A tailored AI workshop for municipal organisations, aligned with the AI Act, the BIO, the Algorithm Register obligation and the administrative reality of public decision-making. No generic training, no vendor pitch: a session run by people who build AI systems for the public sector and understand what an objection procedure or a Wmo (Social Support Act) application involves.
An AI workshop for municipalities is a half-day or full-day session in which we take a civil service organisation through the practical, legal and ethical side of AI in public work. We deliberately work with examples from municipal processes, such as WMO assessment, objection handling, permit pre-screening and council documents, rather than generic corporate use cases. For government, AI is a considerably stricter game than for private organisations: the EU AI package classes many public applications as high-risk by default, the Algorithm Register requires active transparency, and the Baseline Information Security for Government (BIO) sets strict limits on data processing.
We design every workshop around a municipality's current legal reality, not around a tool vendor's marketing. We spend as much time on the AI Act, the Algorithm Register and the DPIA obligation as on prompt engineering or ChatNL. We also say openly when an AI application in the public sector cannot pass legal muster, for example when a social-domain algorithm risks repeating the mistakes we know from SyRI and the childcare benefits scandal.
The second distinguishing point: we come from software development, not management consultancy. We build custom software for municipalities ourselves, and we know how BIO-compliant infrastructure is put together, how DigiD and eHerkenning integrations work, and what separates a production system from a prototype. Our examples come from code, not from a slide deck.
We build the content from seven blocks: an up-to-date picture of AI in 2026, the AI Act and the Algorithm Register for municipalities, BIO and GDPR overlap, bias evaluation for social-domain algorithms, a responsible AI framework, hands-on prompting for local policy tasks, and a closing roadmap. Which blocks receive the most attention depends on the audience. A session for the CIO office and legal affairs leans more heavily on compliance; a session for customer contact, communications or policy advisers leans more on hands-on and use-case work. For longer-running programmes we refer you to our AI implementation programmes, in which we follow up the workshop with guided pilots.
What we will not do: pretend that the AI Act is a voluntary framework. For public organisations, the time for "we'll see how it goes" is over — the legislator, the Dutch Data Protection Authority and the Court of Audit no longer accept that. A good workshop ensures that your organisation not only knows what AI can do, but also gains a clear understanding of the accountability chain that comes with it.
Since February 2025, Article 4 requires that all staff working with AI systems have sufficient AI literacy. For municipalities, this applies not only to CIOs and data specialists but also to civil servants in citizen services, the social domain, permits and communications. A targeted workshop is a pragmatic way to build that AI literacy for an entire department at once, and to demonstrate it on paper to any supervisory authority that asks. See also our standalone AI literacy training.
The Algorithm Register requires municipalities to publicly register high-impact algorithmic decision-making systems, with explainable information on how they work, the dataset used and the option to object. Many organisations do not yet know which existing systems fall under that obligation. The workshop gets that mapping started internally.
Your council sees neighbouring municipalities experimenting with chatbots, objection classification or generative AI for council documents, and wants to know where your own municipality stands. A management team or executive session provides the framework for strategic decisions, without councillors and aldermen first having to work out for themselves what the AI Act or the Algorithm Register means.
Benefit applications, youth care, debt assistance, enforcement: everywhere there are AI applications that seem logical, yet there are also lessons from SyRI, the childcare benefits system and municipal risk-profiling projects. A workshop with rigorous bias evaluation and public-values testing prevents a municipality from building itself into a new scandal. This is not a distant risk: it is the central test for every high-risk application.
We build custom software for municipalities and know the difference between a standard SaaS integration and a BIO-compliant architecture with DigiD/eHerkenning, archive integration and objection-procedure logic. Our examples come from real municipal work, not from a generic corporate sample set.
No partnership with Microsoft, OpenAI, Google or the large consultancy firms. We give you an honest picture of what ChatNL, Copilot, Claude or an open-source model such as Mistral mean for your situation, including data residency, vendor dependency, and whether a self-hosted solution is sometimes wiser than a hyperscale API.
What we advise is consistent with how we work ourselves: code ownership with the municipality, no vendor lock-in, transparent data flows. For regulators, councillors and auditors, that is a much easier story to tell than an AI running in a black box at an external party, especially when a citizen later asks for an explanation of a decision.
These blocks form the building blocks. We combine them depending on the target group, the format (half day, full day or multi-day) and the scoping conversations beforehand. No fixed script, but a solid substantive foundation.
The current state of LLMs, RAG, agents and multimodal models. What is hype and what is realistically deployable in a public-sector context. Includes an overview of ChatNL and EU-oriented models.
What the European AI Act and the Dutch Algorithm Register obligation mean in practice for your municipality. Risk classification, registration duties, transparency and human oversight.
How an AI experiment relates to the Baseline Information Security for Government (BIO). When an adapted DPIA is required. What a data protection impact assessment specifically for municipal AI involves.
Lessons from SyRI, the childcare benefits scandal and municipal risk profiling. How to test a social domain algorithm for proxy discrimination, and how to build a feedback loop with the target group itself.
Translate the concepts from the AI Act, the Algorithm Register and internal frameworks into a workable decision-making structure for your municipality: who approves what, at which point, and with which documentation requirements.
Practical work on official documents: policy papers, council reports, letters to citizens, draft decisions. Working with ChatNL and frontier models within what is permitted on your workplace systems.
To conclude, we design a first roadmap together: which pilots, which registrations in the Algorithm Register, which governance framework, which short-term investments and which longer-term lines.
A conversation with the CIO, legal affairs and the substantive lead. Which departments, which sensitivities, and which recent decisions by the Executive Board or the council are relevant. Based on this, we draw up the programme.
Municipalities are broad organisations with very different relationships to AI. We have designed the modules so we can shift the emphasis by audience: sometimes one session per role group, sometimes a mixed day with sub-sessions.
Employees who help citizens daily with applications, questions and objections. Focus: safe prompting for generating customer letters, the difference between an official decision and an AI suggestion, and handling personal data within the BIO.
A domain where AI could in theory do a great deal, but in practice must be handled with extreme care. Focus: bias evaluation, lessons from SyRI, when a social-domain algorithm falls under the AI Act as high-risk, and what a DPIA and Algorithm Register registration look like.
Permit issuing, BAG data, zoning plans, the environmental plan. Focus: AI for preliminary research, support with large case files, links with DSO integrations and the transparency requirements the Environment and Planning Act imposes.
Press releases, council letters, summaries of council documents, social media. Focus: hands-on prompting for public-facing language, checking factual accuracy, and how AI-generated text can support official neutrality and care without undermining them.
Strategic owners of compliance, security and architecture. Focus: AI Act classification, the Algorithm Register process, DPIA template, BIO impact, supplier assessment, model risk and the in-control statement.
Political and administrative decision-makers who take strategic decisions on AI investments and the associated risks. Focus: policy setting, controllability, public accountability, and the council's powers over AI decisions.
During the workshop we work with use cases that will be recognisable to municipalities. We do not present them as "success stories" (for some we are even deliberately cautious), but as concrete situations on which to apply legal, technical and ethical analysis. That is more instructive than an abstract conversation about "AI possibilities": participants discover for themselves where the dividing line lies between a useful assistant and a high-risk system.
A first classic is WMO application triage. A municipality wants to shorten processing times by having AI assist with incoming applications: a completeness check, an initial indication of the service category, and automatic requests for missing information. The question is not whether that is technically possible (it is), but under which conditions it is permissible. A suggestion to the case worker is fundamentally different from an automated pre-decision. We walk through what that distinction means for the DPIA, for the Algorithm Register and for the objection procedure.
A second is objection classification. AI supports the sorting and summarising of incoming objections so that the legal department can respond to their substance more quickly. Here the bias question comes to the fore immediately: a classification that draws conclusions based on postcode, surname or writing style can unintentionally place entire groups of citizens at the back of the queue. We cover which evaluation steps are mandatory before such a system may go into production, and how to keep monitoring it afterwards.
A third is permit preliminary research. A spatial planning department has AI perform an initial check on a permit application: complete documentation, an indicative test against the zoning plan, and flagging of risk areas. There are opportunities here, but also pitfalls: an incorrect preliminary conclusion can wrongly route an application into the wrong stream. We discuss the architecture that makes such a system auditable, including links with BAG, DSO and the archive.
A fourth is council document summarising. A council clerk's office or board secretariat uses AI to produce draft summaries of council papers, mayor and aldermen letters or implementation documents. This is a relatively low-risk application, as no decision is made about a citizen, but it does raise points of attention around factual accuracy, hallucination and political colouring. We practise with prompts that make the output more useful and easier to verify.
A fifth, and the most critical, is a citizen services chatbot. Many municipalities are experimenting with these, but under the AI Act a chatbot that answers questions about citizens' rights and obligations quickly becomes a high-risk system. We discuss the conditions under which such a chatbot can be justified, what a safe scope definition looks like, and which alternatives exist, for example an internal assistant for contact centre staff rather than a direct citizen-facing AI.
As a sixth, we cover customer letter and decision generation: AI as an assistant for staff who regularly have to draft letters and standard responses. Provided it is paired with human review and checked for plain, understandable language, this usually saves time without major legal risk. One point to watch: archiving and traceability of AI contributions.
These use cases serve as material for analysis. We work through each case against the AI Act classification, the BIO requirements, the Algorithm Register implications and practical feasibility. By the end, your team will share a view of what is sensible for your municipality to examine first, and what is not (yet).
For each application discussed, we start with its classification under the AI Act: prohibited, high-risk, limited risk or minimal risk, and what that means for your municipality. For most public-sector applications, the threshold is higher than in the private sector.
We go through which of your existing and planned systems belong in the Algorithm Register, what information you need to include, and how to build explainability for citizens and councillors without resorting to PR copy.
Many municipal AI questions sit at the intersection of BIO and GDPR. We treat them as a single governance question (data flows, authorisations, extending the DPIA, records management) rather than two separate chapters.
No responsible AI workshop for the public sector is complete without an honest discussion of SyRI and the childcare benefits system. Algorithmic decision-making failed on several levels at once: technically (proxy discrimination, lack of evaluation), legally (no DPIA in the spirit of the law) and in terms of administrative accountability. What was presented as "more efficient enforcement" became a mechanism that affected whole groups of citizens disproportionately and unlawfully.
The relevance is immediate: the technology now in the hands of any policy adviser (generative AI, automated text analysis, classification models) is far more accessible than the SyRI infrastructure ever was. That broadens the risk. A sound AI process ensures that every new application undergoes serious scrutiny: which variables, which proxies, which exclusionary effects, which feedback loop with the target group, which explainability, which possibility of objection.
We do not treat these lessons as a warning paragraph but as a testing tool: a set of questions we put alongside every AI use case. By the end of the workshop, your team holds that test, not as a form that disappears into a drawer but as a habit. For municipalities that regularly make this kind of decision, this fits with our broader practice in AI consultancy.
We come to your town hall with a team of two. Suitable for dedicated sessions with a department, a directorate or a wider group with representatives from different areas. A room with a projector and a laptop per participant is sufficient.
A live video session for distributed teams, also useful for regional partnerships involving several municipalities. Hands-on blocks run in your own browser; a recording is available for colleagues who could not attend.
For larger organisations, sometimes better than one long day: a series of, for example, three half-day sessions in which we address the audiences separately. CIO/CISO and legal affairs in session one, the social domain and citizen services in session two, communications and the executive in session three.
It starts with an intake conversation, usually with the substantive lead, a representative from the CIO portfolio and someone from legal affairs. We discuss the target group, existing AI activities, the Algorithm Register process, ongoing DPIAs and relevant decisions by the Executive Board or the council. On that basis we propose a programme with modules, format and schedule, sent in advance so that nobody has to guess on the day what is coming.
The workshop itself follows. We work in clear blocks: a short introduction to the subject, a longer hands-on or analysis exercise, and at the end a starting point for a follow-up roadmap. For the hands-on blocks, we ask in advance which official documents you would like to use, such as a draft decision, council report, set of citizen questions or policy memo, and build the exercises around them.
Afterwards, you receive a written wrap-up: which modules were delivered, who attended, which priorities came up and which next steps we agreed. This material is useful as documentation for the AI literacy obligation under Article 4 of the AI Act, and as a starting point for an Algorithm Register mapping. For ongoing programmes, we link to an AI implementation programme or our municipal software practice.
In our view, the workshop is not an end goal but a starting point. For some municipalities, that is enough: a solid foundation, a shared understanding and a set of follow-up questions for their own management team. For others, it becomes the trigger to go further: a first pilot, an Algorithm Register project or an extension of existing municipal software with an AI component. Both outcomes are legitimate; we do not push towards a follow-on engagement when it is not needed.
A thirty-minute introductory call in which we go through which departments are taking part, which Algorithm Register project you already have and where the first need lies. We then send a concrete proposal with modules, format and planning.
Appfront uses cookies and similar technologies to keep the website working properly, for analytics and for marketing. You choose what you allow. Read more in our privacy policy.
