Home / Industries / Public Sector
Industry — Public Sector

AI for the Public Sector

Government bodies and municipalities carry huge administrative load and rising citizen expectations. We build AI that cuts backlogs and improves service — with the transparency, accessibility and oversight the public sector rightly demands.

Last updated: 11 June 2026

Why AI here

Serve citizens better — transparently

Government runs on cases, forms and questions — and people wait while staff work through the backlog. AI can take on the repetitive parts: answering routine queries, processing documents and surfacing the right information, so staff focus on the cases that need human judgement.

In the public sector, how matters as much as what. We design for transparency, human oversight, accessibility and EU AI Act / GDPR compliance — with EU-hosted or on-premise options for sovereignty.

Where AI helps

Use cases across the public sector

01

Citizen-service AI chatbot

Grounded assistants that answer citizen questions accurately from official information.

02

Case & document processing

Read, classify and route applications and documents to clear backlogs faster.

03

Knowledge & FOI search

Find and summarise across regulations, policy and records — with sources.

04

Translation & accessibility

Make information available in plain language and multiple languages for everyone.

05

Error & anomaly detection

Flag errors and anomalies in large datasets — with a human reviewing every flag.

06

Policy data analytics

Turn operational data into insight for better, evidence-based policy.

How we work

From use case to transparent production

Step 1

Audit

We map the highest-value use cases, your data and transparency requirements.

Step 2

Build an MVP

By the second call you get a working prototype on your use case — not a spec.

Step 3

Deploy responsibly

We deploy with oversight, documentation and EU-hosted or on-premise options.

Step 4

Monitor

We track accuracy and fairness so the system stays trustworthy.

What you gain

Outcomes that serve citizens

What AI actually changes inside a Dutch gemeente

Walk into almost any Dutch municipality and you find the same bottleneck: cases waiting on a person. A bezwaarschrift (formal objection) sits in a pile for weeks because someone must read it, classify the legal grounds, pull the original decision and route it to the right team. A WOO request lands and a clerk spends two days searching shared drives and email archives that never had proper full-text search. A resident phones the KCC (klantcontactcentrum) for the fourth time because nobody can find their file. None of that is flashy. But this is exactly where AI for the public sector in the Netherlands earns its keep, where AI overheid projects move the numbers a wethouder or gemeentesecretaris actually reports on: processing time per case, the size of the backlog, and how long a citizen waits for a real answer.

The honest version of the pitch is this. AI does not replace the caseworker who decides whether a subsidy is approved or an objection upheld. It removes the reading, sorting and copying that happens before the decision, and the searching that happens around it. When a document-intake model classifies an incoming application, extracts the relevant fields, checks for missing attachments and drops it into the correct queue, the human starts from a prepared file instead of a raw PDF. Multiply that across thousands of cases a year and the backlog stops growing. That is the operational change worth paying for in any gemeente AI project, not a chatbot that sounds clever.

The EU AI Act is the design brief, not a footnote

For private companies, EU AI Act compliance is bolted on. For the public sector it is the specification, because a large share of government use cases land in the higher-risk tiers the regulation was written for. Anything that touches access to public services, benefits eligibility, law enforcement support, or the evaluation of people is treated as high-risk, with obligations around risk management, data governance, logging, human oversight and technical documentation. A system that decides — or strongly influences a decision — about a citizen's entitlement cannot be a black box a vendor refuses to explain.

We build for that from the first line of scope, not as a compliance pass at the end. In practice that means a few concrete commitments on every public-sector engagement:

  • Human-in-the-loop by default for any decision that affects a person. The model prepares, ranks or flags; a named official decides and can override. The override is logged. AI assists, people decide — and the audit trail proves it.
  • Traceability on every output. When a knowledge or FOI assistant answers, it cites the source document and passage, so a caseworker can click through to the regulation behind the answer instead of trusting a confident paragraph.
  • Documented, testable behaviour. We keep the technical documentation, evaluation results and logging the Act expects, so when an auditor, the Autoriteit Persoonsgegevens, or a gemeenteraad asks "how does this system work and where could it go wrong," there is a real answer on file.
  • Bias testing on protected groups, repeated over time. A model that quietly disadvantages residents in one postcode or with one surname pattern is a political and legal liability, not just a technical bug. We test for it before launch and keep monitoring it after.

This is where the Dutch context bites harder than most. The toeslagenaffaire taught every public-sector executive in the country what happens when an opaque, partly automated system produces unfair outcomes at scale and nobody can explain the logic. Responsible AI in Dutch government is not a marketing line; it is the precondition for being allowed to deploy at all. Our AI Audit & Strategy engagement starts by classifying each candidate use case against the Act's risk tiers, so you know on day one whether something is a quick win or a high-risk project needing the full governance wrapper.

Data sovereignty: keeping citizen data in the Netherlands

Public bodies hold the most sensitive data there is: BSN numbers, income, health flags, immigration status, criminal records. Sending that to a US-hosted commercial AI endpoint is a non-starter under the AVG and most municipal information-security policies, and rightly so. The good news: you no longer have to choose between modern AI and keeping data under Dutch or EU control.

We deploy with sovereignty as a hard requirement where it applies. That can mean open-weight language models running on EU-hosted infrastructure, or fully on-premise inside your own data centre or a government cloud, so that no citizen record leaves your control and no third party trains on it. For document processing and case handling, the model can run entirely within your network. The performance gap between hosted frontier models and well-tuned open models has narrowed to the point where, for classifying a vergunning application or extracting fields from an intake form, the on-premise option is more than good enough, and the one that survives a security audit. We design the integration and hosting in our AI implementation work, and the supporting pipelines in our data engineering practice, so the whole stack is something your own ICT team can run and residents' data stays where it belongs.

Where the time actually comes back

It helps to be specific about which workflows move the headline numbers, because not every demo touches them.

Document and case processing — the backlog killer

The single biggest lever in most municipalities is structured intake. Incoming applications, objections, permit requests and correspondence arrive as unstructured PDFs, scans and emails. A processing model reads each one, identifies the type, extracts the fields that matter, checks completeness and routes it to the right team with a draft summary attached. Industry benchmarks for this kind of intelligent document processing typically show 50 to 80 percent reductions in manual handling time per document for high-volume, repetitive flows — and the work that remains is the judgement, not the transcription. The visible result for residents is that a complete file moves through the system in days instead of weeks, and the backlog stops being a quarterly crisis.

Citizen-service assistants that are actually grounded

Citizen services are where most residents meet the municipality, so it is where badly grounded AI does the most damage. A chatbot only earns its place in publieke dienstverlening if it answers from your real, current information and never from invention. We build retrieval-grounded assistants that pull answers from your official policies, the relevant wet- en regelgeving and your own published procedures, with a citation behind every response and a clean handover to a human the moment the question gets complex or sensitive. The outcome is a measurable drop in routine call and email volume at the KCC, faster first responses outside office hours, and — because it is grounded — a system you can defend when someone asks why it said what it said. This natural-language work draws on our LLM optimisation and generative AI capabilities, tuned for accuracy and traceability rather than personality.

Knowledge and FOI search across the archive

WOO (Wet open overheid) requests and policy lookups quietly consume enormous amounts of senior staff time. Semantic search across decades of policy documents, council minutes and correspondence turns a two-day manual hunt into a focused query that returns the relevant passages with sources. Officials still review and decide what is released (the AI does the finding, not the judging), so response times improve and the legal exposure from missed documents drops.

Anomaly and error detection, carefully

Spotting duplicate payments, data-entry errors, or anomalies in large operational datasets is a genuine AI strength and a real money-saver. In the public sector it carries the most political weight, so we treat it with the most caution: anomaly detection here flags items for human review, full stop. It never auto-rejects, auto-sanctions or scores a citizen as a risk. A flag is a prompt for a person to look, never an automated verdict.

Why a boutique partner fits the public sector

The instinct in government is to hire one of the big consultancies — Capgemini, Xebia, the large system integrators — for an AI project. Sometimes that is right. Often it leaves you with a six-figure proof of concept, a rotating cast of junior consultants, and a system nobody internally understands or can maintain once the contract ends. The opposite failure is hiring a web agency that rebranded as "AI" last year and cannot speak credibly about the AI Act or on-premise deployment. Crux Digits is neither a marketing or web shop nor a body-shop integrator.

Crux Digits is the senior-led middle path. We are a boutique AI engineering partner founded in 2022, based at Vlierhoeve 100 in Nieuwegein (province of Utrecht) and working across the Utrecht region, the Netherlands and Europe. The senior people who scope your project stay on it through delivery. Our model is transparent and fixed-step: an AI Audit & Strategy at EUR 2,500 to map and risk-classify the use cases, a Proof of Concept at EUR 20,000 to prove value on your real data, and production from EUR 50,000, all excluding VAT and laid out plainly on our pricing page. No surprise change orders, no vendor lock-in. The deliberate end state is that your organisation owns the solution — code, models and documentation — so it keeps working long after we step back.

We come to public-sector work having already delivered across regulated, accountability-heavy domains. Our 13 case studies span computer vision, natural-language processing, forecasting and predictive maintenance, including municipal-adjacent work such as automatic licence-plate recognition for low-emission zones and road-defect detection from ordinary footage, both built with an audit trail behind every decision. Client names stay confidential, but the pattern transfers directly: read the document, surface the evidence, keep the human in charge.

A practical, low-risk way to start

You do not need an organisation-wide AI strategy before you do anything. Pick one painful, high-volume, lower-risk workflow (usually document intake or grounded citizen-service answers) and prove the value in a contained pilot. That is what the Audit and Proof of Concept steps are for: a small, transparent investment that produces a working system on your own data and a clear view of which use cases are worth scaling and which carry too much risk to automate.

If your backlog is growing, your KCC is overwhelmed, or your FOI response times are slipping, that is exactly the kind of problem we map in a free, no-obligation consultation. You can read how we approach Dutch and EU engagements on our AI consulting page, see the broader pattern of automating repetitive workflows under AI automation, or learn who you would be working with on our about page. Reach Tom Joseph and the team directly at info@cruxdigits.nl or +31 6 44384676 — and we will tell you honestly whether AI is the right tool for the job, including when it is not.

FAQ

Questions, answered

How do you handle transparency and the EU AI Act?

Public-sector use cases often carry higher risk under the EU AI Act, so we design for transparency, human oversight, documentation and explainability from the start.

Can it run on sovereign or on-premise infrastructure?

Yes — where data must stay in the Netherlands or EU, we use EU-hosted or on-premise models so nothing leaves your control.

How do you keep decisions fair?

We test explicitly for bias and keep a human in the loop for decisions that affect citizens — AI assists, people decide.

Will it work with our existing systems?

Yes — we integrate with your case-management and record systems rather than replacing them.

Who builds AI for Dutch government bodies and municipalities?

We do. Crux Digits is a boutique applied-AI firm in the Utrecht region, working with public-sector clients across the Netherlands and the EU, remotely or on-site. A small senior team does the build hands-on, with no offshore hand-off, and you own the source code and IP. We work in Dutch and English, and we'll tell you honestly where AI does and doesn't fit your service.

How do we start, and how do you handle public-sector accountability and the AVG?

We begin with a short paid discovery to scope one use case, then deliver at a fixed price. For the public sector we build in what your role demands: human oversight of decisions, clear records for openness and accountability requests, accessibility that meets Dutch digital-accessibility rules, and AVG data handling from day one. If a task is unsuitable for automation, we'll say so early.

Backlogs or citizen service to improve?

Tell us where the load is — we'll map a transparent, compliant path to value in a free consultation.

Book a free consultation →