infrastructure:

• AI Cloud Incubator at the RZF: The incubator at the RZF enables protected testing of generative AI technologies based on modern cloud architectures as well as on-premises solutions. The focus is on evaluating public-cloud solutions under conditions of public control and strict compliance with security and data-protection requirements, with particular attention to privacy-enhancing technologies such as confidential computing.

• In-house AI hardware at the city of Kaarst site near Düsseldorf: An independent infrastructure with specialised AI hardware is being set up at the new, energy-efficient data centre site of the North Rhine-Westphalia Finance Administration. The aim is to increase digital resilience and secure access to sensitive tax data independently of external infrastructures — an active contribution to the technological sovereignty of the tax authorities.

2: Generative AI: text, research, and assistance systems

In the field of generative AI, various systems are used for text processing, document analysis, and legal research:

• AI decree and handout on the use of chatbots: Since early 2025, the approximately 33,000 employees of the North Rhine-Westphalia tax authorities have been permitted the controlled use of generative AI systems. Their use is permitted for the creation, revision, and structuring of non-personal texts. Permitted areas of application include support in researching general legal issues, the use of formulation aids, the development of argumentation structures, and the creation or editing of internal memos, reports, and public texts. AI-supported image generation is also permitted. A ministerial decree defines the binding framework conditions. The input of personal data, information protected by tax or official secrecy, and other confidential content is expressly prohibited. Content generated by generative AI must always be checked for accuracy, legal admissibility, and compliance with data-protection law. Responsibility for use and further processing lies entirely with the respective employees.

• NRW.Genius: The tax authorities are actively involved in piloting the North Rhine-Westphalian AI assistance system NRW.Genius, developed by the state IT service provider IT.NRW. The system supports analysis and summarisation of texts as well as the preparation of notes at a general administrative level and is to be further developed as a sovereign, data-protection-compliant solution for administration-wide use.

• Juris AI: Since March 2025, an AI-supported research and text function has been available to employees in the Juris legal portal, with a knowledge base specifically geared towards tax law. The AI generates legally aligned suggestions without ‘hallucinated’ content. A particular advantage lies in the continuously updated database, which includes tax laws, case law, and practice-relevant manuals.

• Language Weaver Edge: In addition, AI-supported translation software (Language Weaver Edge) was introduced at the North Rhine-Westphalia tax authorities in December 2024.

3: Analytical AI: pattern recognition and risk management

In addition to generative AI, analytical systems with statistical and predictive capabilities are increasingly being used:

• AI extension in the Risk Management System (RMS): North Rhine-Westphalia is the first German federal state to test a machine-learning module for case forecasting in tax assessment. This module supplements the existing rule-based RMS with a statistical prediction model that estimates the probability of potential risks in the respective tax case. Based on historical assessment data, the system supports decisions on whether a case should be processed automatically (‘auto-case’) or manually. The aim is to identify low-risk cases more reliably, increase the number of automated assessments, and relieve employees of repetitive checking tasks. As part of KONSENS, the joint software development program of the German federal states, the RZF developed the system in collaboration with Bavaria and has been testing it in shadow operation since February 2024; a three-month pilot started in May 2025 in four tax offices in the area of traditional employee assessments. Following a successful pilot phase, a state-wide rollout in North Rhine-Westphalia and provision for other federal states is planned.

• AI at the State Office for Combating Financial Crime (LBF): The LBF, founded at the beginning of 2024, will use AI systems to analyse digital evidence and detect suspicious financial flows — especially in the context of terrorist financing. The use of AI is part of North Rhine-Westphalia’s security package to combat international extremism. In the first expansion stage, an AI-supported software solution is being implemented to identify structures and patterns connected to terrorist financing. There are plans to extend this to other areas of crime such as money laundering and organised crime. The state is cooperating with the Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) to develop domain-specific, robust models. The aim is to create highly specialised tools that effectively support investigations and specifically relieve LBF employees.

4: Empowering employees (‘AI Skilling Initiative’)

A central concern of the North Rhine-Westphalian financial administration is to enable employees to use AI safely, effectively, and responsibly. Developing appropriate skills is not only an expression of modern personnel development, but also a legal requirement under Article 4 of the EU AI Act: organisations that develop or use AI systems must ensure that their employees have sufficient AI skills. Empowering employees is an integral part of the AI strategy and roadmap of the North Rhine-Westphalian financial administration. The administration’s excellent training infrastructure makes it possible to provide 33,000 employees with e-learning courses via its training academy and digital teaching department within a very short time and to document completion directly in the personnel file.

• Large-scale training initiative: Together with GovTech Germany and Microsoft, the North Rhine-Westphalia tax authorities have launched a state-wide training initiative. All employees of the tax administration are offered targeted training on the basics, fields of application, and limits of AI — practical, vendor-neutral, and relevant to everyday work. The content is prepared as e-learning and made available via the administration’s own training systems. The focus is on imparting foundational AI knowledge independent of specific tools.

• Modular training system: The training offer includes digital learning modules, practical application examples, and, in the long term, accompanying support for transferring what has been learned into daily work. The aim is to anchor informed and responsible use of modern AI tools in the administration. The content is vendor-neutral and is being developed in collaboration with GovTech Germany, Microsoft, and the administration’s training specialists. The first e-learning courses are already available. Further content is tailored to typical tax-administration tasks — for example, researching tax-law issues or formulating expert opinions.

Conclusion and Outlook

The use of AI in the tax administration of North Rhine-Westphalia is no longer a vision of the future but a reality — from automated processing and legal research to data-based risk analysis. This demonstrates that efficient, trustworthy, and public-interest-oriented use of AI is possible when technological innovation is combined with strategic leadership, legal safeguards, and clear governance.

The analysis of the North Rhine-Westphalian approach allows for several transferable findings for the administrative-science discourse:

1. Strategic clarity and pragmatism: North Rhine-Westphalia’s AI strategy combines goals of technological sovereignty with concrete operational added value. It eschews ideological framing and is consistently designed for practical effectiveness, feasibility, and scalability — including with a view to federal or sectoral interoperability.

2. Integrated infrastructure and skills development: The combination of technical infrastructure — such as in-house AI hardware and an AI Cloud Incubator — and a comprehensive training initiative is key to anchoring AI systemically in the administration. The aim is not a one-off solution but the sustainable integration of AI into the work culture.

3. Governance and responsibility: Practice in North Rhine-Westphalia shows that effective innovation and stringent control are not contradictory. Clear rules and transparent procedures create trust in the use of algorithm-supported systems in the public sector.

4. People shape outcomes: Tax administration remains a knowledge-based system that relies on the judgement and experience of employees. AI changes processes — particularly by relieving routine activities — but does not replace responsibility and expertise. This tension is understood as a mandate for active design, not as a dilemma. The distinctive quality of North Rhine-Westphalia’s approach lies in the effective interaction between humans and technology. Responsibility cannot be delegated to an AI.

5. Perspective and further development: The role of AI in public administration will continue to evolve — driven by regulatory impulses such as the EU AI Act, progressive standardisation, and especially technological dynamics. For the tax administration, this implies continuous readjustment — technologically, legally, and culturally.

The North Rhine-Westphalia tax administration will continue to play an active role in this: as a practical testing ground for innovative applications, as a transfer partner for other administrations, and as a driving force for a digitally sovereign, learning administration of the future.

It follows an iterative approach based on the ‘human-in-the-loop’ principle: AI is used in a targeted manner to relieve routine tasks, creating scope for demanding specialist work and strengthening administrative efficiency.

Central challenges — such as protecting sensitive data or complying with tax confidentiality — are addressed through secure infrastructures (e.g., dedicated AI hardware in the administration’s data centre in conjunction with specially secured cloud resources) and binding guidelines (e.g., the AI decree and guidance on chatbot use). Comprehensive training for all employees forms the foundation for safe and competent use of AI in everyday work.

In 2030, human expertise will remain the backbone of an efficient tax administration. AI will change work, but it will not replace people. This balance between technological innovation and public responsibility is not a compromise; it is an expression of good, future-oriented governance. The financial administration of North Rhine-Westphalia will therefore continue to put AI into practice with ambition, diligence, and clear judgement.