AI consultant in The Hague
For ministries, government agencies and municipalities: first establish which decision an algorithm affects and who makes it, then build it through to production. We also built Dex for PostNL.
With Dex, PostNL gained a central place for employees’ questions
For PostNL, whose head office is on Waldorpstraat in The Hague, we developed Dex, a self-learning chatbot connected to existing systems so the knowledge base does not need to be stored and maintained in multiple places.
Do our case handlers still make their own decisions, or do they simply sign off on the recommendation?
A model ranks applications, and the algorithm register says a staff member makes the decision. But no one can see how often case handlers depart from the recommendation. We design that human oversight into the process: an explanation for each recommendation, logging of both the recommendation and the decision, and a designated owner.
A dossier for each algorithm that the register, IAMA and legal counsel can use
An overview of all algorithms
From risk-based selection to a traffic light programme: for each algorithm, its purpose, the data it uses, who owns it and the decision it informs.
A register entry for residents
A description for the Algoritmeregister: what the algorithm does, which data it uses, how a staff member intervenes and where people can go with questions.
Groundwork for IAMA and FRIA
For each system, the facts you also need for IAMA and the fundamental rights assessment: its purpose, the people affected and the data used.
Every change traceable
For Zorginstituut Nederland, we built a validation and publication pipeline in GitLab to enable controlled releases of products from the KIK-V programme.
The decision about a resident shapes how we build
-
Define the decision and roles
Which decision the algorithm supports, who makes it, and whether your organisation is the provider, the deployer or both.
-
Build in explanations and logging
A recommendation comes with an explanation for the caseworker and is logged alongside the decision.
-
Register and monitor
A description in the register, monitoring after launch and a review whenever the purpose, data or regulations change.
Would this be worthwhile for your ministry, agency or municipality?
A good fit if
Algorithms are in use or planned that influence decisions about residents or businesses, but you cannot yet answer the council, parliament or auditor when they ask exactly which algorithms are running.
Not a good fit if
It concerns a fixed calculation rule set out in law, such as calculating an administrative fee, that anyone can verify. In that case, a clear description in the register is more useful than an AI project.
What determines the scope
The number of algorithms and suppliers, whether decisions about individuals depend on them, how much documentation already exists, and whether we will also build and manage the systems. You provide a designated owner and access to contracts.
Who decides the risk classification
Your legal counsel or compliance department does; we do not provide legal advice. For each system, we provide the facts: what it is for, what data it uses, which residents it affects and who uses its output.
Buy an AI application, have one built or develop it yourself?
Buy
Choose this if an existing product meets the needs of your process. Specify in the tender that the supplier must provide documentation, logging and scope for an audit.
Have it built
Choose this if the algorithm plays a part in decisions affecting residents. Our advice: specify in the contract which code, data and model documentation you will receive, so you can explain every recommendation.
Develop it yourself
Choose this if your data team can monitor and revise the model after it goes live. If you temporarily lack capacity, we can provide an MLOps engineer or data scientist on secondment.
Consultancy and development from one partner, with management after go-live
Data Science Lab has worked with data and AI since 2016. As a Databricks Partner and Microsoft Partner, we work with clients including PostNL and Zorginstituut Nederland. We can meet at your office, at ours in Amsterdam or online.

Bring one algorithm to the first meeting
Stef Stoelwinder
Business DevelopmentStef is your first point of contact. In our first conversation, we’ll discuss one algorithm or AI idea: which decision it affects, who uses the outcome and what’s already recorded in the register and DPIA.
E-mail info@datasciencelab.nl
Phone +31 (0)20 3636 724
Which algorithm should we focus on first?
Describe the algorithm or idea and the decision that depends on it. We’ll reply by email or arrange an initial conversation.
Message sent
Thanks for your message. We’ll be in touch as soon as possible.
Frequently asked questions
The register focuses on algorithms with a significant impact, including high-risk AI systems. At a minimum, this covers algorithms that contribute to decisions about a person's rights, such as a fine or a grant. Publication is not yet a legal requirement; the Dutch government announced a requirement in 2022, but the law has not been passed.
The municipality of Den Haag, which uses an algorithm to predict how busy polling stations will be, includes algorithms that affect residents' daily lives and adds descriptions twice a year. You can decide who keeps this up to date through AI governance.
No. The European database under the AI Act lists high-risk systems, and a public authority that uses one must record its own use there. From 2 December 2027, Article 26 requires a public authority not to use a high-risk system if it is not listed, and to notify the provider. Separate rules apply to critical infrastructure.
According to its administrators, the Dutch register will remain in place because it also includes simpler, lower-risk algorithms.
Annex III includes systems that public authorities use to assess whether someone is entitled to essential benefits or public services. Under the digital omnibus, the rules for these high-risk systems will not apply until 2 December 2027. Before using certain high-risk systems, a public authority must then carry out a fundamental rights impact assessment (FRIA).
The IAMA, published by the Dutch Ministry of the Interior and Kingdom Relations (BZK), was aligned with the FRIA in February 2026. Your legal adviser determines which obligations apply to your system. We set out all the start dates in the AI Act in practice.
An employee who accepts every recommendation is not really making a decision. In its guidance, the Dutch Data Protection Authority identifies four factors: people, technology and design, process, and governance. Think of the time needed to review a case, an explanation for each recommendation, and the authority to depart from it.
The municipality of Den Haag requires an employee to make the final decision whenever an algorithm involves personal data and has legal consequences. You can have the application record when caseworkers depart from its recommendations, so you can demonstrate this too. We incorporate that into the design. That is how we put Responsible AI into practice.
Leaving out protected characteristics is not enough: a model may still discriminate through a proxy, such as a postcode. Compare how often the model selects people in each group and how often those selections prove justified, and check whether the data is representative. This requires information such as ethnic background, which you can process only under limited circumstances under the GDPR.
The Algorithm Framework published by BZK refers to the risk-profiling assessment framework of the Netherlands Institute for Human Rights. The Dutch Data Protection Authority, based in Den Haag, wants its new AI supervision directorate to ensure that such tools are tested for discrimination and accuracy.
Start with the sources: a chatbot is only as good as the knowledge it can access. At PostNL, Dex answers employees' questions so the service desk can focus on more complex issues. The transparency rules under the AI Act have applied since 2 August 2026: anyone talking to a chatbot must be told they are doing so no later than the start of the conversation.
Since April 2025, public authorities have had a government-wide position on the use of generative AI, with guidance on responsible use.
The Algorithm Framework identifies, among other things, requirements in the specification and award criteria, agreements on data and artefacts, the right to an audit, and a data processing agreement if personal data is involved. Decide in advance whether the algorithm has a determining influence on decisions about individuals.
Specify beforehand what logging and documentation you need to complete the register entry and the FRIA.
What else do you need to put in place besides the algorithm itself?
- AI consultancyHow we decide which AI application to tackle first and what it requires of your organisation.
- The AI Act in practiceThe timeline following the omnibus and how to start by taking stock of the AI already in use.
- Data & AI for governmentHow policy and implementation work from the same information.