AI for government: more time for residents

In government, AI can take over a lot of preparatory work: organizing questions from residents and businesses and preparing a draft answer, rewriting letters in plain language, summarizing case files, preparing meeting documents and answering questions from your own policy and procedures. With us, AI makes no decisions about residents or businesses. A civil servant always checks what AI prepares, and decides.

There's no shortage of work. The questions aren't getting fewer, the rules aren't getting simpler, and experienced people aren't easy to find. Every answer, letter and summary that's ready sooner gives time back for the work where a person makes the difference: the conversation, the judgement, the tailored solution.

This page is for municipalities, provinces, water authorities, implementing agencies and other public bodies. We've worked with organizations in government before, such as the Municipality of Lelystad and the Dutch Police. Where the work touches residents' personal data, extra requirements apply. We arrange those before a single line of code is written.

Because a decision about a resident or business belongs with a person who can explain it and be accountable for it. The AI Act is in line with that. AI in public services can fall under high risk, for example when it helps decide on access to benefits, services or other public provisions. Then strict requirements apply for documentation, human oversight and checks. Read what else the AI Act asks in the EU AI Act in plain language.

We focus on the preparatory work: drafts, summaries, searching policy and organizing documents. There too, we look at which rules apply for each application, and we keep a civil servant in the loop.

On top of that, the AI Act asks organizations that use AI to work on the AI literacy of their people. That applies to government just as much as to a business. And consider registering in the Dutch government's Algorithm Register: the public register in which government organizations describe the algorithms they use. That way residents can see where and how AI is used.

Yes, with the preparation. A Woo request often means searching through a lot of documents: emails, notes, minutes and attachments. AI can search them for what belongs to the request, organize them and prepare a first redaction proposal, with the reason for each passage why it might not be made public.

The case handler then assesses every passage, weighs the interests and decides what is made public. The decision on the request remains a decision of the public authority, based on a person's judgement. AI speeds up the searching and the first proposal, not the responsibility.

Because Woo documents often contain personal data, we only work within the organization's own environment and agreements, with access per role and logging. And we start small: first searching and organizing, and only then redaction proposals.

With three numbers per process step. In the APO method, we give each step a duration, a number of people and a frequency. A worked example, not a client figure: a step takes forty minutes, with two people, eighty times a month. That's 1,280 hours a year, and at € 45 an hour € 57,600 a year.

AI never takes all of that over. We calculate per step with the share AI can prepare, and a person keeps checking. In government, those hours are mostly capacity: time that comes back for complex case files, backlogs and residents who need a conversation.

Want to work it out yourself for one process? Use the AI savings calculator.